Decay cavity temperature acquisition fault diagnosis method and system based on thermal response consistency

Abstract

The application discloses a decay cavity temperature acquisition fault diagnosis method and system based on thermal response consistency, belongs to the field of optical measurement equipment state monitoring and temperature data acquisition fault diagnosis, and aims at solving the problem that real temperature transient and sensor and acquisition channel fault abnormal characteristics are similar and difficult to distinguish and locate. The application synchronously collects multi-position temperature, thermal driving working condition quantity, decay witness quantity and acquisition frame information, constructs a thermal response directed graph, a path consistency tensor and a working condition health path allowed domain, forms diagnosis evidence through leave-one-out reconstruction, node elimination recovery and frozen update, and realizes the technical effects of real working condition temperature change and acquisition fault shunting, fault measurement point positioning and fault type identification.

Description

Technical Field

[0001] This invention relates to the field of condition monitoring and temperature data acquisition fault diagnosis of optical measurement equipment, and particularly to a method and system for fault diagnosis of decaying cavity temperature acquisition based on thermal response consistency. Background Technology

[0002] In the process of spectral detection, gas detection, and cavity loss measurement, the temperature of multiple spatial locations within the ring-down cavity needs to be continuously acquired to characterize the cavity's thermal state and provide a status basis for ring-down time, cavity loss, and related measurement results. Existing temperature acquisition and monitoring typically relies on single-point amplitude, rate of change, or fixed thresholds, directly generating alarms for data exceeding the limits.

[0003] During the operation of the ring-down cavity, changes in laser power, gas flow rate or pressure, and uneven temperature field within the cavity can cause real temperature transients with propagation direction and arrival order. Simultaneously, sensor bias or drift, loose mounting contacts, gain shifts in the acquisition channel, and communication frame drops can also produce abrupt changes, slow deviations, delays, or value hold-offs. These two types of changes are similar in waveform at a single measurement point, making it difficult to distinguish them using fixed thresholds based on the thermal propagation relationship between multiple measurement points. This can easily lead to misinterpreting normal operating condition changes as faults and makes it difficult to locate abnormal measurement points or acquisition channels.

[0004] Furthermore, when updating diagnostic references using online historical data, if suspected fault measurement points and associated paths are not isolated, drift data will enter the health reference and weaken subsequent judgment capabilities; it is also difficult to distinguish between intermittent disconnection and thermal hysteresis or loose installation contact if only timestamps or temperature delays are used for judgment.

[0005] Therefore, there is a need for a fault diagnosis method and system for fluctuating cavity temperature acquisition based on thermal response consistency that can overcome the shortcomings of the existing technologies. Summary of the Invention

[0006] One objective of this invention is to propose a fault diagnosis method and system for temperature acquisition in a ring-down cavity based on thermal response consistency. Addressing the difficulty in distinguishing between actual operating temperature changes and acquisition faults, and in locating fault measurement points and types in existing technologies, this invention proposes the simultaneous acquisition of temperatures at multiple locations, thermally driven operating parameters, ring-down witness parameters, and acquisition frame information. A directed thermal response graph is constructed based on spatial location and airflow direction. Event temperature increments are normalized to form a topologically ordered path consistency tensor. Global operating condition consistency is obtained using the operating condition health path allowable domain. Furthermore, node residuals, path residuals, and node removal recovery are obtained through leave-one-out reconstruction. A dual-gating system for actual events and trusted nodes, along with a freezing and updating technique for suspected points and associated edges, is employed. This invention achieves the technical effects of suppressing false alarms, isolating faulty measurement points or acquisition channels, and maintaining the validity of dynamic health references under changing operating conditions.

[0007] This invention provides a fault diagnosis method for temperature acquisition in a decaying cavity based on thermal response consistency, comprising: S1, simultaneously acquiring temperature measurement point sequences, thermally driven operating parameters, and decay witness parameters at at least three spatial locations in the decaying cavity, detecting the joint changes of the thermally driven operating parameters, and determining thermal response events, their response directions, and event quantization values; S2, constructing a directed thermal response graph based on the spatial locations of the temperature measurement points and the airflow direction, normalizing the temperature increments of each temperature measurement point according to the event quantization values, extracting node responses, and generating path responses from the differences in node responses of adjacent temperature measurement points; S3, organizing the path responses according to the topological order of the directed thermal response graph to obtain a thermal propagation path consistency tensor, and comparing it with the current... S4. For each target temperature measurement point, a leave-one reconstruction is performed based on the remaining reliable temperature measurement points, thermally driven operating conditions, and the directed graph of the thermal response to obtain node residuals and path residuals. After removing the target temperature measurement point and its associated edges, the node removal recovery degree is calculated. S5. Based on the global operating condition consistency degree, the decay witness quantity, the node residuals, the path residuals, and the node removal recovery degree, the actual operating condition temperature change or measurement point acquisition failure is determined, and the update of the healthy path allowable domain is controlled accordingly. S6. A fault fingerprint is generated based on the node residuals and the path residuals, and the fault location and fault type are output.

[0008] Optionally, S1 includes: using at least one of laser power and gas flow rate and gas pressure as the thermally driven operating condition quantity, and using at least one of ring-down time and cavity loss characterization quantity as the ring-down witness quantity.

[0009] For each thermally driven operating condition quantity, calculate the direction of change, magnitude of change, and duration of change relative to the pre-event operating condition baseline. Combine the thermally driven operating condition quantity changes that meet the condition of proximity of the starting point of change within the same detection window into a single thermal response event.

[0010] The earliest starting point of the merged change is taken as the starting time of the thermal response event. The response direction is determined by weighting the change direction of each thermally driven operating condition quantity according to its historical thermal response sensitivity. The event quantization value is determined by weighting the change amplitude of each thermally driven operating condition quantity according to the normalized weighted result of the historical thermal response sensitivity.

[0011] Optionally, S2 includes: using temperature measuring points as nodes, and determining directed edges according to airflow direction and spatial adjacency between nodes;

[0012] The statistical value of the temperature measurement point sequence within the baseline window before the start time is determined as the pre-event temperature baseline. The change in the temperature measurement point sequence relative to the pre-event temperature baseline is used as the temperature increment. The temperature increment is divided by the event quantization value to obtain the normalized temperature increment.

[0013] The response symbol, start delay, time constant, normalized gain, and recovery slope are extracted from the normalized temperature increment as the node response, wherein the start delay is the difference between the time when the temperature increment first crosses the response boundary determined by the pre-event noise distribution and the start time, and the normalized gain is the response amplitude of the normalized temperature increment.

[0014] For each directed edge, calculate the initial delay difference between the downstream node and the upstream node, the normalized logarithmic ratio of the gain, and the normalized temperature increment waveform distance, as the path response.

[0015] Optionally, S3 includes: dividing the operating condition range according to at least one of the laser power range, gas flow rate range, and gas pressure range;

[0016] During the equipment commissioning or calibration phase, based on the path response samples corresponding to the reliable temperature measurement points in the confirmed thermal response events without acquisition faults, the initial quantile boundaries and initial joint covariance boundaries of each operating condition interval are determined respectively.

[0017] During the online operation phase, the path response samples and the path response samples corresponding to the credible temperature measurement points in the confirmed real operating conditions are used together as healthy path response samples. The quantile boundaries and joint covariance boundaries of the node start delay difference, normalized gain logarithmic ratio and normalized temperature increment waveform distance are determined respectively to form the healthy path allowable domain.

[0018] The global operating condition consistency is calculated by combining the proportion of path components located within the quantile boundary and the joint covariance boundary in the current heat propagation path consistency tensor with the normalized distance of each path component to its corresponding boundary.

[0019] Optionally, S4 includes: generating a reconstructed temperature increment of the target temperature measurement point based on the node responses of upstream reliable temperature measurement points pointing to the target temperature measurement point and downstream reliable temperature measurement points pointed to by the target temperature measurement point in the directed thermal response graph, combined with the current thermal driving condition, and forming the node residual by the difference between the measured temperature increment of the target temperature measurement point and the reconstructed temperature increment.

[0020] The path residual is formed by the distance between the current path response of the edge associated with the target temperature measurement point and the boundary corresponding to the healthy path allowable domain.

[0021] Before and after removing the target temperature measuring point and its associated edges, the normalized total distance from the unremoved path to the allowed domain of the healthy path is calculated respectively, and the reduction ratio of the total normalized distance is used as the node removal recovery degree.

[0022] Optionally, S5 includes: performing discrimination in the following order: first, comparing the global operating condition consistency with the operating condition consistency criterion determined by the quantile value of the healthy path response sample, and verifying whether the non-target temperature measurement points that have reached the quantity criterion determined by the number of credible temperature measurement points maintain the response sign corresponding to the response direction and the arrival order along the directed graph of the thermal response, and whether the decay witness quantity maintains the change direction corresponding to the thermally driven operating condition quantity;

[0023] When the aforementioned verification is successful, it is determined to be a true temperature change under actual operating conditions, and the alarm for exceeding the temperature amplitude limit is suppressed.

[0024] If the aforementioned verification fails, the node residuals and path residuals of each target temperature measuring point are compared with the residual criteria determined by the boundary of the healthy path allowable domain. When both exceed the corresponding residual criteria and the node removal recovery indicates that the remaining path returns to the healthy path allowable domain after removal, the target temperature measuring point or its acquisition channel is located as faulty.

[0025] Furthermore, the reference update constraint of the health path allowed domain includes: only after the real operating condition temperature change judgment is established, selecting reliable temperature measurement points and their path responses that have not crossed the corresponding residual criterion and are not in a frozen state as update samples;

[0026] The dual-gated update is performed in the following order: real-world temperature change discrimination, reliable temperature measurement point screening, and updated sample writing.

[0027] When the number of updated samples through the dual gating in any operating condition interval is less than the lower limit of the number of samples determined by the joint covariance dimension and the target coverage level, the updated samples are cached and the existing healthy path allowed domain of the operating condition interval is maintained until the number of cached updated samples reaches the lower limit of the number of samples before the corresponding boundary is updated.

[0028] When the node residual or path residual of any temperature measurement point crosses the corresponding residual criterion, the data of that temperature measurement point, its associated edges, and the fault duration window from the time of crossing to the time when the recovery criterion is established is frozen. The freezing is lifted after the node residual and the path residual both return to the corresponding residual criterion and the node removal recovery degree returns to the recovery interval corresponding to the healthy path response sample distribution.

[0029] Optionally, S6 includes: synchronously acquiring acquisition frame information consisting of frame timestamps and frame sequence numbers;

[0030] The DC component, cumulative slope, and number of isolated pulses are extracted from the node residuals. The start delay, time constant, normalized gain of the thermal response event, and path delay difference are extracted from the node response and the path response. The frame sequence number gap, timestamp interval, and duration of continuous repetition value are extracted from the acquired frame information.

[0031] The DC component continuously crossing the corresponding residual criterion is mapped as a bias fingerprint; the cumulative slope maintaining the same direction in continuous thermal response events is mapped as a drift fingerprint; the isolated pulse is mapped as a sudden change fingerprint; and the continuous increase of the time constant and the path delay difference under frame continuity conditions is mapped as a thermal hysteresis or loose installation contact fingerprint.

[0032] Based on the normalized gain samples of cross-thermal response events of reliable temperature measurement points within the same operating condition range, the gain boundary of the healthy node is determined, and the deviation of the normalized gain of the cross-thermal response events of the target temperature measurement point from the gain boundary of the healthy node is mapped as a gain offset fingerprint.

[0033] Furthermore, the combination of a frame sequence number gap or timestamp interval crossing the continuity boundary determined by the healthy acquisition cycle distribution, and the temperature measurement point sequence maintaining the previous sampled value during the gap or resuming the update after the gap, is mapped as an intermittent disconnection fingerprint;

[0034] For path response features, a normalized distance relative to the boundary of the allowed healthy path domain is used; for node residual features, a normalized distance relative to the residual criterion determined by the residual distribution of healthy nodes is used; for node response features across thermal response events, a normalized distance relative to the healthy node response boundary determined by the historical node response samples of reliable temperature measurement points is used; and for acquisition frame features, a normalized distance relative to the continuity boundary is used.

[0035] The evidence quantity is determined based on the normalized distance, number of consecutive events, number of associated abnormal edges, and number of mutually consistent witnesses corresponding to each fault fingerprint, and the fault confidence is determined based on the empirical distribution of the evidence quantity in historically labeled fault samples.

[0036] When multiple fault fingerprints are established simultaneously, intermittent disconnection is first determined based on the collected frame information, and then thermal hysteresis or loose installation contact is determined based on the time constant and path delay difference under continuous frame conditions. Other fault types are output in descending order of the corresponding amount of evidence.

[0037] On the other hand, the present invention also provides a fault diagnosis system for decay cavity temperature acquisition based on thermal response consistency, including:

[0038] The data acquisition and event detection module is used to synchronously acquire temperature measurement point sequences, thermally driven operating conditions, and decay witness quantities to determine thermal response events; the response graph construction module is used to construct a directed thermal response graph and generate normalized temperature increments, node responses, and path responses; the path consistency analysis module is used to generate a thermal propagation path consistency tensor and determine the global operating condition consistency; the leave-one-out reconstruction module is used to generate node residuals, path residuals, and node removal recovery; the discrimination and reference update module is used to discriminate between actual operating condition temperature changes or measurement point acquisition failures and control the update of the healthy path allowable domain; and the fault fingerprint module is used to generate fault fingerprints and output fault locations and fault types.

[0039] The beneficial effects of this invention are:

[0040] 1. By normalizing the temperature increment of multiple measurement points with the quantization value of thermally driven events and organizing the path response according to the topological order of the directed graph of thermal response, the thermal propagation relationship under different event intensities and operating conditions is comparable; by combining the decay witness quantity, the response sign of non-target measurement points and the arrival order, the real temperature transient with overall propagation consistency can be separated from isolated acquisition anomalies.

[0041] 2. By performing leave-one reconstruction on each target temperature measurement point and simultaneously calculating node residuals, associated edge path residuals, and node removal recovery, abnormal measurement points or acquisition channels can be located by checking whether the remaining paths are consistent after removing candidate points. Fault fingerprints are formed by combining node, path, cross-event gain, and frame continuity features, which can distinguish between offset, drift, abrupt change, thermal hysteresis, loose installation contact, gain offset, and intermittent disconnection.

[0042] 3. By updating the health path allowed domain using only trusted nodes from confirmed real-world events and freezing suspected faulty nodes, associated edges, and corresponding data windows, the existing health domain can be maintained when online samples are insufficient. This reduces the possibility of fault data contaminating the dynamic reference and keeps the health boundary on which continuous diagnosis is based valid. Attached Figure Description

[0043] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0044] Figure 1 This is a flowchart of the fault diagnosis method for damping cavity temperature acquisition based on thermal response consistency according to the present invention.

[0045] Figure 2 This is a flowchart of the S4 leave-one reconstruction and node culling recovery degree calculation of the present invention. Detailed Implementation

[0046] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0047] refer to Figures 1-2 A fault diagnosis method for temperature acquisition in a decaying cavity based on thermal response consistency includes: S1, simultaneously acquiring temperature measurement point sequences, thermally driven operating conditions, and decay witness quantities at at least three spatial locations in the decaying cavity; detecting the joint changes of the thermally driven operating conditions; and determining thermal response events, their response directions, and event quantization values; S2, constructing a directed thermal response graph based on the spatial locations of the temperature measurement points and the airflow direction; normalizing the temperature increments of each temperature measurement point according to the event quantization values; extracting node responses; and generating path responses from the differences in node responses of adjacent temperature measurement points; S3, organizing the path responses according to the topological order of the directed thermal response graph to obtain a thermal propagation path consistency tensor, and integrating it with the current operating conditions. S4. For each target temperature measurement point, a leave-one reconstruction is performed based on the remaining reliable temperature measurement points, thermally driven operating conditions, and the directed graph of the thermal response to obtain node residuals and path residuals. After removing the target temperature measurement point and its associated edges, the node removal recovery degree is calculated. S5. Based on the global operating condition consistency degree, the decay witness quantity, the node residuals, the path residuals, and the node removal recovery degree, the actual operating condition temperature change or measurement point acquisition failure is determined, and the update of the healthy path allowable domain is controlled accordingly. S6. A fault fingerprint is generated based on the node residuals and the path residuals, and the fault location and fault type are output.

[0048] In this specific embodiment, S1 includes:

[0049] In this specific embodiment, the acquisition and event detection module uses a 600mm long ring-down cavity as the execution object. Four temperature measurement points are set at the cavity wall at the inlet end, the middle cavity wall, the cavity wall at the outlet end, and near the optical base. The temperature, laser power, gas flow rate, gas pressure, ring-down time, and cavity loss characterization quantity converted from the ring-down time are sampled at 2Hz using a unified clock. Each frame records the device number, event candidate number, measurement point number, sampling time, frame number, temperature reading, laser power reading, gas flow rate reading, gas pressure reading, ring-down time reading, and cavity loss characterization quantity. The temperature unit is degrees Celsius, the laser power unit is watts, the gas flow rate unit is standard milliliters per minute, the gas pressure unit is kilopascals, and the ring-down time unit is microseconds. The module performs nearest neighbor alignment using frame timestamps and limits the alignment error between channels to within 100ms. Frames exceeding this error are marked as asynchronous frames and are not included in the current event detection window.

[0050] The module maintains a 60-second pre-event baseline window for laser power, gas flow rate, and gas pressure, and applies the formula... Determine the operating condition baseline. Indicates the first The median value of each thermally driven operating condition within the baseline window before the event, with the same unit as the corresponding operating condition. Indicates the index of thermally driven operating conditions. This represents the set of sample sequence numbers within the baseline window prior to the event. Indicates the first The thermal driving condition quantity is at the first The measured values ​​at each sampling time, with the same units as the corresponding operating conditions. Indicates the sampling sequence number. When there are more than 10% of different steps or out-of-bounds values ​​within the baseline window, the previous valid baseline is maintained and the current window is marked as unusable for event start point determination.

[0051] For each thermally driven operating condition, the median value of five consecutive sampling points is used as the current value, and calculations are performed. , and , Indicates the first The direction of change of each thermally driven operating condition quantity is given, and the value is -1, 0, or 1. Indicates the first detection window The current median value of each thermally driven operating condition, with the same unit as the corresponding operating condition value. This indicates the magnitude of the change, and the unit is the same as the corresponding operating condition quantity. Indicates the duration of the change, with the unit being seconds. This indicates the sampling number where the change first crosses the detection boundary. This indicates the last sample number of the continuously out-of-bounds section. The sampling period is 0.5s in this specific implementation. The detection boundary is taken as three times the median absolute deviation of the steady-state sample of the same working condition during the commissioning phase and the upper limit of the instrument resolution.

[0052] The module uses a 10-second joint detection window to collect out-of-bounds sections of various thermally driven operating conditions. The difference between the start times of two sections is set as the condition of proximity of the change start point. The changes of thermally driven operating conditions that meet the condition of proximity of the change start point within the same detection window are merged into a thermal response event. The earliest change start point among all merged sections is taken as the start time of the event. If the same operating condition quantity alternates between positive and negative directions within the detection window and the duration of either direction is less than 2 seconds, the section is discarded as a jitter. If there is no out-of-bounds section that meets the condition of proximity of the change start point and the duration is not less than 2 seconds, the current event candidate is closed and the rolling detection continues. If only one qualified operating condition quantity change is retained after merging, a single-drive thermal response event is established and the flag of the remaining operating conditions not being triggered is written into the event record.

[0053] During the commissioning phase, non-overlapping small steps are applied to the laser power, gas flow rate, and gas pressure. The signed historical thermal response sensitivity is obtained by dividing the median of the peak temperature increment within a 120s observation window at four reliable temperature measurement points by the corresponding step amplitude of the operating condition. The absolute value of this sensitivity is also saved. The thermally driven sensitivity table fields include the type of operating condition, unit, signed historical thermal response sensitivity, absolute value of sensitivity, scale of change in the same unit, applicable power range, applicable flow range, applicable pressure range, version number, and effective time. The scale of change in the same unit is the median of the absolute value of the commissioning step amplitude of the operating condition. During runtime, records are searched according to the current operating condition range and table version. If no match is found, the event candidate status is maintained and the generation of samples for subsequent health domain updates is prohibited.

[0054] The module takes the intersection of the set of qualified operating conditions participating in the merging of this event and the effective set of sensitivity, and only includes those that simultaneously belong to this intersection. are finite numbers and satisfy The operating conditions are included in the effective sensitivity participation set. Operating conditions that are not triggered in this event or that are triggered but do not meet the merging conditions do not participate in directional contribution, weight normalization and event quantification. Indicates the first The signed historical thermal response sensitivity of the effective participating operating condition quantity, in degrees Celsius per unit of that operating condition quantity. Indicates the first The effective lower limit of sensitivity for each effective input condition quantity, expressed in degrees Celsius per unit of that input condition quantity. This represents the highest temperature resolution among the four reliable temperature measurement points, expressed in degrees Celsius. Indicating the number in the sensitivity table The same unit scale of change for each effective participating working condition quantity. This indicates a positive number in the same unit to prevent division by zero, and takes the resolution of the corresponding operating condition instrument; valid participating items are renumbered according to the original operating condition order in the merged record of this event. to , This represents the number of valid sensitivity participation terms in the intersection of the above terms, ranging from 0 to 3. For each valid participation term, let... And order , This represents the dimensionless thermal response direction coefficient, which takes the value of -1 or 1. This represents the dimensionless sensitivity weight. This indicates that the summation index of the valid participants is equal to 1; when When the response direction contribution and event quantization value are not calculated, the sensitivity is set to unavailable and the response direction is set to uncertain, the event candidate state is maintained, the generation of healthy samples is prohibited, and the transmission to S2 or into S5 deterministic discrimination is prohibited; when At that time, the contribution of the response direction is based on synthesis, This indicates the first in this incident The direction of change of each effective participating working condition quantity is taken as -1 or 1, when Press at time Determine the direction of response, when season Concatenate response direction uncertainty flag, This represents the dimensionless, signed directional contribution. This indicates a zero-protection boundary for directional conflicts, with a value of 0.1. Indicates the response direction of a thermal response event and takes values ​​of -1, 0, or 1; the event quantization value is set according to... Sure, This indicates the first in this incident The absolute value of the change amplitude of each effective participating working condition quantity, with the same unit as that working condition quantity. This represents the quantization value of a dimensionless non-negative event; events with uncertain response directions do not undergo S5 deterministic response sign verification; events with a quantization value less than 0.2 are marked as weakly stimulated events and used only for online observation without being included in healthy samples.

[0055] The module synchronously verifies the pre-event baseline, direction of change, and magnitude of change of the decay time and cavity loss characterization quantity. It writes the changes of the two relative to the median value of their 60s baseline window into the decay witness record. When the decay time or cavity loss characterization quantity has at least four consecutive sampling points that exceed three times the median absolute deviation of their respective healthy noise, a witness valid flag is set. If it does not exceed the limit, the original witness quantity is retained and a witness unconfirmed flag is set. The witness quantity in S5 is used to distinguish between overall thermal condition changes and isolated temperature acquisition anomalies, but does not participate in the calculation of event quantification value.

[0056] When there are fewer than three effective synchronous temperature measurement points, all thermally driven operating conditions lack effective frames in the detection window, or the 100ms channel alignment constraint cannot be met for 3 consecutive seconds, the module determines that the current detection cycle does not meet the conditions for generating thermal response events, cancels the event candidate, retains the missing frame and asynchronous diagnostic records, and waits for the next detection window. Event detection is reopened after at least three effective synchronous measurement points and at least one thermally driven operating condition is valid for 4 consecutive frames.

[0057] After the event detection is completed, the module generates a thermal response event record and writes it to the event cache. The record includes the event number, start time, start and end times of the detection window, types of operating conditions involved in merging, directions of change, amplitudes of change, durations of change, response direction, event quantization value, sensitivity table version, decay witness record, and synchronization quality flag. Records with event quantization values ​​not less than 0.2 and qualified synchronization quality, along with the alignment sequence of the four temperature measurement points 60 seconds before and 180 seconds after the start time, are transmitted to S2.

[0058] In this specific embodiment, S2 includes:

[0059] The response graph construction module reads the thermal response event records output by S1, the coordinates of four temperature measurement points, and the steady-state airflow direction in the decaying cavity. Using the temperature measurement points as nodes, it determines the upstream and downstream relationships based on the airflow projection. Construct a directed graph of thermal response. This represents the directed graph of the current device's thermal response. This represents a set of nodes containing fields for measurement point number, three-dimensional coordinates, installation location, and trusted status. This represents a set of directed edges containing upstream node number, downstream node number, axial distance, lateral distance, and valid status fields. When the projected distance of the downstream measuring point along the airflow direction is 50mm to 350mm and the lateral distance is not greater than 150mm, a directed edge is established. When the projected distances are the same, a fixed topology order is determined according to the measuring point number to avoid cyclic edges.

[0060] The module removes out-of-synchronization frames and isolated values ​​outside the historical healthy range of the measurement point within a baseline window 60 seconds before the start of each event. and Calculate the temperature baseline and temperature increment. Indicates the first The pre-event temperature baseline for each temperature measurement point, in degrees Celsius. Indicates the index of the temperature measurement point. Indicates the first The temperature measuring point at the first The measured temperature values ​​at each sampling time are in degrees Celsius. This represents the temperature increment relative to the pre-event temperature baseline, in degrees Celsius. When the effective frame rate of the baseline window is less than 90%, the node is set to temporarily untrusted and the generation of current path responses for its associated edges is stopped.

[0061] Temperature increment by Normalization Indicates the first The temperature measuring point at the first The normalized temperature increment at each sampling time, in degrees Celsius. This represents the dimensionless event quantization value obtained by S1. The dimensionless lower limit of the event quantization value is taken as 0.2 in this specific embodiment. The normalization process ensures that the node response corresponding to different thermal drive change amplitudes remains in the same dimension. Weakly stimulated events with event quantization values ​​lower than this lower limit are used to participate in the display calculation but do not enter the healthy path allowed domain.

[0062] The module determines the response boundary using baseline window temperature increment noise. , Indicates the first The response boundary of each measuring point is given in degrees Celsius. This indicates the operation of median absolute deviation, where the condition is met for the first time. The time when four consecutive sampling points are maintained is determined as the starting point of the node's response. Indicates the first The resolution of each temperature measurement point is in degrees Celsius, and the node start delay is... Take the difference between the response start time and the event start time in seconds. If the response boundary is not crossed within the observation window, write the start delay as missing and set the node to a non-response state.

[0063] For nodes that have already responded, the module will respond with symbols. Take the sign of the absolute peak value of the normalized temperature increment within the observation window, and set the normalized gain accordingly. Take the absolute peak value of the normalized temperature increment within 120 seconds after the starting point and retain its sign. Indicates the first The response sign of each node is either -1 or 1. Indicates the first The normalized gain of each node, in degrees Celsius, is used to determine the moment when the module reaches 63.2% of the final response amplitude from the monotonic principal response segment after the response start point, and the interval relative to the response start point is used as the time constant. , Indicates the first The time constant of each node is in seconds. When the main response segment does not satisfy the requirement of 6 consecutive points changing in the same direction, first-order exponential least squares fitting is used and nodes with a fitting determination coefficient lower than 0.8 are marked as low confidence in time constant.

[0064] Within a 60-second recovery window following the peak response, the module performs a robust linear fit on the normalized temperature increment versus time and uses the slope as the recovery slope. , Indicates the first The recovery slope of each node is expressed in degrees Celsius per second. The node response record consists of measurement point number, response symbol, start delay, time constant, normalized gain, recovery slope, response boundary, fitting quality, and event number. Low confidence time constants do not participate in the thermal hysteresis fingerprint of S6, but the remaining effective node responses can still be used for path calculation.

[0065] For each directed edge Modules are Generate path response, Indicates that it is from the upstream node Pointing to downstream nodes The three-component path response, This represents the downstream temperature measurement point index. The first component is the start-time delay difference between nodes, in seconds, and the second component is the normalized logarithmic ratio of the gain, which is dimensionless. This represents the zero-gain protection value, and in this specific embodiment, it is taken as 0.01 degrees Celsius. This represents the dimensionless distance of the normalized temperature increment waveform.

[0066] Waveform distance according to calculate, This represents the set of common response sampling sequence numbers with a length of 60 seconds, starting from the response start point of the later node. Indicates the number of common response sampling points. and These represent the dimensionless waveforms obtained by dividing the corresponding normalized temperature increment by their respective absolute peak values ​​within the common response window. If any node has no response, the proportion of common effective points is less than 80%, or the absolute value of the gain is less than the zero protection value, the edge is marked as invalid and not replaced with a zero value.

[0067] The module writes the node response table and path response table according to the topological order of the directed graph of thermal response. The path response table fields include event number, edge number, upstream node number, downstream node number, start delay difference, normalized gain logarithmic ratio, waveform distance, valid state and graph version. The module also passes the valid path response, node response, normalized temperature increment sequence and graph version to S3, and passes the list of invalid nodes and their associated edges to S4 and S5 for the confidence quantity criterion.

[0068] In this specific embodiment, S3 includes:

[0069] The path consistency analysis module reads the path response table output by S2 and arranges the initial delay difference, normalized gain logarithmic ratio, and waveform distance of each effective edge according to the topological edge order of the directed graph of thermal response. , Indicates the first Consistency tensor of heat propagation path for each thermal response event Indicates the index of the thermal response event. This represents the set of valid directed edges for the current event. This indicates the number of valid directed edges. Each row of the tensor is bound to a unique edge number in the graph version. Missing edges retain the missing mask and are not included in the calculation of allowed domain distance and component proportion.

[0070] The module divides the operating condition intervals according to at least one of the laser power interval, gas flow rate interval, and gas pressure interval. In this specific implementation, the operating condition interval keys are established using a laser power interval of 20W width, a gas flow rate interval of 100 standard milliliters per minute width, and a gas pressure interval of 5 kPa width. The boundary is assigned to a unique operating condition interval according to the left-closed and right-open rule. The operating condition interval record includes the power interval, flow rate interval, pressure interval, effective edge set, sample number, quantile boundary version, and joint covariance boundary version. When the current operating condition quantity exceeds the calibrated interval, the adjacent interval is selected for diagnostic display only. The automatic discrimination qualification is set to display only, the global operating condition consistency is set to insufficient evidence, and the update of the health path allowed domain for this event is frozen.

[0071] During the equipment commissioning phase, initial healthy path response samples are formed from events that have been manually verified to have no data acquisition faults and whose decay witness quantity is consistent with the direction of the thermally driven operating condition quantity. For each operating condition interval, each side, and each component, the 2.5% quantile value and the 97.5% quantile value are saved as initial quantile boundaries. For the three components of the same side, the robust center vector and covariance matrix are calculated. The robust center is the component median value, and the covariance is the sample covariance after removing the highest 5% of samples with Mahalanobis distance. When the initial sample is less than 80 events, the commissioning state is maintained, the automatic discrimination qualification is set to insufficient samples, the global operating condition consistency is set to insufficient evidence, and the automatic fault discrimination of that operating condition interval is not enabled.

[0072] During the online operation phase, the path responses corresponding to unfrozen trusted nodes in the real operating condition temperature change events confirmed by S5 are appended to the healthy sample cache in the same operating condition interval. The healthy path allowed domain record includes the operating condition interval key, edge number, upper and lower quantile boundaries of the three components, robust center vector, covariance matrix, target coverage level, sample time range, graph version and domain version. A single path record is searched by operating condition interval key, edge number and graph version. If no match is found or the graph version is inconsistent, the allowed domain is set to unavailable and transferred to the calibration queue.

[0073] For the current path component and its allowed range Calculate the normalized distance of the boundary , Indicates the first position of the current path One portion, Component indices representing initial delay difference, normalized gain logarithmic ratio, or waveform distance. and These represent the 2.5% and 97.5% quantile boundaries of the healthy samples in this component, respectively, with the same units as the component. This represents the zero protection quantity in the same unit as the component, taking its instrumental or numerical resolution. It represents the dimensionless distance to the quantile boundary, taking the value of 0 when it is within the quantile interval and increasing in value the further it is from the boundary.

[0074] To eliminate the dimensional differences among the three path components, the module first... Perform component-by-component standardization. Representing an edge No. Dimensionless standardized values ​​of each path component Indicates the edge number 1 The current path component and the unit remains S2. This indicates the health sample number of that side. The median value of each component and its unit is... same, This represents the robust scale obtained by dividing the interquartile range of the healthy sample by 1.349, and the units are... same, This refers to the instrument or numerical resolution for the zero protection quantity within the same unit, and for which that component is taken. and Let these represent the upstream and downstream node indices of a directed edge, respectively. Indicates the path response component index;

[0075] The module will have three sequentially form dimensionless vectors , and according to Calculate the joint covariance distance. Representing an edge The dimensionless Mahalanobis distance of the current path response. This represents the dimensionless, three-dimensional robust center vector of the edge health-normalized vector. This represents the three-dimensional dimensionless covariance matrix corresponding to the health-standardized vector. Denotes the trace of the dimensionless covariance matrix, when Not less than 0.001 hours Pick Otherwise Pick , This represents a dimensionless covariance regularity. Represents a third-order identity matrix, with superscript Indicates transpose, and the joint covariance boundary is taken as the 99th percentile of the Mahalanobis distance of the calibrated healthy samples. , This represents the upper boundary of the dimensionless joint covariance of this edge;

[0076] The module will display currently valid, allowed field records that are available and The set of edges , This represents the set of directed edges that can participate in the consistency calculation. Indicates the dimensionless joint boundary zero protection quantity and takes and order , Indicates the number of paths that can participate in the consistency calculation; when Without calculating path ratios and average out-of-bounds distances, the consistency evidence is directly flagged as insufficient; when In this case, a path is considered an intra-domain path only if the quantile boundary distances of the three components are all 0 and the joint covariance distance is not greater than the corresponding upper boundary. Module calculation and , This represents the dimensionless proportion of the number of paths within the domain to the total number of paths eligible for consistency calculation. express The number of intra-domain paths in the domain express The average dimensionless outbound distance of the path relative to the allowed domain boundary. This indicates an operation that truncates a value to a given lower and upper limit;

[0077] Global operating condition consistency calculate, The value ranges from 0 to 1, representing the global operating condition consistency. A larger value indicates that the current heat propagation relationship is more consistent with the healthy operating condition response. The weight of the path proportion within the domain is indicated and is calibrated to 0.6 in the validation set in this specific implementation. The source of the weight and the applicable graph version are written into the consistency parameter table. When the number of available paths is less than 60% of the number of valid edges in the graph, the consistency is not output and the insufficient evidence flag is set.

[0078] The module writes the heat propagation path consistency tensor, missing mask, quantile boundary distances of each component, joint covariance distance, intra-domain path marker, global operating condition consistency, operating condition interval key, healthy path allowed domain version, automatic discrimination qualification, and reason code into the path consistency event record, and passes the record to S4 and S5. The automatic discrimination qualification includes at least discriminable, display only, pending calibration, inconsistent graph version, and insufficient samples. It is set to discriminable only when the current operating condition is within the calibration range, the graph version is consistent, the allowed domain is available, and there are no less than 80 healthy samples in the interval. All other states keep the global operating condition consistency as insufficient evidence. The healthy path sample cache only accepts the write command after S5 dual gating. The current step does not directly update the boundary based on its own consistency.

[0079] In this specific embodiment, S4 includes:

[0080] The Leave-One Reconstruction Module takes the normalized temperature increment sequence of S2, node response, directed thermal response graph, and health path allowable domain of S3 as inputs. It sequentially sets each current credible temperature measurement point as the target temperature measurement point, temporarily removes the measured sequence of the target measurement point, and retains other credible measurement points, current laser power, gas flow rate, gas pressure, event quantization value, and response direction. The upstream and downstream credible node sets associated with the target measurement point are determined by the current graph version. When there are at least two adjacent credible nodes, a multi-adjacent leave-one model is adopted. When there is only one adjacent credible node, the single adjacent node and three thermally driven operating condition quantities jointly back-off model, which is separately calibrated by the same operating condition health sample, are called. The node residual criterion and node removal recovery criterion are raised to the 99.5th percentile of the corresponding health distribution, respectively. When there are no adjacent credible nodes, the target measurement point is marked as insufficient reconstruction location evidence and is prohibited from outputting non-frame fault locations based on this. However, the original acquisition frame identity of the measurement point is still passed to S6 for use by the direct frame fault branch.

[0081] During the commissioning phase, a leave-one-out reconstruction parameter record is established according to the operating condition interval and the target measurement point. The thermal response events confirmed to be without acquisition failure are used as fitting samples. The normalized temperature increments of other reliable nodes after healthy median time delay correction and the same-scale normalized changes of the three thermally driven operating conditions are used as inputs. The measured normalized temperature increments of the left-out target measurement points are used as outputs. The coefficients are determined by ridge regression with second-order difference smoothing terms. The parameter record includes the target measurement point number, adjacent node number, time delay correction amount, regression coefficient, regularization coefficient, residual quantile boundary, operating condition interval key, graph version, and parameter version. The record is kept unusable when the cross-validation determination coefficient is lower than 0.85.

[0082] Press during runtime Generate the reconstructed normalized temperature increment of the target measurement point. Indicates the first The target measurement point is at the first The reconstructed normalized temperature increment at each sampling time, in degrees Celsius. This represents the intercept of the reconstructed model of the target measurement point, expressed in degrees Celsius. This represents the set of trusted upstream and downstream adjacent nodes of the target measurement point. Indicates adjacent nodes For the target measurement point The dimensionless regression coefficient, This represents the integer sampled displacement calculated from the health path start delay difference. Indicates the first The regression coefficients of the normalized changes in the thermally driven operating conditions on the target measuring point are given, with units in degrees Celsius. Indicates the first The dimensionless change of each thermally driven operating condition quantity after normalization to the same unit of change scale as S1.

[0083] Module by Restore the reconstructed temperature increment, and press... Form a node residual sequence. This represents the reconstructed temperature increment of the target measuring point, expressed in degrees Celsius. This represents the difference between the measured temperature increment and the reconstructed temperature increment at the target measurement point, expressed in degrees Celsius. When the proportion of valid reconstructed points in the event response window is less than 80%, the residual of that node is marked as invalid. When the parameter version does not match the current operating condition range, it will not be called across ranges and will enter the pending calibration state.

[0084] Nodal residual magnitude according to calculate, This represents the root mean square of the nodal residuals of the target measurement point within the current event, expressed in degrees Celsius. This represents the set of valid response sample numbers from the start of the event to 120 seconds after the start. This indicates the number of valid response sampling points. The module also saves the residual mean, robust slope, and absolute peak value for S6 to generate fingerprints. The node residual criterion is determined by the 99th percentile of the healthy leave-one reconstructed samples in the same working condition interval, and the unit is Celsius.

[0085] For all associated edges of the target measurement point, the module reads the quantile boundary distance and joint covariance distance of the current path response relative to the S3 healthy path allowed domain, and then... Path residuals are formed. This represents the dimensionless average path residual of the edges associated with the target measurement point. This indicates that the target measurement point is connected, currently valid, and the allowed domain is available. The set of directed edges, This indicates the number of valid associated edges in the set. This represents the index of the upstream node of the associated directed edge. This represents the index of the downstream node of the associated directed edge. Indicates the path response component index. Represents the edges calculated by S3 No. Dimensionless quantile boundary distance of each component Represents the edges calculated by S3 Dimensionless joint covariance distance Describes the edges determined by S3. Upper bound of dimensionless joint covariance The dimensionless joint boundary zero protection quantity defined in S3 is adopted. Path residuals are invalid when the value is less than 1.

[0086] The module first calculates the normalized total distance for all currently unremoved paths. Then, remove the target measurement point and its associated edges from the tensor and calculate the remaining path. Node removal recovery rate calculate, This represents the dimensionless distance of all valid paths relative to the allowed domain of healthy paths before removal. This represents the dimensionless total distance of the remaining valid paths after removing the target measurement points. This indicates a dimensionless zero protection quantity, taken as 0.001. The value ranges from 0 to 1, representing the node removal recovery degree. The larger the value, the higher the degree of health and consistency of the remaining paths after removing the measurement point.

[0087] When there are fewer than two valid paths remaining after elimination, the module does not use high recovery as evidence of fault and sets the topology insufficiency flag. When the total distance before elimination is less than zero protection amount, the recovery is recorded as 0 to indicate that there is no overall out-of-bounds error attributable to a single node in the original heat propagation path. All target measurement points perform the same leave-one reconstruction and elimination calculation. During the calculation process, the original thermal response directed graph and the current tensor saved by S3 are not modified.

[0088] The reconstructed parameter records are matched in three levels according to the operating condition interval key, target measurement point number, and graph version. The corresponding coefficient is only used when the current laser power, gas flow rate, and gas pressure fall within the recorded interval, the adjacent set of the target measurement point is consistent with the record, the input variables are all within the calibration coverage range, and the standardized root mean square of the most recently confirmed health verification window is not greater than its 95th percentile threshold. The parameter applicability verification does not use the node residual or path residual of the target measurement point of the current event. The residual of the current event is only used as evidence of the fault. If any independent condition is not met, the most recently qualified parameter version in the same operating condition interval is used and the version rollback flag is set. If the most recently qualified version does not exist, the reconstructed model is output as unavailable and the evidence is insufficient. The original alarm is retained and the state is not interpreted as no fault.

[0089] The module also performs comparisons of node residuals with their health residual criteria and path residuals with their health path criteria in the same unit or dimensionless. When a node residual reaches the criterion, a node out-of-bounds flag is set, and when a path residual reaches the criterion, a path out-of-bounds flag is set. The two flags, the effective proportion of associated edges, and the topology insufficiency flag together constitute the leave-one evidence valid state, enabling S5 to distinguish between reconstruction mismatch, insufficient path evidence, and real anomalies of target measurement points.

[0090] The module writes the node residual sequence, root mean square of node residual, mean of residual, residual slope, residual peak, path residual, node removal recovery degree, number of associated abnormal edges, evidence validity status, working condition interval key, graph version and reconstruction parameter version of each target measurement point into the leave-one diagnostic record, and transmits all target measurement point records and the global working condition consistency of S3 to S5 and S6.

[0091] In this specific embodiment, S5 includes:

[0092] The discrimination and reference update module reads the response direction and fading witness record of S1, the node response and trusted status of S2, the global operating condition consistency of S3, the health path allowed domain version, automatic discrimination qualification and cause code, and the leave-one-out diagnostic record of S4. It strictly checks the automatic discrimination qualification first, then performs overall operating condition consistency discrimination and target measurement point fault discrimination. If the qualification is not discriminable, it directly outputs insufficient data evidence or a final state to be calibrated, retains the original alarm, and prohibits alarm suppression, health domain updates, and non-frame automatic location. Only the S6 frame fault independent location branch is allowed to run. Operating condition consistency criteria... Take the 5th percentile value of the global operating condition consistency of health path events within the same operating condition range. This indicates the lower limit of consistency in dimensionless operating conditions;

[0093] Criteria for the number of non-target temperature measuring points Sure, This indicates the minimum number of non-target credible measurement points required to satisfy the joint response verification. Indicates the number of reliable temperature measurement points available for the current event; only if the S1 response direction is... When the value is -1 or 1, the module checks that at least the number of non-target measurement point response symbols are consistent with the S1 response direction, and checks that the node start delay does not exceed a 2s tolerance in reverse order from upstream to downstream according to the directed thermal response graph. Nodes that do not have a comparable order relationship between topological branches are not included in the reverse order count. The response symbol verification is recorded as direction uncertainty, thus failing to form a deterministic overall operating condition conclusion;

[0094] The ring-down witness verification is performed based on the witness direction mapping record of the same operating condition range during the commissioning phase. The record fields include the combination of thermally driven operating condition quantity direction, ring-down time direction, cavity loss characterization quantity direction, effective amplitude lower limit, number of covered events and version number. The module checks item by item whether the ring-down witness quantity maintains the change direction corresponding to the thermally driven operating condition quantity. When at least one of the current ring-down time or cavity loss characterization quantity reaches the effective amplitude lower limit and the direction is consistent with the record, it is recorded as a successful witness. When the mapping record is not hit or both types of witness quantities are lower than the effective amplitude lower limit, it is recorded as an insufficient witness and the isolated temperature exceeding the limit is directly judged as the actual operating condition temperature change.

[0095] When the aforementioned verification is successful, it is determined to be a true temperature change under actual operating conditions, and an alarm for exceeding the temperature amplitude limit is suppressed. Specifically, when... When the response symbol and arrival order of non-target reliable measurement points are verified and the oscillation witness is established, the module writes the real operating condition event suppression flag and suppression validity period to the temperature amplitude alarm. This indicates the global operating condition consistency of the S3 output; the suppression validity period starts from the event start time and ends when the temperature increment of all reliable temperature measurement points returns to their respective response boundaries and remains within 10 seconds; for step events where the thermally driven operating condition quantity continuously maintains a new platform, when the fluctuation of the operating condition quantity within 30 seconds does not exceed the corresponding instrument resolution and the temperature change rate of all reliable measurement points is within the 95th percentile boundary of the healthy steady-state slope of the same operating condition for 30 seconds, the current suppression and solidification event ends and a new read-only operating condition baseline is established for the detection of the next event only. This baseline is not directly written into the healthy path allowable domain; the maximum duration of a single suppression is 300 seconds. If the price does not fall back or enter steady state after reaching this duration, the original amplitude alarm is immediately restored and the acquisition fault or out-of-model anomaly re-judgment is re-entered. The module records the suppression start time, end time, end reason, re-judgment result and reference boundary version, while retaining the original temperature data, node response and path response;

[0096] When the aforementioned overall verification fails, the module calculates the normalized value of the node residual for each target measurement point. and path residual normalization , This represents the dimensionless nodal residual out-of-bounds ratio. This represents the 99th percentile criterion for healthy node residuals within the same operating condition range, expressed in degrees Celsius. This indicates that the temperature resolution is the same as the unit of zero protection. This represents the dimensionless path residual out-of-bounds ratio. This indicates the 99th percentile criterion for residuals in health-related pathways and is dimensionless. This represents a dimensionless zero protection value of 0.001.

[0097] Node removal and recovery criteria For faulty nodes at marked single measurement points within the same operating condition range, the 5th percentile of the recoverability should be removed, but it must not be lower than the 99th percentile of the health event recoverability. Indicates the first The dimensionless restoring limit for each target measurement point; the lower limit of labeled single-point fault samples used to estimate the 5th percentile value for each operating condition interval is 30. If the actual number of fault samples is less than 30, they will be incorporated into the commissioning phase. Initial criteria will be formed by the controlled injection of single-point fault samples in the same operating condition interval, and a low sample flag will be set. If the total number is still less than 30, then... Mark as unavailable, prohibit automatic location of non-frame faults based on this, and output a status pending calibration or manual review; after 30 samples are collected, recalculate and replace the initial criterion with the actual labeled samples, and record the sample index, replacement time, and criterion version; when , , And when there are at least two valid paths remaining after elimination, locate the first... If a temperature measurement point or its acquisition channel fails, it is written into the fault record to be classified. When multiple targets are met simultaneously, all are retained by S6 and sorted according to the amount of fault evidence. When there is only one adjacent trusted node and the S4 backoff model is enabled, the node residual must simultaneously meet the following criteria: the node residual must exceed the 99.5% quantile criterion of the single adjacent healthy distribution, the node removal recovery must exceed the 5% quantile of the single adjacent fault sample and not be lower than the 99.5% quantile of the healthy recovery, and the residual of the unique associated path must exceed the 99.5% quantile criterion of the single associated edge healthy distribution under the same working condition before it is written into the fault record to be classified. When the unique associated path has no valid allowed domain or does not exceed the anomaly criterion, only the topology insufficiency and manual review status are output. The location of non-frame faults is not automatically determined, and frame continuity is still only used by the S6 independent frame fault branch.

[0098] The module generates mutually exclusive event final states based on the priority of data unavailability or asynchrony, actual operating temperature changes, located acquisition faults, suspected multi-point or model-external anomalies, and unconfirmed events: only when the key input is invalid, When insufficient evidence, inadequate witnessing, unusable reconstruction model, or insufficient path evidence result in both the overall operating condition branch and the local fault branch lacking valid evidence and rollback being impossible, the final state of insufficient data evidence is output. Insufficient witnessing only negates the actual operating condition temperature change branch; if the residual of node S4, the associated path, and the recovery degree are all valid and meet the fault candidate conditions, the final state of the located and collected fault is still output and the candidate is passed to S6. When the overall check is successful, the final state of the actual operating condition temperature change is output. When the overall check is unsuccessful and at least one target meets the fault candidate conditions, the final state of the located and collected fault is output. If the verification fails, or if there is a residual exceeding the limit but no target meets the recovery degree or topology conditions, the output will be a suspected multi-point or out-of-model anomaly final state; otherwise, the output will be an unconfirmed event final state; except for the actual operating condition temperature change final state, the original amplitude alarm will not be suppressed. For insufficient data evidence, suspected multi-point or out-of-model anomalies, and unconfirmed events, the relevant samples will be frozen and health domain updates will be prohibited; S6 only uses the non-frame-type fault candidate of S5 for the fault final state that has been located and collected. For the final state of insufficient data evidence or suspected multi-point or out-of-model anomalies, only frame fault independent location will be performed or a record to be reviewed without a definite non-frame-type location will be generated;

[0099] The healthy path allows the domain to use dual-gated updates of real event gates and trusted path gates. Only after the real operating condition temperature change judgment is established will the module select temperature measurement points and their path responses from the event, whose node residuals and path residuals have not exceeded the corresponding criteria, whose node removal recovery is in the healthy distribution recovery range and are not in a frozen state. The responses are first written to the candidate update cache and then the event label, trusted node list, graph version and operating condition range key are verified. If any field is inconsistent, the batch of cache writing is canceled.

[0100] Joint covariance dimension The sample size is determined by the number of valid numerical components in the heat propagation path consistency tensor of the current operating condition interval, with the lower limit of the sample size based on... Sure, This represents the dimensionless dimension covered by the joint boundary. This represents the target coverage level, and in this specific implementation, it is taken as 0.95. This represents the minimum number of event samples required to update the allowed domain of a healthy path once. After the number of cached samples reaches the lower limit, the newly added real-world events are first divided into the first 80% of the fitted set and the last 20% of the hold-out validation set according to the event time order. The hold-out validation set does not participate in the calculation of quantile boundaries, robust centers, covariance matrices, weights, or consistency parameters. The module only recalculates the candidate boundaries with the time decay weights of the original 80% of the healthy samples and the 20% of the newly added fitted set. When the number of cached samples is less than the lower limit, the existing boundaries remain unchanged.

[0101] When the node residual or path residual of any temperature measurement point exceeds the corresponding criterion, the module immediately freezes the measurement point, its associated edges, and the fault duration window starting from the first time it exceeds the limit. The freeze record includes the measurement point number, associated edge number, freeze start time, trigger criterion, current event number, and domain version. During the freeze period, the relevant data is not written to the healthy sample cache. The freeze is lifted when the node residual and path residual both return to the criterion and three consecutive thermal response events remain qualified, and the node removal recovery rate returns to the 5% to 95% quantile recovery range of the healthy event.

[0102] When multiple target measurement points simultaneously meet the fault candidate conditions, the module does not delete any candidate. Instead, it first sorts them from high to low according to the geometric mean of the two residual out-of-bounds ratios, and then processes them from high to low according to the node elimination and recovery degree. When the sorting value and recovery degree are the same, the module maintains the determined order by the measurement point number in ascending order. Each candidate is written to S6, and S6 combines the acquisition frame information and fault fingerprint to complete the final location and type arbitration.

[0103] After each candidate healthy path allowed domain is formed, the module only uses the last 20% of the holdout validation set in the aforementioned time period to recalculate the global operating condition consistency coverage and the one-sided out-of-bounds rate. When the coverage is not lower than the target coverage level and the out-of-bounds rate of any side does not exceed 5%, the incremental domain version is released. After the release, the holdout event is not backfilled into the fitting set of this round but is transferred to the candidate cache of the next update cycle. When the validation fails, the boundary replacement is revoked, the original domain version is retained, and both the fitting set and the holdout validation set are returned to the cache. The validation result, sample segmentation index, revocation reason and the number of samples to be supplemented are written to the healthy domain update log.

[0104] The module outputs event discrimination records, alarm suppression commands, a list of candidate fault measurement points, freeze or unfreeze commands, and health path permission domain update commands. The event discrimination records include the actual operating condition temperature change flag, witness verification results, quantity criterion results, the two residual out-of-bounds ratios of each target measurement point, node elimination recovery degree, and reference boundary version. The list of candidate fault measurement points and leave-one-out diagnostic records are passed to S6, and the health domain update commands are passed back to the health sample cache of S3.

[0105] In this specific embodiment, S6 includes:

[0106] The fault fingerprint module synchronously reads the node residual sequence and path residual of S4, the node response and path response of S2, the fault candidate measurement point list of S5, and the timestamp and frame number of the acquisition frame. It establishes a cross-event fingerprint buffer for each target measurement point. The buffer fields include measurement point number, operating condition interval key, continuous event number, residual feature, response feature, associated edge feature, frame continuity feature, evidence quantity, confidence level, and boundary version. The buffer is updated on a rolling basis according to the most recent 20 hot response events, and frozen events are only used for fault identification and do not enter the health boundary in reverse.

[0107] DC component of nodal residuals calculate, Indicates the first The DC component of the residual at the current event node of each target measurement point, in degrees Celsius, and the cumulative slope. The slope obtained by robustly fitting the DC components of the four most recent consecutive thermal response events within the same operating condition interval to the event sequence number is calculated using Theil-Sen, with units expressed in degrees Celsius per event. If there are fewer than four consecutive events, the cumulative slope is set to unusable, and the bias or drift determination based on this cumulative slope is invalid. Although the cumulative slope is not used for loose installation contacts, the subsequent rules require four consecutive events; if there are fewer than four, the evidence for the output type is insufficient. The cumulative slope is not used for thermal hysteresis; it is determined solely by the rule for three consecutive events within the same operating condition interval, as specified later. If there are fewer than three consecutive events, the evidence for the output type is insufficient, and the thermal hysteresis type determination is invalid. The number of isolated pulses... The number of sampling points whose absolute residual value exceeds three times the node residual criterion and whose two preceding and following sampling points both return to the criterion is taken. This indicates a dimensionless count. The pulse proximity window is fixed as a 5-point window with each isolated pulse sampling point as the center, two sampling points before and after it, and the center point included. When it encounters the boundary of the event observation window, it is truncated according to the boundary. If there are fewer than 3 valid points, the pulse window is invalid.

[0108] The module maps the DC component continuously across the corresponding residual criterion to a bias fingerprint, specifically according to... If and only if Three consecutive events occur, and the cumulative absolute value of the slope does not exceed the 95th quantile boundary of the healthy slope. Indicates the first The offset fingerprint of each target measurement point is established and is set to 0 or 1. S5 uses the same working condition interval node residual criterion and the unit is Celsius. When the condition is not met, the bias fingerprint flag is 0. When the cumulative slope of four consecutive thermal response events keeps the same direction and crosses the 99% quantile boundary of the healthy slope, it is mapped to a drift fingerprint. When the number of isolated pulses in a single event is not less than 1 and the residual of the associated edge path only crosses the boundary within the pulse's neighboring window, it is mapped to a mutation fingerprint.

[0109] The drift and mutation rules are written as follows: When the cumulative slopes of the most recent 4 events have the same sign and ,as well as If and only if , and , This indicates the establishment of a drift fingerprint. This represents the 99th percentile boundary of the cumulative health slope under the same working conditions, expressed in degrees Celsius per event. This indicates the establishment of a mutation fingerprint. Indicates the target measurement point The effective set of neighboring isolated pulses for the current event. This represents the pulse window index in the set. Indicates the first The maximum dimensionless local path residual among all effective associated edges of the target measurement point after recalculation according to the S2 time delay difference definition within the pulse proximity window. This indicates the 99th quantile criterion for healthy local path residuals under the same operating conditions, and it is dimensionless. This represents the maximum dimensionless path residual of all valid associated edges of the target measurement point in the rolling window of the remaining 5 points outside the window of all valid pulses of the current event. This indicates that the residual of the path outside the healthy window under the same working condition is at the 99th percentile and is dimensionless; therefore, the mutation fingerprint must simultaneously satisfy at least one valid pulse window exceeding the limit and all pulse windows not exceeding the limit during the rolling window. If either condition is missing, All establishment flags are set to 0 or 1, and the first establishment event, the number of continuous events, the triggering characteristics, and the corresponding health boundary are stored.

[0110] When frame numbers are consecutive and timestamp intervals are within the 2.5% to 97.5% quantile boundary of the healthy acquisition cycle, the module executes the thermal hysteresis rule and the installation contact loosening rule respectively. The associated edges include all valid incoming and outgoing edges with the target measurement point as upstream or downstream. The absolute deviation of the current edge delay difference minus the median delay difference of the healthy edge in the same direction is taken as the unified delay anomaly, so that the end measurement point can use the incoming edge. The thermal hysteresis rule is fixed to check the three most recent consecutive events under the same operating conditions. It requires that the time constant of the target measurement point and the delay anomaly of at least one valid associated edge increase strictly with each event, and that the third event crosses the 99% quantile boundary of the healthy sample under the same operating conditions. It also requires that the range of the residual DC component of the three events is not greater than the stability boundary. , This represents the 95th percentile of the range of the DC component residuals of any three consecutive events in a healthy sample under the same operating condition, expressed in degrees Celsius. Thermal hysteresis is indicated when all conditions are met. The installation contact loosening rule is used to verify the four most recent consecutive events within the same operating condition interval, without sharing the premise of continuous increase. The time constant difference between adjacent events is set to... , Indicates the first The target measurement point The time constant difference between each event and the previous event, expressed in seconds. This represents a continuous thermal response event index, requiring that the absolute value of the differences between three adjacent events is greater than the time constant difference boundary. And the signs of adjacent differences alternate. This represents the 99th percentile boundary of the absolute value of the adjacent differences in the time constant of healthy samples under the same working condition, in seconds. Simultaneously, it requires that the residual exhibit at least one valid isolated pulse, and that the local associated edge time delay difference be recalculated within each valid pulse neighboring window according to the time delay difference definition of S2. At least one valid associated edge's local time delay anomaly must cross the 99th percentile boundary of the local time delay anomaly of healthy samples under the same working condition, while this edge must not cross this boundary in any of the other 5 rolling windows outside the current event's valid pulse neighboring windows. Only when the above conditions are met is the time delay anomaly considered to have occurred within the pulse neighboring window only, thus marking the installation contact as loose and outputting a status pending on-site verification. If the number of consecutive events is insufficient, the pulse window is invalid, the time constant has low confidence, there are insufficient local valid points on the associated edge, or any rule condition is missing, only the thermal response anomaly is retained, and the type of evidence is insufficient or pending on-site verification is output. If there is no valid associated edge, the type of evidence is insufficient without skipping the stateless state, thus not relying on uncollected mechanical disturbance periods for input.

[0111] Within the same operating condition range, the module determines the 2.5% to 97.5% healthy node gain boundary for each measuring point based on the normalized gain of at least 80 recent reliable temperature measuring point events. When the normalized gain of the target measuring point across thermal response events is outside this boundary for three consecutive events and the response sign, start delay, and frame continuity are all normal, it is mapped as a gain offset fingerprint. When there are fewer than 80 healthy samples or the operating condition range key changes, the module does not make a cross-range judgment but retains the observation flag.

[0112] Before entering the continuous calculation, the acquired frames are stably sorted according to the receiving order and timestamp. The frame sequence number uses a 32-bit unsigned cyclic counter and is set to... , The frame sequence number counting modulus is used only when the timestamp is strictly incrementing and the current frame is marked by the protocol layer as a normal continuation or a single wraparound continuation. and Calculate continuity, Indicates the first The number of missing frames between adjacent valid frames in each acquisition channel, and is a non-negative dimensionless integer. Indicates the first The first channel The frame sequence number of each valid frame. This indicates the timestamp interval between adjacent valid frames, in seconds. Indicates the first The first channel Each valid frame has a timestamp in seconds. Indicates the acquisition channel index. This represents the index of valid frames after sorting. When encountering frame number reset, out-of-order, duplicate frames, multiple undeterminable wraparounds, or timestamp rollbacks, the above two formulas are not calculated. The frame resynchronization flag is set and the frame is prohibited from directly participating in the gap number determination. The calculation is reopened after two consecutive normal continuation frames are completed. The health continuity boundary is taken as the 99th percentile value of the health acquisition cycle distribution of the same channel. When the frame sequence number gap is greater than 0 or the timestamp interval crosses the boundary, and the temperature remains at the previous sampling value during the gap or when the gap is restored and updated, an intermittent disconnection fingerprint is generated.

[0113] For each type of fingerprint, calculate the normalized out-of-bounds distance relative to the corresponding healthy boundary based on the path response features, node residual features, cross-event node response features, and acquisition frame features. The bilateral boundaries are then calculated as follows: calculate, Representation of features The dimensionless normalized out-of-bounds distance, This represents the current path response, node residual, cross-event node response, or acquisition frame feature, while retaining its own unit. and These represent the lower and upper boundaries of the same unit, determined by the corresponding healthy samples, respectively. This represents the zero protection level at the same unit resolution. The upper limit on one side is the difference between the excess amount and the healthy median value. All normalized distances are truncated to 0 to 2 and the larger the value, the higher the degree of abnormality.

[0114] For the duration of consecutive repetitive values, the module counts the cumulative number of sampling points where the temperature reading is exactly the same as the previous frame and the frame number is consecutive, and then sorts them according to... Get the duration. Indicates the first The duration of consecutive repeating values ​​at each measuring point, in seconds. Indicates the number of consecutive repeated sampling points. S1 represents the sampling period defined by S1. When this duration exceeds the 99th percentile boundary of the healthy repetition length and a frame sequence number gap or timestamp interval exceeds the limit in the same period, it is used as an additional rule for the intermittent disconnection fingerprint and the corresponding normalized distance of the collected frame features is added.

[0115] Fault fingerprint The amount of evidence according to calculate, This represents a dimensionless measure of fingerprint evidence, ranging from 0 to 1.4, with larger values ​​indicating stronger evidence. This index indicates the fault type, including bias, drift, abrupt change, thermal hysteresis, loose mounting contact, gain offset, or intermittent disconnection. This represents the average normalized distance of the effective features of the fingerprint, truncated to 0 to 2. This indicates the number of consecutive events that confirm the fingerprint. This indicates the number of anomalous edges that are related to the target measurement point and support the fingerprint. This represents the proportion of abnormal edges when the total number of valid associated edges to the target measurement point is zero. Items are set to 0 and other evidence items are retained for calculation; the witness mapping table is set with atomic predicates fixed according to fault type: the bias adopts the direction of the associated edge residual and... Same number and frame continuity are normal; drift is analyzed using associated edge residuals across event trends. For identical signs and continuous frames, normal behavior is indicated. Abrupt changes are indicated by at least one associated edge local residual exceeding the limit within the same effective pulse neighboring window and no exceeding the limit outside the entire pulse window. Thermal hysteresis is indicated by the associated edge delay anomaly increasing continuously in the same direction as the time constant and the residual DC component satisfying the stability boundary. Loose installation contact is indicated by the associated edge local delay anomaly and the adjacent time constant difference changing alternately in the same event and the anomaly being limited to the same effective pulse neighboring window. Gain offset is indicated by the associated edge normalized gain deviation having the same sign as the target measurement point normalized gain deviation and the response sign being normal to the initial delay. Intermittent disconnection is indicated by the frame sequence number gap or timestamp interval exceeding the limit and the temperature holding value during the same gap or recovery and update after the gap. Each atomic predicate is only counted as the currently available witness quantity when all its inputs and corresponding health boundaries are available. Representation type The number of currently available atomic predicates in the corresponding mapping table. This indicates the number of atomic predicates that are true when checked against the above-mentioned directions, the same event, or the same pulse window relationship. Witnessing ratio The item is set to 0, thereby ensuring the unique reproduction of the witness ratio for the same input; the weights of the four items are determined by maximizing the classification accuracy based on the labeled fault validation set.

[0116] Fault confidence level Sure, Indicates the fault type The confidence level is between 0 and 1. This represents the cumulative empirical distribution function composed of the evidence of historically labeled fault samples of the same type. When there are fewer than 30 historically labeled samples, the empirical distribution formed by the fault samples injected during the debugging phase is adopted and the low sample flag is set. When the confidence level is not lower than 0.95, the fault type is determined; when it is between 0.80 and 0.95, the fault type to be verified is output; when it is lower than 0.80, only the alarm of abnormal measurement point is retained.

[0117] When multiple fault fingerprints are established simultaneously, the module first executes the frame fault independent localization branch: when the frame sequence number gap or timestamp interval overshoot of a certain acquisition channel is combined with the temperature hold value or gap recovery update, the intermittent disconnection is directly located by the measurement point number and acquisition channel number in the frame record. It does not require the S4 residual to be valid or to be pre-entered into the S5 fault candidate list. The node residual and path residual are only used as enhancing evidence. When the intermittent disconnection is established, it is used as the first type and other types are retained as accompanying evidence. When the frames are continuous, thermal hysteresis or loose installation contact is determined based on the time constant and path delay difference. Other non-frame type faults are still required to meet the S5 fault candidate conditions and be sorted from largest to smallest evidence quantity. When the evidence quantity is the same, they are sorted by confidence, number of consecutive events and fault type fixed coding in turn. When the frame identity and residual candidate position conflict, the traceable frame channel identity is used as the intermittent disconnection position and the position conflict pending verification mark is set.

[0118] The module ultimately generates a fault diagnosis result record, which includes the device number, event number, fault measurement point number, acquisition channel number, first fault type, accompanying fault type, evidence quantity, confidence level, number of consecutive events, number of associated abnormal edges, first occurrence time, current freeze status, graph version, health path allowed domain version, and fingerprint boundary version. The result is sent to the status monitoring interface, and the original evidence index is saved to support maintenance personnel in tracing back the corresponding temperature frame, thermally driven operating condition quantity, decay witness quantity, and path response.

[0119] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0120] This invention utilizes thermally driven operating parameters to determine thermal response events and converts the responses of each temperature measurement point into node and path representations related to spatial topology and airflow direction. This allows real-world temperature variations to be confirmed through the combined responses of multiple measurement points and the observation of oscillations, and enables the isolation of single-point or single-channel anomalies through leave-one-out reconstruction and elimination recovery. Therefore, the diagnostic process no longer relies solely on a single temperature amplitude but simultaneously utilizes global thermal propagation consistency and local fault residuals.

[0121] This invention changes the processing structure of dynamic reference and fault identification in terms of path consistency tensor, global operating condition consistency, node removal recovery degree, and dual-gated freeze update; different types of nodes, paths and acquisition frame features adopt corresponding health boundaries, thereby providing technical basis for distinguishing between intermittent disconnection, thermal hysteresis, loose contact and gain offset while keeping the reference data free from suspected fault contamination.

Claims

1. A fault diagnosis method for decaying cavity temperature acquisition based on thermal response consistency, characterized in that, include: S1. Simultaneously acquire temperature measurement point sequences, thermal driving condition quantities, and decay witness quantities at at least three spatial locations of the decay cavity, detect the joint changes of thermal driving condition quantities, and determine thermal response events, their response directions, and event quantization values. S2. Construct a directed thermal response graph based on the spatial location of the temperature measuring points and the airflow direction. Normalize the temperature increment of each temperature measuring point according to the event quantization value, extract the node response, and generate the path response from the differences in the node responses of adjacent temperature measuring points. include: Using temperature measurement points as nodes, directed edges are determined according to airflow direction and spatial adjacency between nodes; The statistical value of the temperature measurement point sequence within the baseline window before the start time is determined as the pre-event temperature baseline. The change in the temperature measurement point sequence relative to the pre-event temperature baseline is used as the temperature increment. The temperature increment is divided by the event quantization value to obtain the normalized temperature increment. The response symbol, start delay, time constant, normalized gain, and recovery slope are extracted from the normalized temperature increment as the node response, wherein the start delay is the difference between the time when the temperature increment first crosses the response boundary determined by the pre-event noise distribution and the start time, and the normalized gain is the response amplitude of the normalized temperature increment. For each directed edge, calculate the initial delay difference between the downstream node and the upstream node, the normalized logarithmic ratio of the gain, and the normalized temperature increment waveform distance, as the path response; S3. Organize the path response according to the topological order of the directed graph of thermal response to obtain the thermal propagation path consistency tensor, and compare it with the healthy path allowable domain of the current operating condition interval to obtain the global operating condition consistency. include: The operating condition range is divided according to at least one of the laser power range, gas flow rate range, and gas pressure range. During the equipment commissioning or calibration phase, based on the path response samples corresponding to the reliable temperature measurement points in the confirmed thermal response events without acquisition faults, the initial quantile boundaries and initial joint covariance boundaries of each operating condition interval are determined respectively. During the online operation phase, the path response samples and the path response samples corresponding to the credible temperature measurement points in the confirmed real operating conditions are used together as healthy path response samples. The quantile boundaries and joint covariance boundaries of the node start delay difference, normalized gain logarithmic ratio and normalized temperature increment waveform distance are determined respectively to form the healthy path allowable domain. The global operating condition consistency is synthesized by combining the proportion of path components located within the quantile boundary and the joint covariance boundary in the current heat propagation path consistency tensor with the normalized distance of each path component to the corresponding boundary. S4. For each target temperature measurement point, perform leave-one reconstruction based on the remaining reliable temperature measurement points, thermal driving conditions, and thermal response directed graph to obtain node residuals and path residuals. Calculate the node removal recovery degree after removing the target temperature measurement point and its associated edges. include: Based on the node responses of upstream reliable temperature measurement points pointing to the target temperature measurement point and downstream reliable temperature measurement points pointed to by the target temperature measurement point in the directed thermal response graph, the reconstructed temperature increment of the target temperature measurement point is generated in combination with the current thermal driving condition. The node residual is formed by the difference between the measured temperature increment of the target temperature measurement point and the reconstructed temperature increment. The path residual is formed by the distance between the current path response of the edge associated with the target temperature measurement point and the boundary corresponding to the healthy path allowable domain. Before and after removing the target temperature measuring point and its associated edge, calculate the total normalized distance from the unremoved path to the allowed domain of the healthy path, and use the reduction ratio of the total normalized distance as the node removal recovery degree. S5. Based on the global operating condition consistency, fading witness quantity, node residual, path residual and node removal recovery degree, determine the actual operating condition temperature change or measurement point acquisition failure, and control the update of the healthy path allowable domain accordingly. S6. Generate fault fingerprints based on node residuals and path residuals, and output fault location and fault type.

2. The method for fault diagnosis of decaying cavity temperature acquisition based on thermal response consistency according to claim 1, characterized in that, S1 includes: Laser power and at least one of gas flow rate and gas pressure are used as the thermally driven operating condition quantity, and at least one of the ring-down time and cavity loss characterization quantity is used as the ring-down witness quantity. For each thermally driven operating condition quantity, calculate the direction of change, magnitude of change, and duration of change relative to the pre-event operating condition baseline. Combine the thermally driven operating condition quantity changes that meet the condition of proximity of the starting point of change within the same detection window into a single thermal response event. The earliest starting point of the merged change is taken as the starting time of the thermal response event. The response direction is determined by weighting the change direction of each thermally driven operating condition quantity according to its historical thermal response sensitivity. The event quantization value is determined by weighting the change amplitude of each thermally driven operating condition quantity according to the normalized weighted result of the historical thermal response sensitivity.

3. The method for fault diagnosis of decaying cavity temperature acquisition based on thermal response consistency according to claim 1, characterized in that, In step S5, the discrimination is performed in the following order: first, the global operating condition consistency is compared with the operating condition consistency criterion determined by the quantile value of the healthy path response sample, and then it is checked whether the non-target temperature measurement points that have reached the quantity criterion determined by the number of credible temperature measurement points maintain the response sign corresponding to the response direction and the arrival order along the directed graph of the thermal response, and whether the decay witness quantity maintains the change direction corresponding to the thermal driving operating condition quantity. When the aforementioned verification is successful, it is determined to be a true temperature change under actual operating conditions, and the alarm for exceeding the temperature amplitude limit is suppressed. If the aforementioned verification fails, the node residuals and path residuals of each target temperature measuring point are compared with the residual criteria determined by the boundary of the healthy path allowable domain. When both exceed the corresponding residual criteria and the node removal recovery indicates that the remaining path returns to the healthy path allowable domain after removal, the target temperature measuring point or its acquisition channel is located as faulty.

4. The method for fault diagnosis of decay cavity temperature acquisition based on thermal response consistency according to claim 3, characterized in that, The reference update constraints for the health path allowed domain include: Only after the actual working condition temperature change judgment is established, reliable temperature measurement points and their path responses that have not exceeded the corresponding residual criterion and are not in a frozen state are selected as update samples. The dual-gated update is performed in the following order: real-world temperature change discrimination, reliable temperature measurement point screening, and updated sample writing. When the number of updated samples through the dual gating in any operating condition interval is less than the lower limit of the number of samples determined by the joint covariance dimension and the target coverage level, the updated samples are cached and the existing healthy path allowed domain of the operating condition interval is maintained until the number of cached updated samples reaches the lower limit of the number of samples before the corresponding boundary is updated. When the node residual or path residual of any temperature measurement point crosses the corresponding residual criterion, the data of that temperature measurement point, its associated edges, and the fault duration window from the time of crossing to the time when the recovery criterion is established is frozen. The freezing is lifted after the node residual and the path residual both return to the corresponding residual criterion and the node removal recovery degree returns to the recovery interval corresponding to the healthy path response sample distribution.

5. The method for fault diagnosis of decay cavity temperature acquisition based on thermal response consistency according to claim 4, characterized in that, S6 include: Synchronously acquire the acquisition frame information consisting of frame timestamp and frame sequence number; The DC component, cumulative slope, and number of isolated pulses are extracted from the node residuals. The start delay, time constant, normalized gain of the thermal response event, and path delay difference are extracted from the node response and the path response. The frame sequence number gap, timestamp interval, and duration of continuous repetition value are extracted from the acquired frame information. The DC component continuously crossing the corresponding residual criterion is mapped as a bias fingerprint; the cumulative slope maintaining the same direction in continuous thermal response events is mapped as a drift fingerprint; the isolated pulse is mapped as a sudden change fingerprint; and the continuous increase of the time constant and the path delay difference under frame continuity conditions is mapped as a thermal hysteresis or loose installation contact fingerprint. The normalized gain sample of the cross-thermal response event of the reliable temperature measurement point within the same operating condition range is used to determine the gain boundary of the healthy node, and the deviation of the normalized gain of the cross-thermal response event of the target temperature measurement point from the gain boundary of the healthy node is mapped as a gain offset fingerprint.

6. The method for fault diagnosis of decay cavity temperature acquisition based on thermal response consistency according to claim 5, characterized in that, The combination of a frame sequence number gap or timestamp interval crossing the continuity boundary determined by the healthy acquisition cycle distribution, and the temperature measurement point sequence maintaining the previous sampled value during the gap or resuming the update after the gap, is mapped as an intermittent disconnection fingerprint. For path response features, a normalized distance relative to the boundary of the allowed healthy path domain is used; for node residual features, a normalized distance relative to the residual criterion determined by the residual distribution of healthy nodes is used; for node response features across thermal response events, a normalized distance relative to the healthy node response boundary determined by the historical node response samples of reliable temperature measurement points is used; and for acquisition frame features, a normalized distance relative to the continuity boundary is used. The evidence quantity is determined based on the normalized distance, number of consecutive events, number of associated abnormal edges, and number of mutually consistent witnesses corresponding to each fault fingerprint, and the fault confidence is determined based on the empirical distribution of the evidence quantity in historically labeled fault samples. When multiple fault fingerprints are established simultaneously, intermittent disconnection is first determined based on the collected frame information, and then thermal hysteresis or loose installation contact is determined based on the time constant and path delay difference under continuous frame conditions. Other fault types are output in descending order of the corresponding amount of evidence.

7. A fault diagnosis system for decaying cavity temperature acquisition based on thermal response consistency, used to execute the fault diagnosis method for decaying cavity temperature acquisition based on thermal response consistency as described in any one of claims 1 to 6, characterized in that, include: The data acquisition and event detection module is used to synchronously acquire temperature measurement point sequences, thermally driven operating conditions, and decay witness quantities, and to determine thermal response events. The response graph construction module is used to construct a directed thermal response graph and generate normalized temperature increments, nodal responses, and path responses. The path consistency analysis module is used to generate the heat propagation path consistency tensor and determine the global operating condition consistency. The Leave-One Reconstruction module is used to generate node residuals, path residuals, and node removal recovery. The discrimination and reference update module is used to discriminate between actual operating temperature changes or measurement point acquisition failures, and to control the update of the health path allowed domain. The fault fingerprint module is used to generate fault fingerprints and output the fault location and fault type.

Images