A data-driven-based industrial equipment fault diagnosis method, device and medium

Abstract

This invention discloses a data-driven industrial equipment fault diagnosis method, device, and medium, relating to the field of equipment fault diagnosis technology. The method includes: establishing a load energy flow diagnostic chain based on net deviation data of action phases; isolating anomalies along the chain to verify the preceding and following stages; determining whether the energy transfer deviation is caused by a single action disturbance; identifying the source of the continuously propagating anomaly and extracting fault evolution characteristics; generating candidate fault diagnosis information; constructing a fault propagation delay evidence chain using the candidate fault diagnosis information and verifying the anomaly propagation sequence; generating delay verification fault information; further verifying the delay verification fault information under similar operating conditions and correcting the diagnostic credibility to obtain industrial equipment fault diagnosis information; generating handling information based on the industrial equipment fault diagnosis information and binding maintenance confirmation content to form a fault diagnosis verification sample record. This invention enables the verification of whether an anomaly continues to propagate along the energy transfer path.

Description

Technical Field

[0001] This invention relates to the field of equipment fault diagnosis technology, and in particular to a data-driven industrial equipment fault diagnosis method, equipment, and medium. Background Technology

[0002] As industrial equipment such as cranes evolves towards automation, continuous operation, and multi-condition collaboration, the operational safety of industrial equipment increasingly relies on real-time status monitoring and data-driven fault diagnosis technologies. Existing technologies collect various operational data through sensors, including motor current, vibration frequency, temperature; reducer vibration frequency, bearing temperature; wheel vibration frequency, bearing temperature; mechanism speed; load status; braking feedback; structural vibration; AI algorithms; and control commands. These data are then combined with anomaly detection, machine learning classification, and remote maintenance platforms to identify faults such as motor malfunctions, reducer malfunctions, wheel malfunctions, wire rope malfunctions, braking malfunctions, transmission malfunctions, structural malfunctions, control system malfunctions, and load malfunctions. Compared to manual inspections and periodic maintenance, data-driven diagnostics enables continuous analysis of the industrial equipment's operation and has become a crucial technological direction for the intelligent operation and maintenance of industrial equipment such as cranes.

[0003] As the operating conditions of industrial equipment in port loading and unloading, engineering construction, and intelligent manufacturing scenarios become increasingly complex, action switching, load inertia continuation, and compound action coupling will cause real-time operating data to exhibit obvious staged and transitive characteristics, which puts forward higher requirements for data synchronization, operating condition identification, and anomaly interpretation in the fault diagnosis process. At the same time, in practical applications, industrial equipment fault diagnosis information not only needs to identify the fault category, but also needs to be verified by combining the anomaly propagation direction, phase delay sequence, and maintenance confirmation content, so as to improve the credibility of diagnostic information, the pertinence of handling, and the utilization value of sample updates. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a data-driven industrial equipment fault diagnosis method to solve the problem that it is difficult to interpret the real-time operating data of industrial equipment under complex operating conditions, verify the fault propagation, and confirm the correlation of maintenance.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a data-driven industrial equipment fault diagnosis method, comprising: collecting real-time operating data of industrial equipment and performing synchronous calibration and operating condition binding based on a unified time reference to obtain a standard action condition data sequence; dividing the standard action condition data sequence into action conditions, establishing a load-action phase continuity reference based on the load continuity relationship between adjacent actions, and performing phase stripping on non-fault disturbances caused by action switching to form action phase net deviation data; establishing a load energy flow diagnosis chain based on the action phase net deviation data, performing anomaly front-to-back level counter-evidence isolation along the load energy flow diagnosis chain, determining whether the energy transfer deviation is caused by a single action disturbance, locking the continuously transmitted anomaly source and extracting fault evolution characteristics to generate candidate fault diagnosis information; constructing a fault propagation delay evidence chain using the candidate fault diagnosis information and verifying the anomaly propagation sequence to generate delay verification fault information, then performing similar operating condition verification on the delay verification fault information and correcting the diagnosis credibility to obtain industrial equipment fault diagnosis information; generating handling information based on the industrial equipment fault diagnosis information and binding maintenance confirmation content to form a fault diagnosis verification sample record.

[0007] As a preferred embodiment of the data-driven industrial equipment fault diagnosis method of the present invention, the specific steps for obtaining the standard operating condition data sequence are as follows: Collect real-time operating data of industrial equipment, and write the local collection time, data source identifier and operating status identifier into the real-time operating data of industrial equipment. Then, map the local collection time to the unified time reference corresponding to the main controller of industrial equipment, and perform time correction in combination with action event anchor points to form a corrected operating frame. The corrected running frames are synchronously resampled, data validity verified, and anomaly sampling reliably corrected according to a unified diagnostic cycle. The operating conditions are then bound together with the action change status and load change status, and continuously encapsulated in a unified time sequence to obtain a standard action operating condition data sequence.

[0008] As a preferred embodiment of the data-driven industrial equipment fault diagnosis method of the present invention, the specific steps for generating the net deviation data of the action phase are as follows: The motion condition binding identifiers in the standard motion condition data sequence are compared frame by frame in chronological order. When the motion condition binding identifier changes, the moment of change is taken as the initial switching point. Then, the process is traced back to the diagnostic moment when the control command data begins to change and extended backward to the diagnostic moment when the mechanism feedback data enters the new motion state. When the load change maintains a continuous bearing state, the traced start point to the extended end point is determined as the motion switching transition zone, and the continuous data between two adjacent motion switching transition zones is divided into a set of motion candidate segments. The load state at the end of the previous action candidate segment and the load state at the beginning of the next action candidate segment in the action candidate segment set are continuously compared. When the change of the load state at the beginning of the next action candidate segment relative to the load state at the end of the previous action candidate segment is within the load continuation judgment range, and there is no unloading breakpoint between the end of the previous action candidate segment and the beginning of the next action candidate segment, it is determined that there is a load continuation relationship between the previous action candidate segment and the next action candidate segment. Adjacent action candidate segments with load continuation relationship are connected in chronological order to form a load-action phase continuity diagram. Based on the load-action phase continuity diagram, historical normal continuity segments are retrieved, and the current action candidate segment is phase-aligned with the historical normal continuity segments to form a load-action phase continuity benchmark. Based on the load-action phase continuity benchmark, non-fault disturbances caused by load continuation during action switching are identified. The non-fault disturbances are extracted from the deviation data of the current action candidate segment, and the action phase net deviation features are extracted. Then, they are encapsulated according to the action occurrence order to form action phase net deviation data.

[0009] As a preferred embodiment of the data-driven industrial equipment fault diagnosis method of the present invention, the specific steps of establishing a load energy flow diagnostic chain based on the net deviation data of the action phase are as follows: Based on the net phase deviation data of the action, phase normalization is performed from the start time to the end time of the action. The net deviation segment is mapped to a set of phase energy port frames. The phase-gated energy transfer confidence between the front-end and back-end ports is calculated according to whether the energy change of the front-end port causes the response change of the back-end port within the allowable phase delay. When the phase-gated energy transfer confidence reaches the energy transfer chaining condition and the response change of the back-end port is later than the energy change of the front-end port, it is determined that there is an effective energy transfer relationship between the front-end and back-end ports, and the corresponding ports are connected to form a candidate load energy flow graph. Based on the candidate load energy flow graph, check step by step along the port connection direction whether the energy change of the preceding port is earlier than the response change of the following port, and check whether the following port responds within the allowable phase delay after the energy of the preceding port increases. If the port connection only shows synchronous fluctuation, does not conform to the forward and backward transmission order, and the response delay exceeds the phase delay interval, then delete the corresponding port connection to form a stable load energy flow graph. Then, select the path from the stable load energy flow graph that continuously transmits from the action phase source end with the earliest net deviation to the final abnormal response end, and write the energy transmission confidence and net deviation propagation status in the path into the path node to form a load energy flow diagnostic chain.

[0010] As a preferred embodiment of the data-driven industrial equipment fault diagnosis method of the present invention, the specific steps for generating candidate fault diagnosis information are as follows: The path nodes in the load energy flow diagnostic chain are divided into the pre-stage observation segment, the current isolation segment, and the post-stage verification segment. The net deviation trajectory of the current isolation segment is replaced with the normal baseline trajectory under the corresponding action phase. At the same time, the real energy change of the pre-stage observation segment is retained to generate the counterfactual post-stage response trajectory. The counterfactual response trajectory is compared with the actual response trajectory of the subsequent verification segment to determine whether the subsequent anomaly has decayed after the current isolation segment is replaced. If the subsequent anomaly has not decayed, the path node corresponding to the current isolation segment is identified as a single action disturbance node, and the subsequent path node is verified along the load energy flow diagnostic chain. If the subsequent anomaly continues to decay, the path node corresponding to the current isolation segment is locked as the source of the continuously transmitted anomaly. The fault evolution features corresponding to the net deviation enhancement process, phase delay change and energy transfer change are extracted from the continuously transmitted anomaly source, and candidate fault diagnosis information is generated based on the fault evolution features.

[0011] As a preferred embodiment of the data-driven industrial equipment fault diagnosis method of the present invention, the specific steps for generating delay verification fault information are as follows: The anomaly source location, anomaly propagation path and candidate fault type in the candidate fault diagnosis information are converted into fault propagation evidence tokens. A time delay pairing relationship is established between adjacent fault propagation evidence tokens according to the propagation direction in the load energy flow diagnosis chain. Then, the phase delay interval is called according to the abnormal arrival time and action phase difference of adjacent fault propagation evidence tokens. The abnormal arrival order of adjacent fault propagation evidence tokens is verified according to the phase delay interval. The time delay pairing relationships that satisfy the sequential propagation relationship are connected to form a fault propagation time delay evidence chain. The abnormal arrival order and phase delay consistency of adjacent fault propagation evidence tokens are checked step by step along the fault propagation delay evidence chain. If the abnormal arrival time of the later fault propagation evidence token is earlier than that of the earlier fault propagation evidence token, and the arrival interval between the later and earlier fault propagation evidence tokens does not fall within the phase delay interval, the corresponding candidate fault diagnosis information is marked as delay inconsistency information and removed from the current fault propagation delay evidence chain. The retained candidate fault diagnosis information is bound to the fault propagation delay evidence chain, and then the candidate fault location, candidate fault type, abnormal arrival order and delay verification status are written to generate delay verification fault information.

[0012] As a preferred embodiment of the data-driven industrial equipment fault diagnosis method of the present invention, the specific steps for obtaining industrial equipment fault diagnosis information are as follows: Based on the time delay verification fault information, generate the current fault review signature, and retrieve the historical review segment that matches the current action succession process from the historical operation record according to the current fault review signature; Based on the maintenance confirmation status of historical review segments, distinguish between historical normal continuation segments and maintenance confirmation fault segments. Calculate the normal counter-evidence support degree of historical normal continuation segments for the current abnormal propagation, and calculate the fault case support degree of maintenance confirmation fault segments for the current abnormal propagation. Then, generate the basic reliability of the delay based on the delay verification status in the delay verification fault information. The reliability of the delay basis is weakened based on the support of normal counter-evidence, and the reliability of the delay basis is strengthened based on the support of fault cases to obtain the corrected diagnostic reliability. Information that the corrected diagnostic reliability meets the diagnostic confirmation conditions is written into the industrial equipment fault diagnosis information.

[0013] As a preferred embodiment of the data-driven industrial equipment fault diagnosis method of the present invention, the specific steps for forming the fault diagnosis verification sample record are as follows: The fault location is read from the fault diagnosis information of industrial equipment, the diagnosis reliability and the basis for abnormal propagation are corrected and written into the same diagnosis identifier to form a diagnosis and treatment trigger package. Then, the treatment priority index is determined according to the diagnosis and treatment trigger package, and treatment information is generated according to the treatment priority index. The inspection and confirmation content is collected by using the information on handling and processing, and the inspection and confirmation content, abnormal segments before maintenance and retest segments after maintenance are bound with the fault diagnosis information of industrial equipment to form a fault diagnosis verification sample record.

[0014] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the data-driven industrial equipment fault diagnosis method as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the data-driven industrial equipment fault diagnosis method as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By dividing the standard action condition data sequence into action condition segments and establishing a load-action phase continuity benchmark by combining the load continuity relationship between adjacent actions, the identification of action switching transition zones, continuous load states, and load continuity disturbances is realized, thereby reducing false alarms during action switching and improving the reliability of anomaly identification under complex working conditions; by establishing a load energy flow diagnostic chain through action phase net deviation data and performing anomaly front-to-back level counter-evidence isolation along the load energy flow diagnostic chain, the verification of whether anomalies continue to spread along the energy transfer path is realized, thereby improving the accuracy of fault source location, enhancing diagnostic credibility, and improving sample update quality. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a data-driven industrial equipment fault diagnosis method.

[0019] Figure 2 A flowchart for generating standard motion condition data sequences.

[0020] Figure 3 A flowchart for generating net deviation data of the action phase.

[0021] Figure 4 A flowchart for creating a fault diagnosis verification sample record.

[0022] Figure 5 This is a graph showing the false alarm rate during action switching.

[0023] Figure 6 This is a comparison diagram of the counterfactual subsequent response trajectory and the actual subsequent response trajectory. Detailed Implementation

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0027] Reference Figures 1-6 This is one embodiment of the present invention, which provides a data-driven industrial equipment fault diagnosis method, including the following steps: S1. Collect real-time operating data of industrial equipment and perform synchronous calibration and operating condition binding based on a unified time reference to obtain a standard action condition data sequence.

[0028] The system collects real-time operating data of industrial equipment and writes the local collection time, data source identifier, and operating status identifier into the real-time operating data of the industrial equipment. Then, it maps the local collection time to the unified time reference corresponding to the main controller of the industrial equipment and performs time correction in combination with action event anchor points to form a corrected operating frame.

[0029] The specific process includes synchronously reading real-time operating data of the industrial equipment according to the diagnostic cycle of the main controller of the industrial equipment. Among them, control command data is obtained from the action control records of lifting, lowering, running, braking and stopping issued by the main controller of the industrial equipment; mechanism feedback data is obtained from the action status, running direction and positioning feedback returned by the corresponding actuator; load response data is obtained from the load change recorded by the load detection device; drive response data is obtained from the drive running status recorded by the drive controller; and structural response data is obtained from the force change recorded by the structural monitoring sensor.

[0030] The local acquisition time, data source identifier, and operating status identifier are written into each piece of real-time operating data of industrial equipment, so that control command data, mechanism feedback data, load response data, drive response data, and structural response data all have traceable acquisition time and operating status. The local acquisition time is mapped to the unified time base corresponding to the main controller of the industrial equipment. The action event anchor point is determined according to the time corresponding to the start, switch, and stop of the action in the control command data and the time of entering the corresponding action state in the mechanism feedback data. The time offset between different data sources is corrected by the action event anchor point, so that the real-time operating data of industrial equipment under the same action event is aligned on the unified time base, forming a corrected operating frame.

[0031] The corrected running frames are synchronously resampled, data validity verified, and anomaly sampling reliably corrected according to a unified diagnostic cycle. The operating conditions are then bound together with the action change status and load change status, and continuously encapsulated in a unified time sequence to obtain a standard action operating condition data sequence.

[0032] The specific process includes: determining continuous diagnostic moments in the corrected operation frame according to a unified diagnostic cycle; aligning control command data, mechanism feedback data, load response data, drive response data, and structural response data in the corrected operation frame to the corresponding diagnostic moments; for continuously changing acquisition content, synchronous resampling is performed using adjacent valid acquisition content; for status-based acquisition content, the most recent valid state before the diagnostic moment is retained and written; verifying the data validity of the resampled corrected operation frame based on whether the acquisition content is within the equipment operating boundary, whether the changes between adjacent diagnostic moments are continuous, and whether the local acquisition moment and data source identifier are valid; for abnormal sampling positions with a single jump and continuous acquisition content before and after, reliable correction is performed using adjacent valid acquisition content; for continuous abnormal sampling positions, the abnormal identifier is retained to form a reliable synchronous diagnostic frame.

[0033] Based on the action change status in the trusted synchronous diagnostic frame, determine the start, hold, switch, and stop status of the action at the current diagnostic time. Combined with the load change status in the trusted synchronous diagnostic frame, determine the load change status at the current diagnostic time. Write the action change status and load change status together into the corresponding diagnostic time to complete the working condition binding. In a unified time order, continuously encapsulate the trusted sampling content, validity verification content, action change status, and load change status corresponding to each diagnostic time to obtain the standard action working condition data sequence.

[0034] It should be noted that the action change state refers to the state of the industrial equipment at the current diagnostic moment, which is determined by the control command data and the mechanism feedback data, including the state of action start, action maintenance, action switching, and action stop; the load change state refers to the state of the industrial equipment at the current diagnostic moment, which is determined by the load response data, including the state of no-load, load-bearing, load continuation, and load change.

[0035] S2. Divide the standard action condition data sequence into action conditions, establish a load-action phase continuity benchmark by combining the load continuity relationship between adjacent actions, and perform phase stripping on non-fault disturbances caused by action switching to form action phase net deviation data.

[0036] The motion condition binding identifiers in the standard motion condition data sequence are compared frame by frame in chronological order. When the motion condition binding identifier changes, the moment of change is taken as the initial switching point. Then, the process is traced back to the diagnostic moment when the control command data begins to change and extended backward to the diagnostic moment when the mechanism feedback data enters the new motion state. When the load change maintains a continuous bearing state, the traced starting point to the extended end point is determined as the motion switching transition zone, and the continuous data between two adjacent motion switching transition zones is divided into a motion candidate segment set.

[0037] The specific process includes: reading the motion condition binding identifiers in the standard motion condition data sequence according to a unified time sequence, and comparing frame by frame whether the motion condition binding identifiers of adjacent diagnostic moments have changed; when the motion condition binding identifier changes from motion hold to motion start, from motion start to motion switch, from motion switch to motion stop, or from one motion state to another, the diagnostic moment when the motion condition binding identifier changes is determined as the initial switching point; around the initial switching point, reading the control command data and mechanism feedback data in adjacent diagnostic moments, looking backward to the diagnostic moment when the control command data began to change as the tracing starting point, and looking backward to the diagnostic moment when the mechanism feedback data stably entered the new motion state as the extension endpoint, so that the data before and after the initial switching point... The control change process and mechanism response process are fully covered; after determining the tracing start point and the extension end point, the load change state between the tracing start point and the extension end point is read. When the load change state remains continuous and there is no change from load to no load, the continuous data between the tracing start point and the extension end point is determined as the action switching transition zone; after determining all action switching transition zones in chronological order, the diagnosis time after the end of the previous action switching transition zone is taken as the start position of the action candidate segment, and the diagnosis time before the start of the next action switching transition zone is taken as the end position of the action candidate segment. The data that maintains a continuous action state between two adjacent action switching transition zones is divided into an action candidate segment, and a set of action candidate segments is formed according to the order of action occurrence.

[0038] A continuous load comparison is performed on the end load state of the previous action candidate segment and the beginning load state of the next action candidate segment in the action candidate segment set. When the change range of the beginning load state of the next action candidate segment relative to the end load state of the previous action candidate segment is within the load continuation judgment range, and there is no unloading breakpoint between the end of the previous action candidate segment and the beginning of the next action candidate segment, it is determined that there is a load continuation relationship between the previous action candidate segment and the next action candidate segment. Adjacent action candidate segments with load continuation relationship are connected in chronological order to form a load-action phase continuity diagram.

[0039] The specific process includes: based on the action candidate segment set, sequentially reading adjacent preceding and following action candidate segments according to the action occurrence order; extracting the end load state from the end position of the preceding action candidate segment and the start load state from the start position of the following action candidate segment; comparing the difference between the start load state of the following action candidate segment and the end load state of the preceding action candidate segment to determine whether the change in the start load state of the following action candidate segment relative to the end load state of the preceding action candidate segment is within the load continuity determination range; simultaneously reading the load change state from the end of the preceding action candidate segment to the start of the following action candidate segment to verify whether the load continuity is maintained between the end of the preceding action candidate segment and the start of the following action candidate segment, and confirming the preceding action candidate segment... There is no unloading breakpoint from the end of the segment to the beginning of the next action candidate segment; when the change in the load state at the beginning of the next action candidate segment relative to the load state at the end of the previous action candidate segment is within the load continuity judgment range, and there is no unloading breakpoint from the end of the previous action candidate segment to the beginning of the next action candidate segment, it is determined that there is a load continuity relationship between the previous action candidate segment and the next action candidate segment; the previous action candidate segment and the next action candidate segment with load continuity relationship are connected in chronological order, and the previous action candidate segment, the next action candidate segment and the corresponding load continuity relationship are recorded in the connection relationship. After completing the continuous load comparison of all adjacent action candidate segments in the action candidate segment set one by one, a load-action phase continuity diagram is formed.

[0040] It should be noted that the load continuity determination interval is the continuous allowable load variation range obtained by statistically analyzing the rated load range of industrial equipment, the normal fluctuation range of the load detection device, and the load change amplitude of adjacent actions in historical normal continuity segments; the load-action phase continuity diagram is used to record the continuous load relationship between adjacent action candidate segments, reflecting the continuity of the load state before and after action switching, thereby distinguishing between normal load continuity disturbances and real abnormal deviations caused by action switching, and improving the accuracy of action phase net deviation data.

[0041] Based on the load-action phase continuity diagram, historical normal continuity segments are retrieved, and the current action candidate segment is phase-aligned with the historical normal continuity segments to form a load-action phase continuity benchmark. Based on the load-action phase continuity benchmark, non-fault disturbances caused by load continuation during action switching are identified. The non-fault disturbances are extracted from the deviation data of the current action candidate segment, and the action phase net deviation features are extracted. Then, they are encapsulated according to the action occurrence order to form action phase net deviation data.

[0042] The specific process includes, according to the load-action phase continuity diagram, first reading the previous action candidate segment and the next action candidate segment corresponding to the load continuity relationship, and searching in the historical normal continuity segments for segments with the same action type, the same action occurrence order, the same load continuity relationship and similar load change state as historical normal continuity segments; taking the action switching transition area in the current action candidate segment as a reference, aligning the current action candidate segment with the historical normal continuity segment according to the corresponding diagnostic times such as action start, action switching, and mechanism feedback to enter a new action state, so that the current action candidate segment and the historical normal continuity segment correspond in the action process position and the load continuity position, forming a load-action phase continuity benchmark.

[0043] Based on the load-action phase continuity benchmark, the deviation data of the current action candidate segment is compared frame by frame with the normal changes in the corresponding phase positions of the historical normal continuity segments. When the deviation data and the normal change trend corresponding to the load continuity relationship are consistent and concentrated in the action switching transition area, the deviation data is identified as a non-fault disturbance caused by load continuity during the action switching process. The non-fault disturbance is extracted from the deviation data of the current action candidate segment, and the deviation content that still exists after extraction is retained and the action phase net deviation feature is extracted. The action phase net deviation feature is encapsulated according to the action occurrence order to form the action phase net deviation data.

[0044] like Figure 5 The graph shows the false alarm rate during action handover, illustrating the impact of phase alignment window length and load continuation judgment interval on the false alarm rate. The graph uses phase alignment window length as the horizontal axis and action handover false alarm rate as the vertical axis, presenting the false alarm rate curves for load continuation judgment intervals of 4%, 6%, and 8%. A magnified view of the area shows the separation of the curves for different load continuation judgment intervals within the 28-40 frame range. As can be seen, the action handover false alarm rate does not decrease monotonically with changes in phase alignment window length, but rather exhibits a differentiated trend under different load continuation judgment intervals. A reasonable phase alignment window length can more fully cover the action handover transition area and isolate load continuation disturbances, thereby reducing the action handover false alarm rate. This demonstrates that load-action phase continuity benchmark and action phase net deviation data can improve the accuracy of non-fault disturbance identification during the action handover phase.

[0045] S3. Establish a load energy flow diagnostic chain based on the net deviation data of the action phase, perform reverse verification isolation of the abnormal stage along the load energy flow diagnostic chain, determine whether the energy transfer deviation is caused by a single action disturbance, lock the source of the continuous transmission abnormality and extract the fault evolution characteristics, and generate candidate fault diagnosis information.

[0046] Based on the net phase deviation data of the action, phase normalization is performed from the start time to the end time of the action. The net deviation segment is mapped to a set of phase energy port frames. The phase-gated energy transfer confidence between the front-end and back-end ports is calculated according to whether the energy change of the front-end port causes the response change of the back-end port within the allowable phase delay. When the phase-gated energy transfer confidence reaches the energy transfer chaining condition and the response change of the back-end port is later than the energy change of the front-end port, it is determined that there is an effective energy transfer relationship between the front-end and back-end ports, and the corresponding ports are connected to form a candidate load energy flow graph.

[0047] The specific process includes: based on the net deviation data of the action phase, determining the complete action interval according to the start time and end time of the action corresponding to each net deviation segment; normalizing the phase of the sampled content within the complete action interval according to the position of the action process, so that net deviation segments of different durations can be compared under a unified phase scale; after completing the phase normalization, extracting the energy change content corresponding to the previous stage port from the net deviation segment, and extracting the response change content corresponding to the subsequent stage port; writing the energy change content, response change content, and net deviation state at the same phase position into the same phase frame to form a phase energy port frame set.

[0048] Based on the phase energy port frame set, adjacent ports are selected according to the transmission order. At each phase position, the energy change between adjacent phase positions of the preceding port and the response change of the following port after phase delay alignment are calculated. The energy change and response change are multiplied phase-by-phase and averaged to obtain the synchronization lag correlation strength between the preceding and following ports. The actual response phase delay is compared with the reference delay under normal operating conditions, and the delay difference is converted into a delay matching degree. Simultaneously, the energy change trajectory of the preceding port and the response change trajectory of the following port are subtracted phase-by-phase and averaged. The degree of trajectory deviation is obtained; the product of the synchronous lag correlation strength and the time delay matching degree is used as the positive transmission term, and the degree of trajectory deviation is used as the negative suppression term. The proportion of the positive transmission term to the sum of the positive transmission term, the negative suppression term, and the stability term is determined as the phase-gated energy transfer confidence. When the phase-gated energy transfer confidence reaches the energy transfer chaining condition, and the response change of the subsequent port is later than the energy change of the preceding port, it is determined that there is an effective energy transfer relationship between the preceding port and the subsequent port. The ports that satisfy the effective energy transfer relationship are connected in the order of front and rear transmission to form a candidate load energy flow graph.

[0049] It should be noted that the allowable phase delay is the normal transmission time range statistically obtained from the phase interval between the energy change of the preceding port and the response change of the following port in historical normal connection segments; the energy transfer chaining condition is the connection judgment condition determined based on the stable distribution of the confidence of phase-gated energy transfer in historical normal connection segments, used to confirm that the energy change of the preceding port can be effectively transmitted to the response change of the following port; the candidate load energy flow graph is used to represent the candidate transmission relationship of the energy change of the preceding port to the response change of the following port in the action phase of industrial equipment, which can provide a basis for screening out false connections and establishing a load energy flow diagnostic chain.

[0050] The expression for calculating the confidence level of phase-gated energy transfer between upstream and downstream ports is as follows: ; in, Indicates the front-end port To the downstream port Phase-gated energy transfer confidence between phases; Indicates the front-end port number; Indicates the downstream port number; It represents a very small positive number.

[0051] It should be noted that, Indicates the front-end port Energy changes and subsequent ports The synchronous lag correlation strength of the response change is expressed as: ; in, This represents the total number of sampling points after phase normalization; Indicates the phase sampling point number; Indicates the front-end port In the Normalized energy values ​​at each phase sampling point; Indicates the front-end port exist Normalized energy values ​​at each phase sampling point; Indicates the downstream port Phase delay Aligned Normalized response values ​​at each phase sampling point; This represents the phase delay corresponding to the energy change at the front-end port being transmitted to the response change at the rear-end port; Indicates the downstream port Phase delay Aligned Normalized response values ​​at each phase sampling point.

[0052] It should be noted that, Indicates the front-end port To the downstream port The degree of matching between the phase delay and the reference delay is expressed as: ; in, Indicates the front-end port To the downstream port The reference phase delay between them under normal operating conditions.

[0053] It should be noted that, Indicates the front-end port Energy changes and subsequent ports The degree of deviation between the response change trajectories after phase delay alignment is expressed as: ; Furthermore, in this expression, each term is calculated from the normalized energy value, the normalized response value, and the normalized phase delay, so the whole is a dimensionless index. The value range is from 0 to 1. The closer the value is to 1, the more likely the energy change of the front-end port is to cause the response change of the back-end port within the phase delay. The more suitable it is to connect the front-end and back-end ports to form a candidate load energy flow graph.

[0054] Based on the candidate load energy flow graph, check step by step along the port connection direction whether the energy change of the preceding port is earlier than the response change of the following port, and check whether the following port responds within the allowable phase delay after the energy of the preceding port increases. If the port connection only shows synchronous fluctuation, does not conform to the forward and backward transmission order, and the response delay exceeds the phase delay interval, then delete the corresponding port connection to form a stable load energy flow graph. Then, select the path from the stable load energy flow graph that continuously transmits from the action phase source end with the earliest net deviation to the final abnormal response end, and write the energy transmission confidence and net deviation propagation status in the path into the path node to form a load energy flow diagnostic chain.

[0055] The specific process includes: based on the candidate load energy flow graph, reading the corresponding front-end port and back-end port for each port connection according to the port connection direction, and determining the diagnostic time when the energy of the front-end port increases, the diagnostic time when the response of the back-end port changes, and the actual response delay between the front-end port and the back-end port respectively; when the diagnostic time when the response of the back-end port changes is later than the diagnostic time when the energy of the front-end port increases, and the actual response delay falls within the allowable phase delay, and the direction of the response change of the back-end port corresponds to the direction of the energy change of the front-end port, the port connection is retained; when the response change of the back-end port is not later than the energy change of the front-end port, or the actual response delay exceeds the phase delay interval, or the front-end port and the back-end port only fluctuate synchronously at the same diagnostic time and lack a forward and backward transmission sequence, the port connection is deleted from the candidate load energy flow graph. After all port connections are checked, a stable load energy flow graph is formed.

[0056] In the stable load energy flow diagram, the path node with the earliest net deviation is found according to the order of action phase occurrence, and the path node with the earliest net deviation is determined as the action phase source. From the action phase source, the connection along the reserved port is traced step by step backward. When tracing to each subsequent port, it is checked whether the subsequent port continues the net deviation propagation state of the previous port within the phase delay interval. If the response change of the subsequent port is later than the energy change of the previous port and the net deviation propagation state can continue, then the subsequent port is added to the continuous transmission path. If the subsequent port does not continue the net deviation propagation state, the tracing in the corresponding direction is stopped. When the continuous transmission path reaches the final abnormal response end, the energy transmission confidence and net deviation propagation state corresponding to each path node in the continuous transmission path are written into the path node to form a load energy flow diagnostic chain.

[0057] It should be noted that the phase delay interval is the normal response delay range statistically obtained from the phase interval between the energy change of the preceding port and the response change of the following port in the historical normal connection segment; the stable load energy flow graph is used to retain the effective energy transfer relationship that conforms to the preceding and following transmission sequence and the phase delay interval, and can eliminate the false connection formed by synchronization fluctuations, providing a reliable basis for accurately establishing the load energy flow diagnostic chain.

[0058] The path nodes in the load energy flow diagnostic chain are divided into a pre-stage observation segment, a current isolation segment, and a post-stage verification segment. The net deviation trajectory of the current isolation segment is replaced with the normal baseline trajectory under the corresponding action phase. At the same time, the real energy change of the pre-stage observation segment is retained to generate the counterfactual post-stage response trajectory.

[0059] The specific process includes: selecting path nodes to be verified one by one, starting from the action phase source end, according to the sequential order of path nodes in the load energy flow diagnostic chain; reading the energy transfer confidence, net deviation propagation state, phase delay relationship, and action phase position corresponding to the path node to be verified; dividing the continuous path content that has undergone energy changes and been transmitted to the path node to be verified through port connection before the path node to be verified into the pre-observation segment; dividing the action phase content from the beginning to the end of the net deviation trajectory in the path node to be verified into the current isolation segment; and dividing the response that continues along the load energy flow diagnostic chain after the path node to be verified into the current isolation segment. The continuous path content that should change is divided into subsequent verification segments. Normal change content that matches the current isolation segment's action phase, load continuity, and has no recorded abnormal response is found in historical normal continuation segments and used as the normal baseline trajectory for the current isolation segment. The net deviation trajectory in the current isolation segment is replaced with the normal baseline trajectory, while keeping the actual energy change in the preceding observation segment unchanged. Based on the energy transfer confidence and phase delay relationship already written in the load energy flow diagnostic chain, the replaced current isolation segment change is transmitted to the subsequent verification segment, and a predicted response change is generated for the subsequent verification segment after the current isolation segment recovers its normal baseline trajectory, resulting in the counterfactual subsequent response trajectory.

[0060] The counterfactual response trajectory is compared with the actual response trajectory of the subsequent verification segment to determine whether the subsequent anomaly has decayed after the current isolation segment is replaced. If the subsequent anomaly has not decayed, the path node corresponding to the current isolation segment is identified as a single action disturbance node, and the subsequent path node is verified along the load energy flow diagnostic chain. If the subsequent anomaly continues to decay, the path node corresponding to the current isolation segment is locked as the source of the continuously propagating anomaly.

[0061] The specific process includes aligning the counterfactual response trajectory with the actual response trajectory of the subsequent verification segment at the same action phase position point by point, comparing whether the response intensity in the subsequent verification segment decreases due to the replacement of the net deviation trajectory of the current isolation segment with the normal baseline trajectory, and then comparing whether the response duration shortens and whether the response change direction returns from the abnormal direction to the normal direction. When the counterfactual response trajectory is still close to the actual response trajectory of the subsequent verification segment, and the abnormal intensity, duration, and change direction in the subsequent verification segment do not significantly weaken, it indicates that the replacement of the net deviation trajectory of the current isolation segment has not changed the subsequent abnormal behavior, and the path node corresponding to the current isolation segment does not belong to the continuous transmission source. The path node corresponding to the current isolation segment is identified as a single action disturbance node, and the subsequent path node is verified along the load energy flow diagnostic chain. When the counterfactual response trajectory shows a decrease in abnormal intensity, a shortening of abnormal duration, and a decline in abnormal change direction relative to the actual response trajectory of the subsequent verification segment, and the subsequent path node also shows the same weakening trend, it indicates that the net deviation trajectory of the current isolation segment has a transmission effect on the subsequent abnormality, and the path node corresponding to the current isolation segment is locked as the source of the continuously transmitted abnormality.

[0062] It should be noted that continuous attenuation refers to the continuous decrease in the net deviation intensity, response duration, or energy transfer confidence of subsequent path nodes during the process of propagating along the load energy flow diagnostic chain. It originates from the net deviation propagation state of each path node, the comparison results between the counterfactual subsequent response trajectory and the true response trajectory, and the stepwise changes in energy transfer confidence.

[0063] like Figure 6 This diagram compares the counterfactual subsequent response trajectory with the true subsequent response trajectory to illustrate the effectiveness of isolation between the preceding and following stages in verifying the continuous propagation of the anomaly source. The diagram uses diagnosis time as the horizontal axis and normalized subsequent response amplitude as the vertical axis, displaying the true subsequent response trajectory, the counterfactual subsequent response trajectory, and the normal baseline trajectory. The true subsequent response trajectory continuously deviates from the normal baseline trajectory in its later stages, while the counterfactual subsequent response trajectory significantly reverts to the normal baseline trajectory after being replaced by the normal baseline trajectory in the current isolation segment. This indicates that the net deviation trajectory in the current isolation segment has a propagating effect on subsequent anomalies. This trajectory comparison visually demonstrates whether the anomaly continues to spread along the load energy flow diagnostic chain and supports the process of locking down the continuously propagating anomaly source, thereby verifying the effectiveness of improving the accuracy of fault source location and enhancing diagnostic reliability.

[0064] The fault evolution features corresponding to the net deviation enhancement process, phase delay change and energy transfer change are extracted from the continuously transmitted anomaly source, and candidate fault diagnosis information is generated based on the fault evolution features.

[0065] The specific process includes: after identifying the continuously propagating anomaly source, reading the net deviation propagation state, energy transfer confidence, and phase delay relationship of each path node from the continuously propagating anomaly source backward along the load energy flow diagnostic chain; comparing the changes in net deviation intensity and duration of the continuously propagating anomaly source in adjacent action phases to determine whether the net deviation has developed from a local short-term change to a continuously increasing change, thus obtaining the net deviation enhancement process; and comparing the actual response delay and phase delay interval between the continuously propagating anomaly source and each subsequent path node to determine whether there is response advance, response lag, or delay in anomaly propagation. The phase delay change is obtained by stretching the image; the changes in the energy transfer confidence between each path node during the propagation process are compared to determine whether there is continuous attenuation, local amplification, or transmission interruption in energy transfer, thus obtaining the energy transfer change; the net deviation enhancement process, phase delay change, and energy transfer change are combined in the order of the load energy flow diagnosis chain to form fault evolution characteristics, and based on the location of the continuously propagating abnormal source, abnormal propagation path, and final abnormal response end pointed to by the fault evolution characteristics, candidate fault diagnosis information containing candidate fault location, candidate fault type, and abnormal propagation basis is generated.

[0066] S4. Construct a fault propagation delay evidence chain using candidate fault diagnosis information and verify the abnormal propagation order to generate delay verification fault information. Then, review the delay verification fault information under similar working conditions and correct the diagnostic credibility to obtain industrial equipment fault diagnosis information. Generate handling information based on the industrial equipment fault diagnosis information and bind the maintenance confirmation content to form a fault diagnosis verification sample record.

[0067] The anomaly source location, anomaly propagation path, and candidate fault type in the candidate fault diagnosis information are converted into fault propagation evidence tokens. A time delay pairing relationship is established between adjacent fault propagation evidence tokens according to the propagation direction in the load energy flow diagnosis chain. Then, the phase delay interval is called according to the anomaly arrival time and action phase difference of adjacent fault propagation evidence tokens. The anomaly arrival order of adjacent fault propagation evidence tokens is verified according to the phase delay interval. The time delay pairing relationships that satisfy the sequential propagation relationship are connected to form a fault propagation time delay evidence chain.

[0068] The specific process includes: reading the anomaly source location, anomaly propagation path, and candidate fault type item by item from the candidate fault diagnosis information; and writing the path node corresponding to the anomaly source location, the continuous path nodes in the anomaly propagation path, and the candidate fault type into the corresponding fault propagation evidence token according to the propagation direction in the load energy flow diagnosis chain, so that each fault propagation evidence token is bound to the path location, anomaly arrival time, and action phase position; reading adjacent fault propagation evidence tokens according to the propagation direction in the load energy flow diagnosis chain, establishing a time delay pairing relationship between the previous fault propagation evidence token and the next fault propagation evidence token, and based on the previous fault propagation... The abnormal arrival time of the evidence token, the abnormal arrival time of the subsequent fault propagation evidence token, and the action phase difference call phase delay interval; when the abnormal arrival time of the subsequent fault propagation evidence token is later than that of the previous fault propagation evidence token, the abnormal arrival interval falls within the phase delay interval, and the path position of the subsequent fault propagation evidence token is downstream of the path position of the previous fault propagation evidence token, it is determined that the previous fault propagation evidence token and the subsequent fault propagation evidence token satisfy the sequential propagation relationship, the corresponding delay pairing relationship is retained, and the delay pairing relationships that satisfy the sequential propagation relationship are connected in order of propagation direction to form a fault propagation delay evidence chain.

[0069] It should be noted that the sequential propagation relationship refers to the transmission relationship along the load energy flow diagnostic chain, where the upstream path node appears first and the downstream path node appears subsequently within the phase delay interval. This relationship is obtained by jointly verifying the path position, anomaly arrival time, and phase delay interval of adjacent fault propagation evidence tokens.

[0070] The abnormal arrival order and phase delay consistency of adjacent fault propagation evidence tokens are checked step by step along the fault propagation delay evidence chain. If the abnormal arrival time of the later fault propagation evidence token is earlier than that of the earlier fault propagation evidence token, and the arrival interval between the later and earlier fault propagation evidence tokens does not fall within the phase delay interval, the corresponding candidate fault diagnosis information is marked as delay inconsistency information and removed from the current fault propagation delay evidence chain. The retained candidate fault diagnosis information is bound to the fault propagation delay evidence chain, and then the candidate fault location, candidate fault type, abnormal arrival order and delay verification status are written to generate delay verification fault information.

[0071] The specific process includes, during the step-by-step verification along the fault propagation delay evidence chain, reading adjacent fault propagation evidence tokens according to the connection order in the fault propagation delay evidence chain, and extracting the abnormal arrival time of the previous fault propagation evidence token, the abnormal arrival time of the next fault propagation evidence token, the arrival interval between the previous and next fault propagation evidence tokens, and the corresponding phase delay interval; when the abnormal arrival time of the next fault propagation evidence token is later than that of the previous fault propagation evidence token, and the arrival interval between the previous and next fault propagation evidence tokens falls within the phase delay interval, the abnormal arrival time of the adjacent fault propagation evidence tokens is determined. If the arrival order and phase delay consistency are established, the corresponding candidate fault diagnosis information is retained. When the abnormal arrival time of the subsequent fault propagation evidence token is earlier than that of the previous fault propagation evidence token, and the arrival interval between the subsequent and previous fault propagation evidence tokens does not fall within the phase delay interval, the corresponding candidate fault diagnosis information is marked as delay inconsistency information, and the corresponding delay pairing relationship is removed from the current fault propagation delay evidence chain. The retained candidate fault diagnosis information is bound to the fault propagation delay evidence chain, and the candidate fault location, candidate fault type, abnormal arrival order, and delay verification status are written into the same record to generate delay verification fault information.

[0072] It should be noted that candidate fault locations refer to industrial equipment mechanisms, components, or connection locations that may fail, initially determined based on the location of the continuously transmitted anomaly source and the anomaly propagation path; candidate fault types refer to fault categories obtained through preliminary matching based on fault evolution characteristics.

[0073] Based on the time delay verification fault information, a current fault review signature is generated, and historical review segments matching the current action sequence are retrieved from the historical operation records according to the current fault review signature.

[0074] The specific process includes reading the candidate fault location, candidate fault type, abnormal arrival order, time delay verification status, and fault propagation time delay evidence chain based on the time delay verification fault information. Then, according to the propagation direction in the fault propagation time delay evidence chain, the abnormal source location, the arrival interval between adjacent fault propagation evidence tokens, the action phase position, and the phase delay interval are organized. The action succession process, abnormal propagation order, and time delay verification content in the same fault propagation process are written into the same review record to form the current fault review signature.

[0075] When retrieving historical data from the historical operation record based on the current fault review signature, the sequence of changes in the action condition binding identifier in the current fault review signature is read, and the action start, action hold, action switch, and action stop are organized into the current action sequence according to their occurrence order. Then, the action condition binding identifiers in the historical operation record are read one by one, and the historical action sequence is extracted in the same way. The historical action sequence is compared with the current action sequence item by item, and historical segments with consistent action state change types, consistent action switch order, and load continuity relationship between adjacent actions are retained. Subsequently, the load change state and action switch transition area in the retained historical segments are read to check whether the load change state in the historical segments maintains continuous bearing, and whether the action switch transition area in the historical segments corresponds to the action switch transition area corresponding to the current fault review signature. The abnormal arrival order and phase delay interval in the historical segments are compared, and historical segments with abnormal arrival order consistent with the current fault review signature and arrival interval falling within the phase delay interval are retained to obtain the historical review segment.

[0076] Based on the maintenance confirmation status of historical review segments, distinguish between historical normal continuation segments and maintenance confirmation fault segments. Calculate the normal counter-evidence support degree of historical normal continuation segments for the current abnormal propagation, and calculate the fault case support degree of maintenance confirmation fault segments for the current abnormal propagation. Then, generate the basic reliability of the delay based on the delay verification status in the delay verification fault information.

[0077] The specific process includes: based on the maintenance confirmation status of historical review segments, reading the maintenance confirmation content and operating status corresponding to each historical review segment; classifying historical review segments without recorded maintenance confirmation content and whose subsequent operating status remains stable into historical normal continuation segments; and classifying historical review segments with recorded maintenance confirmation content that matches the candidate fault location and candidate fault type into maintenance confirmation fault segments; and comparing each historical review segment with the current anomaly propagation process item by item to calculate propagation similarity, specifically comparing the similarity between historical review segments and the current anomaly propagation process in terms of net deviation intensity change, anomaly arrival order, and propagation delay. The closer the net deviation intensity change, the more consistent the anomaly arrival order, and the closer the propagation delay falls within the phase delay, the better. The higher the similarity of propagation within a given interval, the better. The propagation similarity corresponding to historical normal continuation segments is summarized, and the normal counter-evidence support of these segments for the current abnormal propagation is calculated. This indicates the degree to which the current abnormal propagation closely approximates the normal action continuation process. Simultaneously, the propagation similarity corresponding to maintenance-confirmed fault segments is summarized, and the fault case support of these segments for the current abnormal propagation is calculated. This indicates the degree to which the current abnormal propagation closely approximates the confirmed fault process. Based on the delay verification status in the delay verification fault information, the consistency between the abnormal arrival order and phase delay is determined. If the consistency is achieved, a higher basic delay confidence score is generated; otherwise, a lower score is generated.

[0078] The expression for calculating propagation similarity is: ; in, Indicates the first The similarity of propagation between historical review segments and the current abnormal propagation process; Indicates the number of the historical review segment; Indicates the number of sampling points for net deviation intensity; Indicates the serial number of the sampling point for net deviation intensity; This indicates that the current abnormal propagation process is in the [number]th [stage]. Normalized net deviation intensity at each sampling point; Indicates the first The historical review segment in the Normalized net deviation intensity at each sampling point; Indicates the number of nodes with propagation delay; Indicates the sequence number of the node representing the propagation delay; This indicates that the current abnormal propagation process is in the [number]th [stage]. Normalized propagation delay at each propagation node; Indicates the first The historical review segment in the Normalized propagation delay at each propagation node.

[0079] The expression for calculating the normal proof support of historical normal continuation fragments for the current anomalous propagation is: ; in, Indicates the support level of normal counter-evidence; This represents a set of normally continuing historical segments.

[0080] The expression for calculating the support of the fault segment confirmed by maintenance for the current anomaly propagation fault case is as follows: ; in, Indicates the support level of failure cases; This indicates the number of faulty segments confirmed during inspection.

[0081] The reliability of the delay basis is weakened based on the normal counter-evidence support and strengthened based on the failure case support to obtain the corrected diagnostic reliability. Information that meets the diagnostic confirmation conditions based on the corrected diagnostic reliability is written into the industrial equipment failure diagnosis information.

[0082] The specific process includes: using the latency-based reliability as the initial reliability basis for the current candidate fault diagnosis information, and combining it with historical normal continuity segments to weaken the normal counter-evidence support of the current abnormal propagation. When the normal counter-evidence support increases, it indicates that the current abnormal propagation is closer to the normal action continuity process, and the weakening of the latency-based reliability increases accordingly. When the normal counter-evidence support decreases, it indicates that the current abnormal propagation lacks normal action continuity counter-evidence, and the weakening of the latency-based reliability decreases accordingly. After completing the normal counter-evidence weakening, the fault case support of the current abnormal propagation is enhanced by combining it with maintenance-confirmed fault segments. When the fault case support increases, it indicates that the current abnormal propagation is closer to the confirmed fault process, and the enhancement of the weakened latency-based reliability increases accordingly, resulting in the corrected diagnostic reliability. The corrected diagnostic reliability is compared with the diagnostic confirmation conditions. When the corrected diagnostic reliability meets the diagnostic confirmation conditions, the corresponding candidate fault location, candidate fault type, abnormal propagation basis, latency verification status, and corrected diagnostic reliability are written into the industrial equipment fault diagnosis information.

[0083] It should be noted that the diagnostic confirmation conditions are reliable judgment conditions determined based on the stable distribution of the corrected diagnostic reliability in historical maintenance confirmed fault segments, the counter-evidence distribution of historical normal succession segments, and the safety requirements for industrial equipment operation. These conditions are used to confirm that candidate fault diagnosis information can be written into the industrial equipment fault diagnosis information.

[0084] The fault location is read from the fault diagnosis information of industrial equipment, the reliability of the diagnosis and the basis for the abnormal propagation are corrected and written into the same diagnostic identifier to form a diagnosis and treatment trigger package. Then, the treatment priority index is determined according to the diagnosis and treatment trigger package, and treatment information is generated according to the treatment priority index.

[0085] The specific process includes reading the fault location, correcting the diagnostic credibility, and the basis for abnormal propagation from the fault diagnosis information of industrial equipment, and verifying whether the fault location is consistent with the abnormal source location and abnormal propagation path in the basis for abnormal propagation; if consistent, the fault location, corrected diagnostic credibility, basis for abnormal propagation, and time delay verification status are written into the same diagnostic identifier, so that the location information, credibility information, and propagation evidence in the same fault diagnosis process are bound together to form a diagnostic handling trigger package.

[0086] The basic handling level is determined based on the operational impact of the fault location. The urgency of the handling is increased or decreased based on the reliability of the corrected diagnosis. The anomaly propagation evidence is used to determine whether the anomaly has been continuously transmitted along the load energy flow diagnostic chain. When the operational impact of the fault location is high, the reliability of the corrected diagnosis is high, and the anomaly propagation evidence shows that the anomaly is continuously transmitted, the handling priority index is increased. The handling information, which includes the handling object, handling order, handling method, and retesting requirements, is generated according to the handling priority index.

[0087] The inspection and confirmation content is collected by using the information on handling and processing, and the inspection and confirmation content, abnormal segments before maintenance and retest segments after maintenance are bound with the fault diagnosis information of industrial equipment to form a fault diagnosis verification sample record.

[0088] The specific process includes: recording the actual repair location, discovered phenomena, repair methods, component replacements, and repair confirmation time according to the fault location and handling requirements corresponding to the handling information, forming the repair confirmation content; extracting the net deviation data of the action phase, load energy flow diagnostic chain, fault propagation delay evidence chain, and abnormal response data corresponding to the abnormal propagation basis from the standard action condition data sequence before maintenance, based on the same diagnostic identifier in the industrial equipment fault diagnosis information, forming the pre-maintenance abnormal segment; re-collecting the real-time operating data of the industrial equipment according to the same action sequence, and performing synchronous calibration and operating condition binding according to a unified time reference, extracting the retest data at the same action phase position as the pre-maintenance abnormal segment, forming the post-maintenance retest segment; binding the repair confirmation content, the pre-maintenance abnormal segment, and the post-maintenance retest segment with the industrial equipment fault diagnosis information according to the same diagnostic identifier, recording the fault location, corrected diagnostic credibility, abnormal propagation basis, repair confirmation content, and retest comparison content, forming the fault diagnosis verification sample record.

[0089] This embodiment also provides a computer device applicable to the data-driven industrial equipment fault diagnosis method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the data-driven industrial equipment fault diagnosis method proposed in the above embodiment.

[0090] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0091] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the data-driven industrial equipment fault diagnosis method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0092] In summary, this invention achieves the identification of action switching transition zones, continuous load states, and load continuity disturbances by dividing the standard action condition data sequence into action condition segments and establishing a load-action phase continuity benchmark based on the load continuity relationship between adjacent actions. This reduces false alarms during action switching and improves the reliability of anomaly identification under complex working conditions. Furthermore, by establishing a load energy flow diagnostic chain through action phase net deviation data and performing anomaly isolation along the load energy flow diagnostic chain, this invention verifies whether anomalies continue to spread along the energy transfer path, thereby improving the accuracy of fault source location, enhancing diagnostic credibility, and improving sample update quality.

[0093] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A data-driven method for fault diagnosis of industrial equipment, characterized in that, include: Collect real-time operating data of industrial equipment and perform synchronous calibration and operating condition binding based on a unified time reference to obtain a standard action condition data sequence; The standard action condition data sequence is divided into action conditions. The load-action phase continuity benchmark is established by combining the load continuity relationship between adjacent actions. The non-fault disturbances caused by action switching are phase stripped to form action phase net deviation data. A load energy flow diagnostic chain is established based on the net deviation data of the action phase. The abnormality is isolated before and after the abnormality along the load energy flow diagnostic chain. It is determined whether the energy transfer deviation is caused by a single action disturbance. The abnormality source that is continuously transmitted is locked and the fault evolution characteristics are extracted to generate candidate fault diagnosis information. By using candidate fault diagnosis information to construct a fault propagation delay evidence chain and verify the abnormal propagation order, delay verification fault information is generated. Then, the delay verification fault information is reviewed under similar working conditions and the diagnostic credibility is corrected to obtain industrial equipment fault diagnosis information. Based on the industrial equipment fault diagnosis information, handling information is generated and maintenance confirmation content is bound to form a fault diagnosis verification sample record.

2. The data-driven industrial equipment fault diagnosis method as described in claim 1, characterized in that, The specific steps for obtaining the standard operating condition data sequence are as follows: Collect real-time operating data of industrial equipment, and write the local collection time, data source identifier and operating status identifier into the real-time operating data of industrial equipment. Then, map the local collection time to the unified time reference corresponding to the main controller of industrial equipment, and perform time correction in combination with action event anchor points to form a corrected operating frame. The corrected running frames are synchronously resampled, data validity verified, and anomaly sampling reliably corrected according to a unified diagnostic cycle. The operating conditions are then bound together with the action change status and load change status, and continuously encapsulated in a unified time sequence to obtain a standard action operating condition data sequence.

3. The data-driven industrial equipment fault diagnosis method as described in claim 2, characterized in that, The specific steps for generating the net deviation data of the action phase are as follows: The motion condition binding identifiers in the standard motion condition data sequence are compared frame by frame in chronological order. When the motion condition binding identifier changes, the moment of change is taken as the initial switching point. Then, the process is traced back to the diagnostic moment when the control command data begins to change and extended backward to the diagnostic moment when the mechanism feedback data enters the new motion state. When the load change maintains a continuous bearing state, the traced start point to the extended end point is determined as the motion switching transition zone, and the continuous data between two adjacent motion switching transition zones is divided into a set of motion candidate segments. The load state at the end of the previous action candidate segment and the load state at the beginning of the next action candidate segment in the action candidate segment set are continuously compared. When the change of the load state at the beginning of the next action candidate segment relative to the load state at the end of the previous action candidate segment is within the load continuation judgment range, and there is no unloading breakpoint between the end of the previous action candidate segment and the beginning of the next action candidate segment, it is determined that there is a load continuation relationship between the previous action candidate segment and the next action candidate segment. Adjacent action candidate segments with load continuation relationship are connected in chronological order to form a load-action phase continuity diagram. Based on the load-action phase continuity diagram, historical normal continuity segments are retrieved, and the current action candidate segment is phase-aligned with the historical normal continuity segments to form a load-action phase continuity benchmark. Based on the load-action phase continuity benchmark, non-fault disturbances caused by load continuation during action switching are identified. The non-fault disturbances are extracted from the deviation data of the current action candidate segment, and the action phase net deviation features are extracted. Then, they are encapsulated according to the action occurrence order to form action phase net deviation data.

4. The data-driven industrial equipment fault diagnosis method as described in claim 1, characterized in that, The specific steps for establishing a load energy flow diagnostic chain based on the net deviation data of the action phase are as follows: Based on the net phase deviation data of the action, phase normalization is performed from the start time to the end time of the action. The net deviation segment is mapped to a set of phase energy port frames. The confidence of phase-gated energy transfer between the front-end and back-end ports is calculated according to whether the energy change of the front-end port causes the response change of the back-end port within the allowable phase delay. When the confidence level of phase-gated energy transfer reaches the chaining condition for energy transfer, and the response change of the downstream port is later than the energy change of the upstream port, it is determined that there is an effective energy transfer relationship between the upstream and downstream ports, and the corresponding ports are connected to form a candidate load energy flow graph. Based on the candidate load energy flow graph, check step by step along the port connection direction whether the energy change of the front-end port is earlier than the response change of the rear-end port, and check whether the rear-end port has a corresponding response within the allowable phase delay after the energy of the front-end port rises. If the port connection only shows synchronous fluctuation, does not conform to the front-end and rear-end transmission order and the response delay exceeds the phase delay interval, then delete the corresponding port connection to form a stable load energy flow graph. Then, select a path from the earliest net deviation action phase source end to the final abnormal response end from the stable load energy flow graph, and write the energy transfer confidence and net deviation propagation status in the path node to form a load energy flow diagnostic chain.

5. The data-driven industrial equipment fault diagnosis method as described in claim 1, characterized in that, The specific steps for generating candidate fault diagnosis information are as follows: The path nodes in the load energy flow diagnostic chain are divided into the pre-stage observation segment, the current isolation segment, and the post-stage verification segment. The net deviation trajectory of the current isolation segment is replaced with the normal baseline trajectory under the corresponding action phase. At the same time, the real energy change of the pre-stage observation segment is retained to generate the counterfactual post-stage response trajectory. By comparing the counterfactual response trajectory with the actual response trajectory of the subsequent verification segment, it can be determined whether the subsequent anomaly is attenuated after the current isolation segment is replaced. If the subsequent anomaly does not decay, the path node corresponding to the current isolation segment is identified as a single action disturbance node, and the subsequent path node is verified along the load energy flow diagnostic chain; if the subsequent anomaly continues to decay, the path node corresponding to the current isolation segment is locked as the source of the continuously propagating anomaly. The fault evolution features corresponding to the net deviation enhancement process, phase delay change and energy transfer change are extracted from the continuously transmitted anomaly source, and candidate fault diagnosis information is generated based on the fault evolution features.

6. The data-driven industrial equipment fault diagnosis method as described in claim 1 or 5, characterized in that, The specific steps for generating delay verification fault information are as follows: The anomaly source location, anomaly propagation path and candidate fault type in the candidate fault diagnosis information are converted into fault propagation evidence tokens. A time delay pairing relationship is established between adjacent fault propagation evidence tokens according to the propagation direction in the load energy flow diagnosis chain. Then, the phase delay interval is called according to the abnormal arrival time and action phase difference of adjacent fault propagation evidence tokens. The abnormal arrival order of adjacent fault propagation evidence tokens is verified according to the phase delay interval. The time delay pairing relationships that satisfy the sequential propagation relationship are connected to form a fault propagation time delay evidence chain. The abnormal arrival order and phase delay consistency of adjacent fault propagation evidence tokens are checked step by step along the fault propagation delay evidence chain. If the abnormal arrival time of the later fault propagation evidence token is earlier than that of the earlier fault propagation evidence token, and the arrival interval between the later and earlier fault propagation evidence tokens does not fall within the phase delay interval, the corresponding candidate fault diagnosis information is marked as delay inconsistency information and removed from the current fault propagation delay evidence chain. The retained candidate fault diagnosis information is bound to the fault propagation delay evidence chain, and then the candidate fault location, candidate fault type, abnormal arrival order and delay verification status are written to generate delay verification fault information.

7. The data-driven industrial equipment fault diagnosis method as described in claim 6, characterized in that, The specific steps for obtaining industrial equipment fault diagnosis information are as follows: Based on the time delay verification fault information, generate the current fault review signature, and retrieve the historical review segment that matches the current action sequence in the historical operation record according to the current fault review signature; Based on the maintenance confirmation status of historical review segments, distinguish between historical normal continuation segments and maintenance confirmation fault segments. Calculate the normal counter-evidence support degree of historical normal continuation segments for the current abnormal propagation, and calculate the fault case support degree of maintenance confirmation fault segments for the current abnormal propagation. Then, generate the basic reliability of the delay based on the delay verification status in the delay verification fault information. The reliability of the delay basis is weakened based on the normal counter-evidence support and strengthened based on the failure case support to obtain the corrected diagnostic reliability. Information that meets the diagnostic confirmation conditions based on the corrected diagnostic reliability is written into the industrial equipment failure diagnosis information.

8. The data-driven industrial equipment fault diagnosis method as described in claim 7, characterized in that, The specific steps for forming the fault diagnosis verification sample record are as follows: The fault location is read from the fault diagnosis information of industrial equipment, the diagnosis reliability and the basis for abnormal propagation are corrected and written into the same diagnosis identifier to form a diagnosis and treatment trigger package. Then, the treatment priority index is determined according to the diagnosis and treatment trigger package, and treatment information is generated according to the treatment priority index. The inspection and confirmation content is collected by using the information on handling and processing, and the inspection and confirmation content, abnormal segments before maintenance and retest segments after maintenance are bound with the fault diagnosis information of industrial equipment to form a fault diagnosis verification sample record.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the data-driven industrial equipment fault diagnosis method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the data-driven industrial equipment fault diagnosis method according to any one of claims 1 to 8.

Images