Test equipment fault identification method based on diagnostic semantic adaptation

Abstract

This invention provides a fault identification method for test equipment based on diagnostic semantic adaptation, belonging to the field of test equipment operation status monitoring technology. It includes: S1, collecting multi-measurement channel sensor signals and image information generated during the test and performing preprocessing; S2, performing measurement channel mapping, unit conversion, range matching, and normalization processing on multi-source asynchronous data; S3, determining the fault probability of the test equipment, obtaining the fused fault confidence level, and generating a structured test equipment fault diagnosis record package; S4, analyzing the fault types of the test equipment and executing the test equipment fault identification alarm program. This invention uses an edge-side temporal fault discrimination network to perform real-time fault screening on a unified state vector, and employs a multimodal transfer diagnostic network for fault verification and incremental learning, forming a parameter optimization closed loop to improve the accuracy and robustness of fault identification.

Description

Technical Field

[0001] This invention relates to the field of testing equipment operation status monitoring technology, specifically to a testing equipment fault identification method based on diagnostic semantic adaptation. Background Technology

[0002] As testing equipment evolves towards intelligence and unmanned operation, various types of testing equipment, such as hydraulic testing benches, motor testing benches, and pneumatic testing platforms, all require real-time monitoring of their operating status, fault identification, and remote alarm capabilities. However, existing technologies still have many limitations in terms of multimodal monitoring, cross-platform compatibility, and the reliability of remote alarms.

[0003] Currently, most testing equipment still relies on traditional threshold monitoring or single data sources for fault diagnosis. Alarm elements are mostly simple text or audio-visual prompts, lacking rich media information support such as images and trend curves, making it difficult to accurately determine the nature and location of faults in remote scenarios. Furthermore, differences in communication protocols, range settings, and data structures between various testing devices mean that monitoring systems often can only be used with specific models of testing equipment, resulting in poor system portability and difficulty in meeting the engineering requirements of adapting a single system to multiple testing devices. For example, patent publication number CN112983932B discloses a data acquisition and safety measurement and control scheme for a hydraulic testing bench. Although it implements the basic operation monitoring and control logic of the testing equipment, its technical solution is mainly aimed at a single hydraulic testing bench architecture and does not solve the problem of rapid adaptation and universal monitoring between different types of testing equipment. For example, US patent number 11,688,273 B2 discloses a monitoring and alarm method based on video streams, which can realize image-based alarm triggering. However, this solution does not involve multimodal fusion of sensor data and image data, nor does it have an alarm adaptive mechanism under bandwidth-limited conditions, and it lacks collaborative optimization of edge testing equipment and cloud joint learning.

[0004] Therefore, existing technologies have not yet formed a general intelligent monitoring system for test equipment that can be rapidly deployed across test equipment, integrate multimodal data for real-time diagnosis, and support rich media adaptive alarms and edge-cloud collaborative learning. It is necessary to propose new technical solutions to improve the diagnostic accuracy, alarm reliability, and system versatility of test equipment under unattended conditions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention aims to provide a fault identification method for test equipment based on diagnostic semantic adaptation. This method utilizes an edge-side temporal fault discrimination network to perform real-time initial fault screening on a unified state vector. A cloud-based multimodal migration diagnostic network is then employed for fault verification and incremental learning. Updated differential parameters are distributed to the edge-side temporal fault discrimination network, forming a closed-loop parameter optimization mechanism that integrates edge and cloud capabilities. This improves the accuracy and robustness of fault identification while maintaining real-time performance. Furthermore, through an intermediate adaptation layer and a self-describing parameter set mechanism for test equipment, automatic identification and parameter loading for various models and communication protocols of test equipment are achieved. This allows the same monitoring and diagnostic software to be quickly deployed across multiple types of test equipment, enhancing portability and versatility.

[0006] Specifically, the present invention provides a method for fault identification of test equipment based on diagnostic semantic adaptation, which includes the following steps: S1: The test equipment is used to collect multi-channel sensor signals and image information output by the image acquisition unit during the test, and the multi-source asynchronous data is processed using a unified time base; the multi-channel sensor signals include the load of the test equipment. Sensor signals, temperature of test equipment Sensor signals and vibration of test equipment Sensing signals; S2: Obtain the multi-source asynchronous data processed in step S1, based on the self-describing parameter set of the experimental equipment. The process involves protocol parsing, unit conversion, range verification, rate of change verification, and semantic mapping of measurement channels for the raw sampled values ​​of each communication protocol, channel number, and dimension; and will be integrated with the load of the test equipment. Temperature of test equipment Vibration of test equipment The relevant measurement channels are bound to the main measurement channel and normalized respectively; they are uniformly mapped to a standardized data structure with consistent semantic positions under a general signal structure. Construct a unified state vector for the experimental equipment Output the main variable applicability identifier vector. and measurement channel validity identifier set ; S3: Based on the unified state vector and measurement channel validity sequence output in step S2, perform timing fault discrimination within the sliding time window to obtain the failure probability of the test equipment. When the probability of equipment failure... After the initial screening trigger conditions are met, keyframes are matched according to timestamps and image features of key monitoring areas are extracted. The image difference metric relative to the baseline image is then calculated. The failure probability of the test equipment is determined by a unified fusion function. Mapped to the real number domain and compared with visual difference measures By performing unified-scale fusion, the fusion failure confidence of the test equipment is obtained. Once the fused fault confidence level meets the fault establishment criteria, a structured test equipment fault diagnosis record package is generated. S4: Perform fault type analysis on the obtained unified state vector, fault confidence, and image information to obtain the fault type index. Fault assessment parameters and fault risk level ;Execute the test equipment fault identification and alarm procedure.

[0007] Preferably, step S2 specifically includes: S21: Obtain the self-describing parameter set of the test equipment Through the protocol parsing function The raw data in step S1 is parsed into a vector of raw signal sample values. Physical quantity verification is performed to obtain the validity identifier of the measurement channel; S22: Define the cutoff function Obtain the standardized vector of the measurement channel of the test equipment. Perform dimensional unification mapping and normalization standardization on the physical quantities of each measurement channel; S23: Output the main variable applicability identifier vector of the test equipment according to the consistency rule. Includes: Validation results of load host variables Validation results of temperature main variable Results of validation of vibration main variables .

[0008] Preferably, the dimensionless mapping in step S22 specifically involves: ; ; ; in, A standardized data structure under a general signal structure; For measuring the channel mapping matrix; Standardized vectors for measuring channels of the test equipment; Index for discrete sampling time; Index of the main measurement channels related to the load of the test equipment; Index of the main temperature measurement channel of the test equipment; Index of the measurement channels for the original vibration quantities of the test equipment; For mapping rules; For the self-describing parameter set of the test equipment; Discrete sampling time The corresponding equivalent load of the test equipment; Standardize vector elements for the load measurement channel of the test equipment; The effective pressure-bearing area of ​​the actuator.

[0009] Preferably, step S3 specifically includes: S31: Obtain the validity identifier of the measurement channel. Then, the effective sequence of synchronous construction and time window synchronization is achieved. Use an edge-side timing fault discrimination network Obtain the failure probability of the test equipment Perform edge-side initial screening to determine fault triggering; S32: When the initial screening fault on the edge side is triggered, key frames are obtained by matching the timestamps, resulting in a set of images of the key monitoring area; operators are extracted based on the image feature vectors. Extract and fuse images of key monitoring areas to obtain a single image feature vector. The fusion fault confidence level is obtained using a unified fusion function. Perform fault condition determination; S33: Once it is determined that the fusion fault confidence level meets the fault establishment conditions, a structured test equipment fault diagnosis record package is generated.

[0010] Preferably, in step S32, the unified fusion function determines the fusion failure confidence level of the test equipment as follows: ; ; ; in, Discrete sampling time The corresponding fusion fault confidence level; To standardize the failure confidence level; The fusion weights are for the fault probability branch on the sensing side; A function that maps fault probability values ​​to the real number field. This represents the probability of equipment failure during testing. For visual enabling indicators; The fusion weights for the visual difference branches; For measuring image differences in key monitoring areas; For bias terms; It is the natural logarithm function; For input feature variables.

[0011] Preferably, step S4 specifically includes: S41: Upload the fault diagnosis record package to the cloud-based multimodal migration diagnostic network. Output the failure probability vector for each failure category. Select the index with the highest probability. Determine the initial fault type; S42: Construct out-of-limit parameters for physical quantities based on the deviation of the uniform state vector of the test equipment. Determine the parameters for constructing a fault assessment. Analyze the fault risk level Division; S43: Establish a mirror copy of the edge network structure that is consistent with the cloud-based structure, and perform periodic incremental training using confirmed valid samples to obtain updated edge network parameters. ; S44: Triggers rich media alarm communication and executes the test equipment fault identification alarm program.

[0012] Preferably, the fault assessment parameters in step S42 for: ; in, For fault assessment parameters; Prior severity weights for fault types; Fault type; For fault confidence weights; To integrate fault confidence; For physical over-limit weights; The physical quantity parameter that exceeds the limit corresponding to the current fault diagnosis record; It is a nonlinear saturation mapping function.

[0013] Preferably, step S44 specifically includes: Based on fault type index Fault assessment parameters and fault risk level The rich media alarm communication of the triggering test equipment enables adaptive selection and delivery of alarm elements; based on the fault risk level... Pre-set alarm element priority rules: fault risk level For a Level 1 fault, send a text description, keyframe images, and voice prompts to provide evidence of the fault location; fault risk level. For a level 2 fault, send a text description, a compressed thumbnail of the keyframe, and a summary of the trend curve; fault risk level. In the case of a Level 3 fault, only necessary text alarm messages are sent; determine the maximum data payload that can be sent at the current moment. The candidate alarm elements corresponding to the current fault risk level are identified as having a total of Optimize the fault alarm element group using candidate elements.

[0014] Preferably, the optimization of the fault alarm element group in step S44 is as follows: ; in, To find the maximum value of the function; For the first The information utility value of each candidate alarm element; The data size of this candidate alarm element; For the first Whether each candidate alarm element is selected into the current alarm package; This represents the total number of current candidate alarm elements; Use constraint symbols for modeling; Index for candidate alarm elements; This represents the maximum data payload allowed for transmission.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention achieves automatic identification and parameter loading of various models and communication protocols of test equipment by using an intermediate adaptation layer and a test equipment self-description parameter set mechanism, combined with a general signal structure and a modular protocol plug-in framework. This enables the same monitoring and diagnostic software to be quickly ported and deployed among multiple types of test equipment, improving portability and versatility.

[0016] (2) This invention uses an edge-side temporal fault discrimination network to perform real-time fault screening of a unified state vector, and generates a fault diagnosis record package after a fault is established and uploads it to the cloud. The cloud-based multimodal transfer diagnosis network is used for fault verification and incremental learning. The updated differential parameters are sent to the edge-side temporal fault discrimination network to form a parameter optimization closed loop of edge-cloud collaboration, which improves the accuracy and robustness of fault identification while taking into account real-time performance.

[0017] (3) This invention uses rich media alarm communication to adaptively select alarm elements such as text, image summary, curve information and voice prompts according to the fault risk level and network bandwidth. It also combines hierarchical alarm strategy, receipt record and multi-measurement channel redundant push mechanism. At the same time, it uses data storage and visualization to uniformly archive and display operation and alarm information. Even in weak network and unattended scenarios, it can still achieve reliable delivery and closed-loop management of alarm information, and improve maintainability and availability. Attached Figure Description

[0018] Figure 1 This is a control block diagram for a test equipment fault identification method based on diagnostic semantic adaptation. Figure 2 A flowchart of the intermediate adapter layer provided by the present invention; Figure 3A flowchart illustrating the rich media alarm communication process provided by this invention; Figure 4 This is a waveform diagram collected under normal operating conditions in the vibration channel of a certain production line. Figure 5 The waveforms collected under the condition of jamming in the vibration channel of a certain production line are shown. Figure 6 Vibration time-domain waveform diagram for a complete discrimination window under normal operating conditions; Figure 7 Vibration time-domain waveform diagram for a complete discrimination window of the stuck working condition; Figure 8 The diagram shows the discrimination output results of the edge-side timing fault discrimination network for the test segments of the two operating condition windows. Detailed Implementation

[0019] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0020] This embodiment proposes a fault identification method for test equipment based on diagnostic semantic adaptation. Taking a hydraulic test bench as an example, this invention can also be applied to motor test benches, pneumatic test platforms, power system test devices, and other test equipment capable of status monitoring through sensor data and image data. To facilitate unified modeling and portable deployment across test equipment, this embodiment selects three types of engineering physical quantities that are common to most test equipment and can be processed throughout the entire process as unified input variables. Specifically, these are: test equipment load. The load strength borne by the test equipment can be obtained from equivalent values ​​of pressure, thrust / torque; the temperature of the test equipment. Indicates the temperature of critical media or critical components; vibration of testing equipment. The mechanical vibration intensity of the test equipment is represented using a standardized notation. Furthermore, image keyframes are introduced. As a visual modal input, it supports the identification of visual faults such as leakage, splashing, and component displacement; the embodiment also requires load on the test equipment. Temperature of test equipment Vibration of test equipment This indicates that the corresponding engineering physical quantities have undergone dimensional unification and normalization processing to obtain the normalized test equipment load. Normalized test equipment temperature Vibration of normalized test equipment And used to construct the unified state vector of the test equipment. The portability mentioned in this manual refers to the conversion of raw data from test equipment of different models, communication protocols, and range settings into a standardized data structure and unified state vector that is semantically, dimensionally, and structurally consistent, through an intermediate adaptation layer, test equipment self-describing parameter sets, measurement channel mapping rules, and a general signal structure. This enables the same monitoring, diagnosis, and alarm software to be quickly deployed across devices simply by loading parameters, switching protocol parsing functions, and binding master variables, without having to rewrite the core fault identification process.

[0021] In this embodiment, single-channel vibration data collected online at a hydraulic valve station on a production line is selected as the data source. The vibration data is acquired by a single-axis accelerometer mounted on the outer wall of the valve body. To verify the feasibility of this invention under single-channel conditions, a complete discrimination window is selected for both normal and stuck operating conditions, with each discrimination window containing 12,000 continuous sampling points. This embodiment does not elaborate on the multi-class training sample construction process, but only utilizes the two real operating condition windows mentioned above to verify that the original vibration data input in step S1, the intermediate adaptation layer unit conversion and standardized output in step S2, and the edge-side fault discrimination process in step S3 can all be executed and can distinguish between normal and stuck operating conditions; the specific implementation process is as follows... Figure 1 As shown, the process involves collecting multi-channel sensor signals and image information generated during the experiment, performing preprocessing; mapping measurement channels, converting units, matching ranges, and normalizing multi-source asynchronous data; determining the probability of equipment failure, obtaining fused fault confidence, and generating a structured equipment fault diagnosis record package; analyzing the fault types of the equipment and executing the equipment fault identification and alarm program; specifically, the process includes the following steps: Step S1: Use the test equipment to collect the multi-measurement channel sensor signals and image information output by the image acquisition unit generated during the test, and process the multi-source asynchronous data with a unified time reference.

[0022] Step S11: Set the standard sampling period for the test equipment as follows: The time base starts at Then the discrete sampling time is: , ; Set the test equipment to exist Each sensing acquisition and measurement channel, at discrete sampling times The vector of original signal sample values ​​acquired ;in, This is the vector of original signal sample values; For the first Each measurement channel at discrete sampling time The original signal sample value, Belongs to 1, 2, ... ; This represents the total number of sensor acquisition and measurement channels. Index for discrete sampling time; For measurement channel index; This is the transpose symbol.

[0023] The raw signal sample values ​​are the original values ​​obtained by restoring the analog-to-digital converter (ADC) count values, register readings, or protocol text data output from the acquisition card. The raw signal sample values ​​from multiple measurement channels should at least include values ​​sufficient for subsequent construction of a unified input variable test equipment load. Temperature of test equipment Vibration of test equipment The relevant measurement channel data.

[0024] To accommodate the differences in measurement channel configurations among various types of testing equipment and the limitations of edge computing power, a full-data acquisition measurement channel set was developed. Data is collected, and a subset of fault diagnosis measurement channels that participate in online diagnostic calculations are pre-defined. Subsequent fault identification uses only a subset of the fault diagnosis measurement channels. The corresponding measurement channel data is collected, while the unselected measurement channels continue to be collected for archiving and backtracking, status display, or rich media curve summary generation.

[0025] Step S12: In actual test equipment, each measurement channel may have different sampling frequencies or use asynchronous reporting methods. To ensure that subsequent steps are based on a unified timestamp... Feature computation and modeling align the discrete values ​​of the asynchronous measurement channel to ensure they are aligned at each discrete sampling time. All can provide the corresponding sample values. ; for the first There are 3 measurement channels, and let the set of their actual arrival discrete sampling times be denoted as . The corresponding set of original signal sample values ​​is The original signal sample values ​​are processed using a nearest neighbor alignment strategy, specifically as follows: ; ; in, For the first Each measurement channel at discrete sampling time The original signal sample value; Discrete sampling times with a unified timestamp; For the first The actual sampling time of each measurement channel; This refers to the actual sampling sequence number of the measurement channel; In order to be in The actual sampling sequence number with the smallest absolute value of its time difference; This is the original signal sample value corresponding to the nearest actual sampling time; When the function reaches its minimum value The value of .

[0026] When improved time alignment accuracy is required and linear interpolation conditions are met, linear interpolation is used: if there exists a condition that satisfies... and If the adjacent valid sampling points are, then the first All measurement channels at the same time The alignment sample value at that location is taken as: ; in, These are the original signal sample values; A preset time alignment tolerance upper limit is used to limit the uniformity of time. The maximum permissible time deviation between the sample point and the adjacent sample point that can be used for interpolation.

[0027] If the above linear interpolation conditions are not met, the process reverts to the nearest neighbor alignment strategy. Specifically, if it is difficult to obtain effective sampling points for nearest neighbor alignment or linear interpolation within the preset alignment tolerance range, the measurement channel is determined to have missing data samples at a unified timestamp, and the measurement channel is marked as missing in the signal data record. This is to facilitate the generation of a validity identifier in subsequent step S2 and to mask the impact of the missing measurement channel on online fault identification calculations in step S3. Through the above formula, the input multi-source asynchronous measurement channel data is organized into a standard sampling period. Periodic, with discrete sampling times The original signal sampling vector after unification of timestamps Sequence output.

[0028] After the timestamp alignment described above, a raw sampling window of uniform length is obtained for both operating conditions. Here, only the first three points of the unified raw signal are shown: Normal operating condition. Stuck working condition .

[0029] Step S13: Output the sequence of image frames from the test equipment and assign a timestamp to each frame. Let the image frame number be... The image frame sequence is The corresponding timestamp is Cache and record the image frames within the window over a recent period. The index relationship. To ensure that image evidence consistent with the time of the fault can be extracted when a subsequent fault is triggered, for any time to be evidenced... If the fault trigger time is output from subsequent steps, the keyframe matching index is set as follows: ; ; in, Match the index for the keyframe; When the function reaches its minimum value The value; Indexed by timestamp; The time is pending evidence collection; These are keyframes for the image.

[0030] will use the standard sampling period For a period of time, with The original signal sampling vector with a unified timestamp And the necessary operational status identifiers, encapsulated as timestamped signal data records. for: ; in, For timestamped signal data records; The raw data payload must contain at least the aligned raw signal sample vectors. and selectively includes extended original data fragments. Expanding the original data fragment Used to store raw text fragments or byte sequences related to vendor interfaces or communication protocols. If extended raw data fragments are not enabled, then... ; For identification of testing equipment; This is an optional indicator of the operating status of the test equipment.

[0031] The final output of this step is: based on a unified timestamp. Organized aligned sequence of original signal sample vectors or in a subset of the fault diagnosis measurement channels A subset sequence of images, a sequence of images. and the relationship between unified timestamp image indexes Signal data records with a unified timestamp .

[0032] Step S2: See Appendix Figure 2 The flowchart of the intermediate adapter layer provided by this invention for performing test equipment self-description, protocol conversion, and signal standardization is as follows: Multi-source asynchronous data processed in step S1 is obtained, based on the test equipment self-description parameter set. The process involves protocol parsing, unit conversion, range verification, rate of change verification, and semantic mapping of measurement channels for the raw sampled values ​​of each communication protocol, channel number, and dimension; and will be integrated with the load of the test equipment. Temperature of test equipment Vibration of test equipment The relevant measurement channels are bound to the main measurement channel and normalized respectively; then they are uniformly mapped to a standardized data structure with consistent semantic positions under a general signal structure. Construct a unified state vector for the experimental equipment Output the main variable applicability identifier vector. and measurement channel validity identifier set The purpose of this step is to shield the underlying differences between various models, communication protocols, and range settings of test equipment, so that subsequent edge-side fault identification always receives input data that is semantically consistent, dimensionally consistent, and structurally consistent.

[0033] The above processing first converts heterogeneous raw channel data into physical quantities with engineering significance based on the self-describing parameter set of the test equipment. Then, according to the mapping rules, load, temperature and vibration information from different sources but with equivalent diagnostic semantics from each device are uniformly bound to the main variables. Among them, the load component achieves semantic unification across pressure, thrust or torque measurement forms through equivalent load transformation, and the vibration component achieves a unified characterization of time-domain vibration energy through vibration intensity extraction within a statistical time window. Finally, the data output by each device is converted into a unified state vector with consistent semantics, dimensions and structure, thereby realizing diagnostic input across test equipment.

[0034] Step S21: Receive the signal data record with a unified timestamp output from step S1. Fault diagnosis measurement channel subset Image indexing relationship with unified timestamps and according to the test equipment markings Retrieve the matching test equipment self-description parameter set from the local cache or cloud-based test equipment self-description repository. Test equipment self-describing parameter set Automatically generated or pre-configured by the test equipment during the initialization phase, including at least: protocol type identifier. Unit conversion factor for each measurement channel Unit conversion bias Upper limit of physical quantity range Lower limit of the range of physical quantities Measurement channel sampling parameters Measurement channel mapping rules and rules for generating key monitoring areas If the identification of the test equipment cannot be obtained... If the matching test equipment self-description parameter set or the text is missing key fields, a preset general template is loaded to complete the minimal protocol parsing and main measurement channel binding, and the test equipment is marked as an unregistered test equipment; at this time, the full data acquisition and archiving functions are retained, but only the alarm strategy based on conservative thresholds is executed.

[0035] Based on the protocol type identifier in the self-description parameter set of the test equipment Dynamically load the corresponding protocol parsing function And combined with the measurement channel sequence correction rules The raw data payload in step S1 is parsed into a vector of raw signal sample values ​​arranged in a fixed order according to the measurement channels. for: ; in, This is the vector of original signal sample values ​​after protocol parsing and sequence correction; The raw data payload output in step S1; For optional extension of the original data fragment; for protocol type The corresponding protocol parsing function; The measurement channel sequence correction function is used to ensure that the data output by each test device has a consistent channel sequence within step S2.

[0036] By encapsulating protocol differences in Internally, all protocols, such as Modbus, CANopen (a high-level communication protocol based on CAN bus), EtherCA (Ethernet Control Automation Technology), and vendor-defined protocols, output structurally consistent raw signal sampling vectors in the main process, thus masking underlying communication differences. If protocol parsing fails, the corresponding measurement channel is marked as invalid at that moment, and the reason for the failure is written to the operation log or an extended raw data segment for retrospective review.

[0037] For the first Each measurement channel is based on the self-describing parameter set of the test equipment. Unit conversion factor given Unit conversion offset After correction, the measurement channel value in the physical quantity unit is obtained as follows: ; in, For the first Each measurement channel at discrete sampling time The corresponding physical quantity value; This is the original sampled value after parsing the measurement channel; This is used to convert the original sampled values ​​to the corresponding engineering units; It is used to compensate for zero bias or fixed bias.

[0038] The above formula forms the basis for subsequent physical quantity verification and normalization in this step. In this embodiment, the sensor sensitivity corresponding to the main vibration measurement channel is 100 mV / g, and the measurement range is 50 g. Therefore, the self-describing parameter set of the test equipment configures the unit conversion factor of the main vibration measurement channel as follows: Based on the aligned vibration sequence obtained in step S1, the sampled value of the second sampling point under normal operating conditions is... Substituting into the above formula, we get The sampled value of the second sampling point under the stuck condition is Substituting into the above formula, we get This demonstrates that after reading the original sampled values ​​from the actual production line, the intermediate adapter layer can convert the original voltage values ​​into acceleration physical quantities based on the sensor sensitivity parameters, thereby completing the conversion from the original sampled values ​​to unified engineering physical quantities.

[0039] To reduce the risk of false triggering caused by sensor configuration errors, drift, or communication resolution failures, the range of physical quantities after unit conversion is verified; based on the upper and lower limits of the physical quantity's range... , The specific steps for generating the range calibration identifier are as follows: ; in, For the first Each measurement channel at discrete sampling time Range calibration mark at the location, This indicates that the current physical quantity value of the channel is within a reasonable range. This refers to the upper limit of the range of a physical quantity. This is the lower limit of the range of a physical quantity.

[0040] Perform rate-of-change verification on key physical quantities such as temperature rise curves, pressure curves, and vibration amplitude sequences of the test equipment. Let the first... Maximum allowable rate of change per unit time for the measurement channel The change rate verification identifier element is: ; in, For the first Each measurement channel at discrete sampling time Change rate verification identifier at the location; The standard sampling period is set by step S1; The physical quantity value at the previous discrete sampling time; This is to measure the maximum rate of change of the channel per unit time under the current engineering scenario.

[0041] The above verification formula is used to suppress unreasonable jumps caused by communication jitter, short-term spikes, or transient parsing errors; the combined formula yields the... The validity identifier element for each measurement channel is: ; in, For the first At discrete sampling time The rate of change check flag at the location, when This indicates that the current sample in this channel can proceed to the subsequent standardization and unified state vector construction process. This indicates that the current sample in that channel has been determined to be invalid.

[0042] The measurement channel validity identifier vector, composed of the comprehensive validity identifiers of each measurement channel, is as follows: ; in, This is a vector indicating the validity of the measurement channel.

[0043] Step S22: After completing protocol parsing, unit conversion, and validity determination, perform dimensional unification and normalization standardization on the physical quantity values ​​of each measurement channel; set the truncation function. ,in, It is the independent variable of the function; To find the minimum value of the function; To find the maximum value of the function; then the first... The standardized results for each measurement channel are set as follows: ; in, For the first Each measurement channel at discrete sampling time The standardization results; This is a truncation function; To avoid dividing by zero for extremely small positive numbers.

[0044] The standardized vector of the measurement channel of the test equipment is composed of the standardized results of each channel. .

[0045] In this embodiment, based on the sensor's measurement range of 50g, the upper and lower limits of the vibration main measurement channel's range are configured in the self-describing parameter set of the test equipment as follows: Substituting the unit conversion results above, the channel-level standardized result for the second sampling point in the normal operating condition window is: The channel-level normalization result of the second sampling point in the stuck condition window is: The main purpose of this standardization process is to enable data from various devices, sensors, and raw dimensions to be incorporated into the subsequent unified state construction and edge-side fault identification processes in a consistent format.

[0046] When measuring channel validity identifier set In such cases, the standardized results are retained for backtracking, but in subsequent main variable construction or diagnostic calculations, the measurement channel validity identifier vector is used. Shielding. To achieve a unified data structure across testing equipment, a pre-defined general signal structure is provided, the number of its general measurement channel sets denoted as... Based on the self-describing parameter set of the test equipment. Measurement channel mapping rules Automatically generate measurement channel mapping matrix ,in Indicates the first The measurement channel of the test equipment is mapped to the first... One general measurement channel, otherwise Then the standardized vector of the measurement channel of the test equipment. Mapped to a general structure: ; in, A standardized data structure under a general signal structure; This is the measurement channel mapping matrix.

[0047] The core of this invention's portability lies here: through the self-describing parameter set of the experimental equipment. The given mapping rules The original measurement channels in each device, each with its own channel number, arrangement order, and physical meaning, are uniformly mapped to positions with consistent semantics under a general signal structure, thus providing a foundation for the subsequent construction of a unified state vector.

[0048] To achieve portability of this embodiment, this embodiment adopts a main measurement channel binding method to load the test equipment for the three common physical quantities. Temperature of test equipment Vibration of test equipment The corresponding main measurement channel index is determined by the mapping rules. Automatically provided, specifically: ; in, Index of the main measurement channels related to the load of the test equipment; Index of the main temperature measurement channel of the test equipment; This is the index for the measurement channel of the original vibration quantity of the test equipment, which is the acceleration or velocity measurement channel; For mapping rules.

[0049] The improvement in the above formula lies in the fact that this invention does not directly model all the original measurement channels separately, but first models them based on the self-describing text. Identify which channels represent load, temperature, and vibration respectively, and then bind these semantically equivalent but different physical quantities as the main variable inputs required for subsequent diagnosis.

[0050] The main temperature measurement channel of the test equipment, and the normalized temperature measurement of the test equipment are as follows: .

[0051] Test equipment load main measurement channel: First, obtain the physical quantity values ​​of the relevant measurement channels of the test equipment load. When the measurement channel is a pressure measurement channel, it is combined with the effective pressure-bearing area of ​​the actuator. The equivalent load of the test equipment is calculated as follows: ; in, Discrete sampling time The corresponding equivalent load of the test equipment; Standardize vector elements for the load measurement channel of the test equipment; The effective pressure-bearing area of ​​the actuator.

[0052] The improvement of the above formula lies in the fact that it is transformed into an equivalent load through engineering physical relationships and then normalized, so that the mechanical quantities from each source in each test device have consistent semantics in a unified state vector.

[0053] Normalization is performed using the equivalent load range corresponding to the pressure range, specifically as follows: ; ; in, The result is the normalized equivalent load of the test equipment; This is the upper limit of the equivalent load of the test equipment; This is the lower limit of the equivalent load of the test equipment; For the first Upper limit of the load measurement range of the test equipment in the measurement channel; For the first Lower limit of the load measurement range of the test equipment in the measurement channel.

[0054] When the main load measurement channel of the test equipment is a thrust or torque equivalent measurement channel, directly set the equivalent load of the test equipment. And the results of the equivalent load normalization of the test equipment Winning Normalization is complete.

[0055] Main measurement channel for vibration of the test equipment: Assume the original physical quantity of vibration of the test equipment is... To ensure that the RMS window length matches the sampling frequency, the vibration statistical time window length is set to be... The vibration intensity of the RMS test equipment is: ; ; in, Discrete sampling time The corresponding vibration intensity of the test equipment; For the current statistics window, the first Vibration physical quantity values ​​at historical sampling points; This serves as an index for the sampling points within the vibration statistics window; This represents the number of sampling points included within the vibration statistics window; The sampling frequency of the main vibration measurement channel; This represents the length of the vibration statistics time window.

[0056] To utilize existing range text data for normalization, the upper limit of the amplitude of the vibration measurement channel of the test equipment is set as follows: In this embodiment, vibration intensity calculations are performed on 12,000 vibration sequences for both the normal operating condition window and the stuck operating condition window. Using the unit-converted acceleration sequence as input, the vibration intensity calculation result corresponding to the normal operating condition window is: The vibration intensity calculation result corresponding to the stuck working condition window is as follows: .

[0057] ; in, This is the upper limit of the amplitude of the vibration measurement channel of the test equipment; For the first Lower limit of vibration measurement range for testing equipment in the measurement channel; For the first Upper limit of vibration measurement range for the test equipment in the measurement channel; This is a function that maximizes the value of a function.

[0058] In this embodiment, the upper limit of the amplitude of the vibration measurement channel of the test equipment is... The normalized result of the vibration of the test equipment is then taken as: ; in, This represents the normalized vibration result of the test equipment.

[0059] The improvement in the above formula lies in combining the sampling frequency and upper range limit in the self-describing text to uniformly map the vibration intensity into a normalized main variable that can directly participate in cross-device diagnostics. In this embodiment, the vibrator intensity is normalized. This demonstrates that the intermediate adapter layer of this invention can uniformly convert real production line vibration data into normalized vibration master variables based on the sensor range parameters configured in the self-describing parameter set of the test equipment, and output a unified input format compatible with the subsequent edge-side timing fault discrimination network.

[0060] This outputs the unified state vector of the test equipment. ,in, To unify the state vector of the test equipment; These represent the normalized test equipment load, normalized test equipment temperature, and normalized test equipment vibration, respectively. This formula is the final output of the unified state vector formula chain of the intermediate adaptation layer of this invention. In other words, this invention does not directly model the original channels of each device separately, but first completes unit conversion, range verification, and channel mapping through the test equipment self-describing parameter set and protocol parsing function, and then uniformly transforms the load information from various sources such as pressure, thrust, or torque into normalized load components, which, together with the temperature component and vibration component, constitute a unified state vector consistent across devices.

[0061] Step S23: To satisfy the consistency rule of full data collection being traceable and diagnosis using only a subset, output the availability identifiers of three common physical quantities, specifically: ; ; ; ; in, A mask for the validity of the main variables; This is an indicator function; it takes the value 1 if the condition is true, and 0 otherwise. The result of the validity check of the load master variable at the current moment; The result of the validity check of the temperature main variable at the current moment; This is the result of the validity check of the vibration main variable at the current moment.

[0062] It is necessary to determine the above formula. , , Does it belong to both the subset of fault diagnosis measurement channels and satisfy the validity check? This formula only allows the main variable to participate in the edge-side fault identification calculation if it is selected into the diagnostic link and the current sample is valid. Subsequent steps can only perform a validity mask on the main variable. The components with a value of 1 are calculated or weighted, and unselected measurement channels do not enter the diagnostic calculation link, thus maintaining executability and portability under the constraints of measurement channel differences and edge computing power.

[0063] Combining image acquisition calibration information with the key monitoring area generation rules in the self-describing parameter set of the experimental equipment Automatically generate configurations for key monitoring areas The critical monitoring area configuration is used to indicate multiple key monitoring locations subsequently cropped from the keyframe images, including valve assemblies, oil pipe joints, pressure gauges, and leakage risk points. When changes in ambient lighting, camera installation posture, or site layout cause calibration parameter updates, the intermediate adaptation layer automatically updates the critical monitoring area configuration.

[0064] Output of this step: Unified state vector of the test equipment Main variable applicability identifier vector Measurement channel validity identifier set Standardized data structures under general signal structures Configuration of key monitoring areas And the standardized sequence of the auxiliary measurement channels that are retained.

[0065] Step S3: Based on the unified state vector and measurement channel validity sequence output in step S2, perform timing fault discrimination within the sliding time window to obtain the failure probability of the test equipment. When the probability of equipment failure... After the initial screening trigger conditions are met, keyframes are matched according to timestamps and image features of key monitoring areas are extracted. The image difference metric relative to the baseline image is then calculated. The failure probability of the test equipment is determined by a unified fusion function. Mapped to the real number domain and compared with visual difference measures By performing unified-scale fusion, the fusion failure confidence of the test equipment is obtained. Once the fused fault confidence level meets the fault establishment conditions, a structured test equipment fault diagnosis record package is generated.

[0066] The aforementioned unified fusion function first maps the fault probability output by the edge-side temporal fault discrimination network to the real number domain, placing it in a fusion space that can be uniformly computed with the visual difference metric extracted from the key monitoring area image. Then, it generates a fused fault confidence score based on the weights of each branch and re-compresses it to a unified scale. This achieves unified judgment of various representations, namely "temporal sensing anomaly intensity" and "visual state deviation degree." Furthermore, when keyframe matching fails, key regions cannot be extracted, or visual branches are unavailable, the fusion expression can automatically degenerate into a judgment form that relies solely on the sensor-side fault probability, thus balancing multimodal verification capability with the robustness of independent edge-side discrimination.

[0067] Step S31: Use a sampling point length of... The sliding time window at discrete sampling times Constructing edge-side timing input sequences ;in, This is a sequence of uniform state vectors within the current sliding time window; This represents the number of sampling points included in the edge-side fault identification window. Step S2 outputs the measurement channel validity flag as follows: Then, synchronous construction: ,in, This is used to suppress or shield missing measurement channels, over-range measurement channels, or rate of change fault measurement channels during edge-side fault identification. This is a set of identifiers for measuring channel validity.

[0068] Let the edge-side timing fault discrimination network be denoted as Its network parameters are The edge-side temporal fault discrimination network is a one-dimensional convolutional temporal network, a gated recurrent temporal network, or an equivalent temporal discrimination implementation of both, used for fast initial fault screening of a unified state vector sequence under limited edge computing resources. The fault probability of the test equipment in the current window is then set as follows: ; in, This represents the probability of equipment failure during testing. This is a sequence of uniform state vectors within the current sliding time window; A valid sequence synchronized with the time window.

[0069] Set the edge-side initial screening fault trigger condition to ,in, The edge-side fault probability threshold is issued by the cloud and updated with the edge-side time-series fault identification network version. When the triggering condition is met, the initial edge-side fault triggering time is set to [time value missing]. If the cloud threshold has not yet been issued, or if the issued threshold does not match the current edge-side time-series fault identification network version, the local preset default threshold will be used for the judgment, and the threshold configuration and network version information will be written to the operation log or fault diagnosis record package.

[0070] In this embodiment, to verify the executability of the edge-side temporal fault discrimination network under single-channel conditions, the normal operating condition window and the stuck operating condition window are divided into training and testing segments according to time sequence. The first 70% of each class of 12,000-point sequences is used as the training segment, and the last 30% is used as the testing segment. Subsequently, within each data segment, the time window length is slid according to step S31. A temporal input sequence is constructed for the edge side, where the training sample label corresponding to normal operating conditions is denoted as 0, and the training sample label corresponding to jamming conditions is denoted as 1. The training samples are input into the edge-side temporal fault discrimination network. In this embodiment, a one-dimensional convolutional temporal network is used to train the network parameters. After training, the temporal input sequence obtained from the test segment is input into the edge-side temporal fault discrimination network, which outputs the fault probability corresponding to the test window. .in , Take the edge-side fault probability threshold. If the stuck working condition window is detected, it is determined to be a fault.

[0071] Step S32: When at time... When the edge-side initial screening fault is triggered, if visual-assisted diagnosis is enabled, the image cache index relationship of step S1 is invoked. The keyframes obtained by matching timestamps are: ; in, Keyframe images corresponding to the moment when the initial screening fault is triggered on the edge side; To the time of triggering the initial screening fault on the edge side The image frame number with the smallest absolute time difference; For Keyframe images with numbered tags; When the function reaches its minimum value The value; The keyframe acquisition time is the time closest to the fault trigger time; This is the time when the fault was triggered.

[0072] Configure the key monitoring areas based on the output of step S2. By cropping the keyframes, a set of images of the key monitoring area is obtained: ; in, This is a set of images of key monitoring areas corresponding to the time when the fault is triggered in the initial screening on the edge side, which consists of cropping results of multiple key monitoring areas. Trim operators for key monitoring areas; For the first Configuration of key monitoring areas; Number of key monitoring areas; Index the monitored area.

[0073] Let image feature vector extraction operator Then, a single image feature vector is obtained by extracting and fusing images from each key monitoring area: ; in, This is the image feature vector corresponding to the moment when the initial screening fault on the edge side is triggered. For image feature extraction and feature fusion operators.

[0074] The baseline characteristics are established during the stable operating phase when the test equipment begins to run. And calculate the image difference metric: ,in, This indicates the deviation parameter of the current key monitoring area image relative to the reference image; for Norm; its physical meaning is: when the visual state of key monitoring components such as hydraulic valve groups, oil pipe joints, pressure gauges, and leakage risk points differs significantly from normal operating conditions, It will increase.

[0075] To reflect the structure of prioritizing sensing and using images as an auxiliary tool, a vision enable indicator is set. If image processing is enabled and keyframes and key monitoring area image features are successfully extracted, then... ,otherwise Specifically, when keyframe timestamp matching fails, key monitoring area is configured as empty, key monitoring area cropping fails, or image feature extraction fails, all of these situations will cause... Furthermore, image-related fields are not written into subsequent fault diagnosis record packages; based on this, a unified fusion function is used to obtain the fused fault confidence score as follows: ; ; ; in, Discrete sampling time The corresponding fusion fault confidence level; This represents the probability of equipment failure during testing. For visual enabling indicators; The fusion weights are for the fault probability branch on the sensing side; The fusion weights for the visual difference branches; For bias terms; For measuring image differences in key monitoring areas; This is a function that maps fault probability values ​​to the real number domain, used to map probability values ​​in the interval (0, 1) to the real number domain, enabling it to be fused with image difference measures in a unified linear space. A fault confidence level with a uniform scale is used to re-compress the fusion score into the interval (0, 1); It is the natural logarithm function; For input feature variables.

[0076] The fault condition is set as follows: ,in, To determine the threshold, updates are sent from the cloud. When the vision enable indicator... When the above formula automatically degenerates into a fault confidence expression determined solely by the fault probability of the sensing side, it allows fault determination to be completed independently by the sensing side branch even when visual-assisted verification is not enabled or the image is unavailable.

[0077] Step S33: Once the fused fault confidence score meets the fault establishment condition, a structured test equipment fault diagnosis record package is generated. To avoid continuous faults causing a single record package to grow indefinitely, only the minimum segmentation rule is retained. The fault start time is defined as the first time the condition is met. timestamp Let the point of recovery be the point where the continuity is first satisfied. Each discrete time point is below the judgment threshold. Discrete time index Let the upper limit of the length terminate at point . ,in, The time of fault initiation The corresponding discrete index; This represents the maximum discrete window length that a single fault diagnosis record package is allowed to cover. Therefore, the terminating discrete index of the current fault diagnosis record package is set to... ,in, This is the terminating discrete index of the current fault diagnosis record package.

[0078] The fault diagnosis record package must include at least the following fields: Test equipment identification. Fault start time Fault termination time Fault trigger time Record the unified state vector sequence within the window. The corresponding measurement channel validity identifier sequence Edge-side failure probability Fusion Fault Confidence And the keyframe images corresponding to when visual-assisted review is enabled. Image set of key monitoring areas Image feature vectors Image difference measurement If visual-assisted verification is not enabled, image-related fields will be empty or not written to the record package. If the length limit is reached... And the termination and the termination time still satisfy the condition. If so, the fault is determined to be in a continuous state, and is... The next fault diagnosis record package will continue to be segmented and encapsulated, serving as the starting index.

[0079] In this embodiment, since the fusion fault confidence level of the normal operating condition window is 0.0392, which does not reach the fault establishment threshold, no fault diagnosis record package is generated. However, the fusion fault confidence level of the stuck operating condition window is 0.9124, which reaches the fault establishment threshold, so a corresponding fault diagnosis record package is generated. The fault diagnosis record package includes at least: test equipment identifier, start and end times of the discrimination window, vibration main measurement channel identifier, standardized vibration sequence within the window, edge-side fault probability, and current edge-side parameter version number. When image-assisted verification is not enabled, image-related fields are empty or not written to the record package. The output of this step is: edge-side fault probability. Fusion Fault Confidence And a structured test equipment fault diagnosis record package.

[0080] Step S4: Analyze the fault types of the test equipment based on the unified state vector, fault confidence and image information; incrementally update the mirror copy of the edge-side temporal fault discrimination network according to the confirmed valid samples, generate differential parameters and send them to the edge side, thereby forming a parameter optimization closed loop of edge-cloud collaboration; trigger rich media alarm communication according to the optimization results and execute the test equipment fault identification alarm program.

[0081] Step S41: The cloud receives the fault diagnosis record package uploaded from the edge side and extracts the unified state vector sequence of the test equipment within the record window. Fusion Fault Confidence And the corresponding keyframe images when visual-assisted review is enabled. Or key monitoring area image features. The current test equipment data in the fault diagnosis record package constitutes the target domain data in the migration diagnosis task, while the historical mature test equipment fault samples pre-stored in the cloud database constitute the source domain data. In this embodiment, the cloud diagnosis network is explicitly defined as a cloud multimodal migration diagnosis network. It is used to verify the fault type of the current fault diagnosis record package.

[0082] Cloud-based multimodal transfer diagnostics network It includes a multimodal feature extraction unit, a domain adaptation unit, and a fault classification unit, and its training objective can be abbreviated as: ; in, The total loss function of the cloud-based multimodal migration diagnostic network; Diagnostic network parameters for multimodal migration in the cloud; This is source domain fault sample data; Record data for fault diagnosis in the target domain; The source domain classification loss is used to enable the network to identify common failure modes; This is the domain adaptation loss, used to reduce the distribution difference between the source domain and the target domain in the feature space; This is the balance coefficient.

[0083] After inputting the current fault diagnosis record package into the cloud-based multimodal migration diagnostic network, the network outputs the fault probability vector for each fault category. ,in, This is the probability vector of the fault category corresponding to the current fault diagnosis record; The total number of preset fault categories, The current fault diagnosis record belongs to the first... The probability of a type of failure.

[0084] Based on the failure probability vector Select the index with the highest probability. Determine the initial fault type. This embodiment of the invention pre-configures a graded fault mode table for hydraulic test benches, covering component-level, hydraulic circuit-level, and system-level faults; when ported to non-hydraulic test equipment, only the fault mode table needs to be replaced. Based on the determined fault type... Refer to the table below to obtain the corresponding standard fault names and basic severity assessment results. Take values ​​from 0 to 100.

[0085] Table 1. Typical Failure Modes and Severity Mapping Table for Hydraulic Testing Rigs Those skilled in the art will understand that when transferred to non-hydraulic testing equipment, such as a motor test bench, Table 1 can be configured accordingly to fault modes such as electrical short circuit and bearing wear.

[0086] Step S42: To avoid false alarms or false negatives due to relying solely on the basic severity assessment results, the cloud platform further constructs physical quantity exceedance parameters based on the deviation of the unified state vector of the test equipment, and sets them as follows: ; in, The physical quantity parameter that exceeds the limit corresponding to the current fault diagnosis record; The weighting coefficient for the load deviation term; This represents the weighting coefficient for the temperature deviation term; Let be the weighting coefficient of the vibration deviation term, and satisfy . ; To normalize the load on the test equipment; To normalize the temperature of the test equipment; To normalize the vibration of the test equipment; This is the baseline value for the normal operating load of the test equipment; This is the baseline value for the normal operating temperature of the test equipment; This is the reference value for the normal operating conditions of the test equipment vibration.

[0087] Normal operating conditions are defined as: within a continuous preset time window, the fault confidence level is consistently lower than the fault establishment threshold, and the test equipment is in a stable operating state; during this period, the time average of each component of the unified state vector is taken.

[0088] Since the severity of the fault mechanism itself is usually higher than that of a single measurement deviation, the weighting relationship is set as follows: And determine the parameters for constructing the fault assessment. for: ; in, For fault assessment parameters; Fault type; Based on the severity assessment results; The fault confidence is obtained in step S3; Prior severity weights for fault types; For fault confidence weights; For physical over-limit weights; The physical quantity parameter that exceeds the limit corresponding to the current fault diagnosis record; This is a nonlinear saturation mapping function, which aims to map physical quantities with potentially large ranges of values ​​beyond their limits. Compressing the parameters to a limited range (0, 1) prevents the sudden and drastic changes in individual physical quantities from having an excessively large proportion in the fault assessment parameters and thus obscuring other information.

[0089] The improvement to the above formula lies in incorporating the severity of the fault mechanism, the current fault establishment parameters, and the deviation parameters of the unified state vector relative to normal operating conditions into the risk assessment. This ensures that cloud-based decision-making considers not only the fault category itself but also the deviation parameters of the current operating conditions and the fault judgment results from the edge side. Based on the fault assessment parameters... Classify the faults.

[0090] Set the first-level fault threshold and Level 2 fault threshold and satisfy The fault risk level is... It can be set as: Level 1 fault, when At that time; Level 2 fault, when When; Level 3 fault hour.

[0091] Step S43: To ensure the continuous evolution of the edge-side temporal fault discrimination network, a mirror copy with the same structure as the edge side is maintained in the cloud, and periodic incremental training is performed using confirmed valid samples to obtain updated edge-side network parameters. Cloud-based calculation of differential parameters: Only the difference parameter The message is sent out. Upon receiving it, the edge node performs an online update operation: The cloud distributes the differential parameters along with the parameter version number. When the edge detects a version mismatch, it continues to use the current local parameters to perform initial fault screening and writes the version inconsistency information to the operation log for subsequent backtracking and resynchronization.

[0092] This step outputs: Fault Type Index Fault assessment parameters Fault risk level And the differential parameters sent to the edge side With corresponding version information.

[0093] Step S44: See Appendix Figure 3 The flowchart illustrates the workflow for hierarchical alarm, multi-measurement channel push, and receipt reception in the rich media alarm communication provided by this invention; in this embodiment, a fault type index is output according to steps S41 and S42. Fault assessment parameters Fault risk level Subsequently, rich media alarm communication is triggered to adaptively select and accurately deliver alarm elements. The goal of this step is to dynamically determine the combination of alarm elements based on the fault risk level and link bandwidth constraints under network conditions, and to improve the reliability of alarm information delivery in unattended scenarios through receipt detection and a multi-communication channel redundant push mechanism.

[0094] According to the fault risk level Pre-set alarm element priority rules: When a fault is determined to be Level 1, send a text description, keyframe image, and voice prompt to provide evidence of the fault scene; when a fault is determined to be Level 2, send a text description, keyframe compressed thumbnail, and trend curve summary; when a fault is determined to be Level 3, only send the necessary text alarm information.

[0095] Real-time monitoring of uplink bandwidth used from the current edge node to the remote monitoring terminal , This is estimated based on the effective throughput or network interface statistics of alarm messages sent in the recent period. The maximum allowable transmission delay for alarm messages is set to... Then the maximum data payload allowed to be sent at the current moment is ; in, The maximum data payload allowed for transmission represents the maximum data payload available for sending alarm messages at the current moment while satisfying the maximum allowed transmission delay constraint. This represents the available uplink bandwidth from the current edge node to the remote monitoring terminal. This is the network fluctuation coefficient, used to reserve margin for link jitter, transient congestion, and protocol overhead.

[0096] The physical meaning of the above formula is that it unifies the available bandwidth, latency constraints, and security margin of the link into the upper limit of alarm packets allowed to be sent at the current moment. To maximize the value of alarm information under limited bandwidth, an adaptive selection model for alarm elements is constructed.

[0097] Suppose there are a total of candidate alarm elements corresponding to the current fault risk level. The candidate element, the first The size of each element is denoted as . Information utility value is denoted as Introducing binary selection variables ,in Indicates selection to send the first One alarm element, This indicates that the element will not be sent. To ensure that the semantics of the fault alarm are not lost, a mandatory text selection constraint is further added, meaning that the selection variable corresponding to the text alarm element is always 1; therefore, the fault alarm element group optimization is as follows: ; in, For the first The information utility value of each candidate alarm element is used to characterize the important parameters of the element for remote maintenance personnel to judge faults. The data size of this candidate alarm element; For the first Whether each candidate alarm element is selected into the current alarm package; This represents the total number of candidate alarm elements. By solving the above 0 / 1 constraint optimization problem, the final alarm package content combination is obtained. .

[0098] The innovation of the above formula lies in the fact that, during the alarm phase, this invention does not send fixed preset content, but rather dynamically selects efficient alarm elements based on the fault risk level and real-time bandwidth conditions. When bandwidth is sufficient, the alarm packet... It can contain various information such as text descriptions, keyframe images, trend curve summaries, and voice prompts; when bandwidth is limited, the optimization result will be automatically compressed into smaller payload elements such as text descriptions, compressed keyframe thumbnails, or key curve summaries; when bandwidth further decreases, only necessary text information is retained. This is based on the maximum data payload currently allowed to be sent. If the payload is less than the minimum required for a text alarm, and even sending a standard text alarm would fail to meet the latency constraints, then a degraded encoding mode is entered. In degraded encoding mode, long natural language text is no longer sent; instead, the fault information is compressed and encoded into a short status code before being sent. The short status code includes at least the core fields such as the test equipment identifier, fault risk level, fault type index, and trigger time.

[0099] This embodiment supports multiple communication measurement channels, including industrial Ethernet, 4G / 5G wireless networks, and SMS gateways. High-bandwidth measurement channels are prioritized for push operations by default. To the remote monitoring terminal. To ensure accurate delivery of critical alarms, a receipt waiting timer is started after the alarm is sent, with a waiting threshold set. If in If no automatic feedback response signal is received from the remote monitoring terminal within the specified time, the redundancy mechanism is automatically triggered: First, it immediately switches to a low-bandwidth backup measurement channel, such as an SMS gateway or narrowband IoT measurement channel. Second, it forcibly degrades the alarm elements into core text information or short status codes and retransmits them. Once a terminal feedback signal is received, retransmission stops, and the closed-loop status of this communication is marked as delivered. If no acknowledgment is received within the preset number of retries, the alarm is marked as awaiting manual review or unconfirmed delivery, and the communication log is written to the operation record for subsequent tracking. Output: Alarm packet content combination. Send communication channel selection results and degradation strategy execution results, as well as receipt reception results and final delivery status flags.

[0100] Data generated throughout the entire monitoring process is archived in a unified manner, and an intuitive status retrospective interface is provided to operations and maintenance personnel. Simultaneously, confirmed valid historical samples are fed back to the cloud for diagnostics and collaborative learning, used for subsequent parameter updates, thus forming a data closed loop.

[0101] The unified state vector sequence of the test equipment output in step S2 The system continuously stores timestamps to record the changing trends of normalized test equipment load, temperature, and vibration. Simultaneously, the standardized sequences of auxiliary measurement channels retained in step S2 are archived for status retrospection, offline verification, and auxiliary display. It should be noted that the archived non-diagnostic measurement channel data is only used for historical querying and auxiliary analysis and does not participate in the online fault identification and risk assessment calculations of steps S3 to S4. Furthermore, the fault diagnosis record package generated in step S3 is archived, retaining the keyframe image, fault confidence level, and original waveform fragment corresponding to the fault trigger time. The diagnostic conclusions output in step S4 are also recorded synchronously, including at least the fault type index, basic severity assessment results, fault assessment parameters, and parameter version information. In addition, based on the acknowledgment results returned by the remote monitoring terminal in step S44, the corresponding alarm records are marked with a delivery status, and combined with the computer verification results, the relevant historical records are marked as valid or invalid samples to complete data cleaning.

[0102] Based on archived data, status backtracking services are provided to operations and maintenance personnel. To avoid making the manual too lengthy, the interface display format and types of statistical charts will not be elaborated further here. However, it should be understood that any trend display, fault reproduction, and historical query that can be achieved based on archived data falls under the status backtracking implementation methods of this step. Historical data marked as valid samples are periodically organized and fed back to cloud-based diagnostics and collaborative learning to drive subsequent updates of cloud-based multimodal migration diagnostic network parameters and edge-side temporal fault discrimination network mirror copy parameters. Thus, not only is the archiving and backtracking of monitoring data completed, but it also serves as a data entry point for continuous optimization of edge-cloud collaboration, gradually accumulating samples and continuously optimizing parameters as runtime increases.

[0103] like Figure 4 The image shows a comparison of waveforms acquired from 480,000 points in the vibration channel of a production line. It illustrates the original vibration signal characteristics under normal operating conditions, allowing for a direct observation of the overall differences in vibration amplitude characteristics under normal operating conditions. Figure 5 The image shows a comparison of waveforms collected from 480,000 points in the vibration channel of a production line. It illustrates the vibration signal morphology under jamming conditions and allows for a direct observation of the overall difference in vibration amplitude characteristics between jamming and normal conditions, providing an intuitive background for the data source of the example.

[0104] like Figure 6 The image shows a complete discrimination window under normal operating conditions, with vibration time-domain waveforms from 12,000 consecutive sampling points. The waveform morphology reveals a certain regularity in the vibration amplitude distribution under normal operating conditions, but this is difficult to accurately distinguish with the naked eye and cannot be determined solely through manual observation. Figure 7The image shows a complete discrimination window for the stuck condition, with vibration time-domain waveforms from 12,000 consecutive sampling points. The waveforms reveal differences in vibration amplitude distribution between the two conditions. The overall amplitude of the stuck condition is slightly different, but these differences are difficult to discern accurately with the naked eye, highlighting the limitations of relying on manual observation and emphasizing the necessity of introducing an edge-side timing fault discrimination network for automatic initial screening.

[0105] like Figure 8 The image shows the discrimination output of the edge-side temporal fault discrimination network for two test segments under different operating conditions: the fault probability for the normal operating condition is 0.0392, and the fault probability for the stuck operating condition is 0.9124, with a discrimination threshold set to 0.50. This demonstrates that, using only single-channel vibration data, the edge-side temporal fault discrimination network can effectively distinguish between normal and stuck operating conditions.

[0106] The beneficial effects of the embodiments of the present invention are as follows: The embodiments of the present invention achieve automatic identification and parameter loading of test equipment of various models and communication protocols through a test equipment fault identification method based on diagnostic semantic adaptation, enabling the same monitoring and diagnostic software to be quickly ported and deployed among multiple types of test equipment, thereby improving the portability and versatility of the present invention; the cloud-based multimodal migration diagnostic network is used for fault verification and incremental learning, forming a closed loop of edge-cloud collaborative parameter optimization, which improves the accuracy and robustness of fault identification while taking into account real-time performance; the embodiments use rich media alarm communication to adaptively select alarm elements such as text, image summaries, curve information and voice prompts, which can improve the efficiency of fault alarm.

[0107] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for fault identification of test equipment based on diagnostic semantic adaptation, characterized in that, It includes: S1: Use experimental equipment to collect multi-measurement channel sensor signals and image information output by the image acquisition unit during the test, and process the multi-source asynchronous data with a unified time reference; Multi-channel sensor signals include test equipment load Sensor signals, temperature of test equipment Sensor signals and vibration of test equipment Sensing signals; S2: Obtain the multi-source asynchronous data processed in step S1, based on the self-describing parameter set of the experimental equipment. The process involves protocol parsing, unit conversion, range verification, rate of change verification, and semantic mapping of measurement channels for the raw sampled values ​​of each communication protocol, channel number, and dimension; and will be integrated with the load of the test equipment. Temperature of test equipment Vibration of test equipment The relevant measurement channels are bound to the main measurement channel and are then normalized and standardized. A standardized data structure that maps uniformly to a general signal structure and has consistent semantic positions. Construct a unified state vector for the experimental equipment Output the main variable applicability identifier vector. and measurement channel validity identifier set ; S3: Based on the unified state vector and measurement channel validity sequence output in step S2, perform timing fault discrimination within the sliding time window to obtain the failure probability of the test equipment. When the probability of equipment failure... After the initial screening trigger conditions are met, keyframes are matched according to timestamps and image features of key monitoring areas are extracted. The image difference metric relative to the baseline image is then calculated. The failure probability of the test equipment is determined by a unified fusion function. Mapped to the real number domain and compared with visual difference measures By performing unified-scale fusion, the fusion failure confidence of the test equipment is obtained. ; Once the fused fault confidence level meets the fault establishment criteria, a structured test equipment fault diagnosis record package is generated. S4: Perform fault type analysis on the obtained unified state vector, fault confidence, and image information to obtain the fault type index. Fault assessment parameters and fault risk level ;Execute the test equipment fault identification and alarm procedure.

2. The test equipment fault identification method based on diagnostic semantic adaptation according to claim 1, characterized in that: Step S2 is as follows: S21: Obtain the self-describing parameter set of the test equipment Through the protocol parsing function The raw data in step S1 is parsed into a vector of raw signal sample values. Physical quantity verification is performed to obtain the validity identifier of the measurement channel; S22: Define the cutoff function Obtain the standardized vector of the measurement channel of the test equipment. Perform dimensional unification mapping and normalization standardization on the physical quantities of each measurement channel; S23: Output the main variable applicability identifier vector of the test equipment according to the consistency rule. Includes: Validation results of load host variables Validation results of temperature main variable Results of validation of vibration main variables .

3. The test equipment fault identification method based on diagnostic semantic adaptation according to claim 2, characterized in that: The dimensionless mapping in step S22 is specifically as follows: ； ； ； in, A standardized data structure under a general signal structure; For measuring the channel mapping matrix; Standardized vectors for the measurement channels of the test equipment; Index for discrete sampling time; Index of the main measurement channels related to the load of the test equipment; Index of the main temperature measurement channel of the test equipment; Index of the measurement channels for the original vibration quantities of the test equipment; For mapping rules; For the self-describing parameter set of the test equipment; Discrete sampling time The corresponding equivalent load of the test equipment; Standardize vector elements for the load measurement channel of the test equipment; The effective pressure-bearing area of ​​the actuator.

4. The test equipment fault identification method based on diagnostic semantic adaptation according to claim 1, characterized in that: Step S3 is as follows: S31: Obtain the validity identifier of the measurement channel. Then, the effective sequence of synchronous construction and time window synchronization is achieved. Use an edge-side timing fault discrimination network Obtain the failure probability of the test equipment Perform edge-side initial screening to determine fault triggering; S32: When the initial screening fault on the edge side is triggered, key frames are obtained by matching the timestamps, resulting in a set of images of the key monitoring area; operators are extracted based on the image feature vectors. Extract and fuse images of key monitoring areas to obtain a single image feature vector. The fusion fault confidence level is obtained using a unified fusion function. Perform fault condition determination; S33: Once it is determined that the fusion fault confidence level meets the fault establishment conditions, a structured test equipment fault diagnosis record package is generated.

5. The test equipment fault identification method based on diagnostic semantic adaptation according to claim 4, characterized in that: In step S32, the unified fusion function determines the fusion failure confidence level of the test equipment as follows: ； ； ； in, Discrete sampling time The corresponding fusion fault confidence level; To standardize the failure confidence level; The fusion weights are for the fault probability branch on the sensing side; A function that maps fault probability values ​​to the real number field. This represents the probability of equipment failure during testing. For visual enabling indicators; The fusion weights for the visual difference branches; For measuring image differences in key monitoring areas; For bias terms; It is the natural logarithm function; For input feature variables.

6. The test equipment fault identification method based on diagnostic semantic adaptation according to claim 1, characterized in that: Step S4 is as follows: S41: Upload the fault diagnosis record package to the cloud-based multimodal migration diagnostic network. Output the failure probability vector for each failure category. Select the index with the highest probability. Determine the initial fault type; S42: Construct out-of-limit parameters for physical quantities based on the deviation of the uniform state vector of the test equipment. Determine the parameters for constructing a fault assessment. Analyze the fault risk level Division; S43: Establish a mirror copy of the edge network structure that is consistent with the cloud-based structure, and perform periodic incremental training using confirmed valid samples to obtain updated edge network parameters. ; S44: Triggers rich media alarm communication and executes the test equipment fault identification alarm program.

7. The test equipment fault identification method based on diagnostic semantic adaptation according to claim 6, characterized in that: Fault assessment parameters in step S42 for: ； in, For fault assessment parameters; Prior severity weights for fault types; Fault type; For fault confidence weights; To integrate fault confidence; For physical over-limit weights; The physical quantity parameter that exceeds the limit corresponding to the current fault diagnosis record; It is a nonlinear saturation mapping function.

8. The test equipment fault identification method based on diagnostic semantic adaptation according to claim 6, characterized in that: Step S44 is as follows: Index by Fault Type Fault assessment parameters and fault risk level The rich media alarm communication of the triggering test equipment enables adaptive selection and delivery of alarm elements; based on the fault risk level... Pre-set alarm element priority rules: fault risk level For a Level 1 fault, send a text description, keyframe images, and voice prompts to provide evidence of the fault location; fault risk level. For a level 2 fault, send a text description, a compressed thumbnail of the keyframe, and a summary of the trend curve; fault risk level. In the case of a level 3 fault, only necessary text alarm messages are sent; Determine the maximum data payload allowed to be sent at the current moment. The candidate alarm elements corresponding to the current fault risk level are identified as having a total of Optimize the fault alarm element group using candidate elements.

9. The test equipment fault identification method based on diagnostic semantic adaptation according to claim 8, characterized in that: The optimization of the fault alarm element group in step S44 is as follows: ； in, To find the maximum value of the function; For the first The information utility value of each candidate alarm element; The data size of this candidate alarm element; For the first Whether each candidate alarm element is selected into the current alarm package; This represents the total number of current candidate alarm elements; Use constraint symbols for modeling; Index of candidate alarm elements; This represents the maximum data payload allowed for transmission.

Images