Fault monitoring method and system using AI intelligent image recognition

Abstract

The invention discloses a fault monitoring method and system using AI intelligent image recognition, and the method achieves the continuous monitoring of the subtle change of equipment through collecting the time-space information image data of the operation state of the equipment and extracting the dynamic characteristics of motion trail, deformation rate, spectrum change and the like. The dynamic features are associated with the historical state of the device, and a dynamic incidence matrix is constructed to analyze potential trend deviation points in a time sequence, so that possible faults are predicted, and not only occurred problems are identified. An abnormal mode matching rule is generated by calculating a prediction window and combining the prediction window with equipment operation environment parameters, meanwhile, a response threshold value is adjusted to adapt to equipment performance attenuation characteristics, and finally an early warning signal is generated through cross validation of the abnormal mode matching rule and the prediction window. According to the method, the abnormal development trend can be identified before the equipment has an obvious fault, early warning is given out in advance, the transformation from post-event identification to pre-event early warning is realized, and the fault prevention capability is remarkably enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault monitoring, and specifically relates to a fault monitoring method and system using AI intelligent image recognition. Background Art

[0002] Currently, image recognition-based fault detection methods are widely used in the field of industrial equipment operating status monitoring. Existing technologies typically use cameras to capture images of the equipment during operation and combine them with image processing algorithms to identify abnormalities such as component displacement, surface damage, or temperature abnormalities. These methods primarily rely on comparing the current image with a standard template to determine whether there are any features that significantly deviate from normal conditions, thereby enabling fault identification.

[0003] However, this monitoring method, which focuses on identifying existing faults, has obvious limitations: it is unable to detect potential risk trends in a timely manner before a fault occurs, resulting in delayed response and difficulty in effectively preventing the occurrence of sudden faults. Summary of the Invention

[0004] The purpose of the present invention is to provide a fault monitoring method and system using AI intelligent image recognition, which can identify abnormal development trends before obvious equipment failures occur, thereby issuing early warnings, improving the safety of equipment operation and maintenance efficiency, and solving the problems raised in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a fault monitoring method using AI intelligent image recognition, comprising the following steps: Collecting image data of spatiotemporal information of the operating status of the device, and extracting dynamic features of the image data, wherein the dynamic features include motion trajectory, deformation rate, and spectral change; Constructing a dynamic correlation matrix through the mapping relationship between the dynamic features and the historical status of the equipment, analyzing the time series of the dynamic correlation matrix, and identifying potential trend deviation points; Calculating a prediction window based on the trend deviation point, the prediction window is generated by comparing the difference between the current state and a preset threshold, and combining the prediction window with the device operating environment parameters to generate an abnormal pattern matching rule; The response threshold of the abnormal pattern matching rule is adjusted to adapt to the device performance attenuation characteristics, and an early warning signal is generated through cross-validation of the abnormal pattern matching rule and the prediction window.

[0006] Preferably, the image data of the temporal and spatial information of the operating status of the equipment is collected, including: Arrange multi-angle imaging devices in the equipment area to obtain continuous frame image sequences at different viewing angles; Performing pixel-level time alignment on the image sequence to extract displacement changes between adjacent frames; Divide the dynamic area based on the displacement change, and mark the area where the change amplitude exceeds the set reference as the area of ​​interest; Infrared thermal imaging data is superimposed in the area of ​​interest to generate a fused image containing temperature distribution and deformation information.

[0007] Preferably, extracting the dynamic features of the image data includes: Selecting a number of interest points in the fused image, analyzing the position offset of each interest point in consecutive frames, and calculating a motion vector; Determine the motion trajectory of each point of interest using the motion vector, form a trajectory line by connecting the positions in consecutive frames, and analyze the trajectory line to obtain the deformation rate; Spectral changes are detected for the infrared thermal imaging data superimposed within the area of ​​interest to identify temperature fluctuation trends.

[0008] Preferably, a dynamic association matrix is ​​constructed by mapping the dynamic features to the historical status of the device, including: Collecting data on the deformation rate and spectral change, and recording the equipment operating parameters at the corresponding time, to form a basic data pair; Calculating a similarity score between each data point and the historical state based on the basic data pair, and matching the current dynamic feature with the historical state using the similarity score to generate a matching matrix; The matching matrix is ​​normalized to obtain a final dynamic correlation matrix.

[0009] Preferably, analyzing the time series of the dynamic correlation matrix to identify potential trend deviation points includes: Arranging the dynamic correlation matrix in chronological order to generate a time series set, and calculating an average correlation strength within a sliding window based on the time series set; Compare the average correlation strength of the current window with that of the previous window. If the difference exceeds the set range, mark the window as an abnormal fluctuation range. The earliest time point at which a significant change occurs within the abnormal fluctuation range is located as a potential trend deviation point.

[0010] Preferably, calculating the prediction window according to the trend deviation point includes: Obtain the potential trend deviation point and extract the state change rate within a set range before and after the time point; Comparing the state change rate with a preset threshold to determine whether there is an accelerating change trend; The starting point and the ending point of the prediction window are determined based on the duration of the accelerating change trend to form a prediction window interval.

[0011] Preferably, combining the prediction window with the equipment operating environment parameters to generate an abnormal pattern matching rule includes: Collecting multi-dimensional parameters of the environment in which the device is located within the prediction window, including temperature, humidity, and vibration intensity; Classify and label the image features in the prediction window based on the environmental parameters to form a state sequence with environmental labels; Count the recurring patterns of state sequences under the same environmental conditions and define similarity indicators to measure the consistency of states between different time points; The state combinations that appear frequently and meet the similarity threshold are summarized as abnormal patterns, and the corresponding matching rules are generated.

[0012] Preferably, adjusting the response threshold of the abnormal pattern matching rule comprises the following steps: Record the state change curve of the equipment during multiple operation cycles and extract the basic response level at the end of each cycle; Calculating a long-term attenuation factor based on the base response level to quantify the trend of device performance degradation over time; Introducing the attenuation factor into the threshold adjustment of the current matching rule to update the similarity judgment benchmark; The updated benchmark is used to replace the original threshold and dynamically evaluate the degree of match between the new acquisition state and the abnormal pattern.

[0013] Preferably, generating an early warning signal by cross-validation of the abnormal pattern matching rule and the prediction window includes: Compare the acquisition state with the high-frequency state combination in the abnormal pattern matching rule in real time within the current prediction window, and count the number of matches; Calculating a matching density based on the number of matches, comparing the matching density with an updated benchmark, and confirming the presence of an abnormal development trend if the matching density exceeds the benchmark; A graded warning signal is triggered according to the abnormal development trend, and the signal strength increases with the matching difference to achieve warning output.

[0014] On the other hand, the present invention proposes a fault monitoring system using AI intelligent image recognition, comprising: A dynamic feature extraction module is used to collect image data of the spatiotemporal information of the operating status of the device and extract dynamic features of the image data, wherein the dynamic features include motion trajectory, deformation rate and spectral change; A trend identification module is used to construct a dynamic correlation matrix based on the mapping relationship between the dynamic features and the historical status of the equipment, analyze the time series of the dynamic correlation matrix, and identify potential trend deviation points; a matching rule generation module, configured to calculate a prediction window based on the trend deviation point, the prediction window being generated by comparing the difference between the current state and a preset threshold, and combining the prediction window with equipment operating environment parameters to generate an abnormal pattern matching rule; The early warning signal generation module is used to adjust the response threshold of the abnormal pattern matching rule to adapt to the equipment performance attenuation characteristics, and generate an early warning signal through cross-validation of the abnormal pattern matching rule and the prediction window.

[0015] Technical effects and advantages of the present invention: Compared with the existing technology, the fault monitoring method and system using AI intelligent image recognition proposed in the present invention have the following advantages: The present invention realizes continuous monitoring of subtle changes in the equipment by collecting spatiotemporal information image data of the equipment's operating status and extracting dynamic features such as motion trajectory, deformation rate, and spectral changes. These dynamic features are associated with the equipment's historical status, and a dynamic correlation matrix is ​​constructed to analyze potential trend deviation points in the time series, thereby predicting possible failures rather than just identifying problems that have already occurred. By calculating the prediction window and combining it with the equipment's operating environment parameters to generate abnormal pattern matching rules, while adjusting the response threshold to adapt to the equipment's performance attenuation characteristics, and finally using the abnormal pattern matching rules and the prediction window for cross-validation to generate early warning signals; this method effectively improves the safety of equipment operation and maintenance efficiency, can identify abnormal development trends before obvious equipment failures occur, and issue early warnings, realizing the transition from post-identification to pre-warning, and significantly enhancing fault prevention capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of the fault monitoring method using AI intelligent image recognition in the present invention; Figure 2 This is a block diagram of the fault monitoring system using AI intelligent image recognition in the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] The present invention provides Figure 1The fault monitoring method shown here uses AI-powered intelligent image recognition. This method can identify potential abnormal trends and issue warnings in advance based on dynamic features and historical correlation analysis in image data before a significant equipment failure occurs, effectively improving equipment operation safety and the timeliness of maintenance responses. The details are as follows:

[0019] In this embodiment, a fault monitoring method using AI intelligent image recognition includes the following steps: Step 1: Collect image data of the spatiotemporal information of the equipment's operating status; specifically including: Multi-angle imaging devices are placed in the equipment area to obtain continuous frame image sequences from different perspectives; this operation ensures that the spatial displacement and morphological changes of the equipment during operation are fully captured from multiple directions.

[0020] Perform pixel-level time alignment on the image sequence to eliminate the time deviation caused by shooting frequency or transmission delay, so that each frame image is consistent in the time dimension. Use the difference method to extract the displacement change between adjacent frames, and the expression is ; Represents the image intensity value at the coordinate (x, y) at time point t, is the intensity value at the corresponding position in the next frame, and D(x,y) represents the degree of change of that pixel over time. This formula can be used to quantify the severity of changes in a local area and identify abnormal movement or deformation of the device surface.

[0021] The dynamic area is divided based on the displacement change, and the area where the change amplitude exceeds the reference value δ is marked as the focus area; the reference value δ can be set according to the maximum allowable fluctuation range during normal operation of the equipment. If it exceeds this value, it is considered that there is dynamic behavior worthy of attention.

[0022] Overlaying infrared thermal imaging data within the area of ​​interest generates a fused image that includes both temperature distribution and deformation information. This not only preserves the spatial details of the visible light image but also incorporates temperature variation, a key physical parameter, enabling earlier detection of potential thermal anomalies or structural fatigue issues within the equipment.

[0023] Step 2: Extracting dynamic features of the image data, including motion trajectory, deformation rate, and spectral change; specifically, including: Select several points of interest in the fused image. These points of interest are usually located in the key structural parts of the equipment or in the areas with obvious changes in the area of ​​interest. Then, the motion vector of each point of interest in the continuous frames is calculated based on the optical flow method. ,in Indicates the coordinate position of a point of interest in the previous frame image, The coordinate position of the interest point in the next frame of the image is represented by . This formula can be used to quantify the motion intensity of the interest point in the image space, thereby reflecting the displacement of the equipment components in actual operation.

[0024] Use motion vectors to determine the trajectory of each point of interest , i=1,2,...,n, where n is the total number of interest points; the trajectory line is formed by connecting the positions of corresponding interest points in consecutive frames, which can intuitively show the movement path and direction of the local area of ​​the device and help identify abnormal movement patterns.

[0025] Analyze the trajectory to obtain the deformation rate ,in, and where represents the trajectory length at the i-th and i+1-th time points, respectively, and Δt represents the time interval between adjacent acquisition moments. This formula measures the rate of change of trajectory length over time, thereby reflecting the deformation trend of the equipment structure during operation. It is an important basis for determining potential structural fatigue or loosening.

[0026] For the infrared thermal imaging data superimposed in the area of ​​interest, the spectrum change S is detected using the spectrum analysis method, and the difference in average brightness value in different time periods is calculated. ,in, Indicates the average brightness value of the selected area in the current time period. represents the average brightness value over the previous time period, and S represents the absolute difference between the two. This formula is used to measure the degree of change in temperature distribution in infrared images and can effectively detect phenomena such as localized temperature rise or abnormal heat dissipation in equipment.

[0027] Step 3: Construct a dynamic association matrix based on the mapping relationship between the dynamic features and the historical status of the device; specifically, the following steps are included: Collect deformation rate and spectral change S, and record the equipment operating parameters at the corresponding time t (such as load, speed, etc.), forming basic data ; This data combines the image analysis results with the actual working conditions to provide multi-dimensional information support for subsequent trend modeling.

[0028] Based on basic data , calculate the similarity score C between each data point and the historical state data point, defined as ,in, Represents a combination of data features under a specific historical state. This formula uses interpolation and normalization to vary the similarity score between 0 and 1. Larger values ​​indicate closer states, enabling quantitative comparison of device operating trends.

[0029] Use the similarity score C to match the current dynamic features with the historical state to generate a matching matrix M, where , i and j represent the similarity scores at different time points. This matrix reflects the correlation between states at different moments and is the basis for establishing time series dependencies; Apply normalization to the matching matrix M so that the sum of all elements in the matrix is ​​equal to 1, and obtain the final dynamic association matrix A, that is, ,This matrix not only reflects the similarity between the current state and the historical state, ,but also establishes the dynamic connection between different time points, ,providing a structured basis for the subsequent identification of potential ,trend deviations.

[0030] Step 4: Analyze the time series of the dynamic correlation matrix to identify potential trend deviation points; specifically including: Arrange the dynamic correlation matrix A in chronological order to generate a time series set , where n represents the total number of time points collected during the entire monitoring process. This operation transforms the originally static matrix data into a dynamic sequence with time dependence, providing a structured basis for subsequent trend analysis.

[0031] Calculate the average correlation strength within the sliding window based on the time series set T , the expression is , where m represents the window length, that is, the number of time points included in each calculation; to Represents the dynamic correlation matrix value corresponding to each time point within the window. This formula is used to measure the overall similarity between the device status and the historical status over a period of time, and can effectively reflect the stability or volatility of the operating trend; Set the current window With the previous window If the difference between the two exceeds the set benchmark ε (i.e. ), the system status is considered to have changed significantly, and the window is marked as an abnormal fluctuation range. The setting of the benchmark ε can be adjusted according to the trend fluctuation range during normal equipment operation, ensuring that anomalies can be sensitively detected while avoiding false alarms.

[0032] Locate the earliest time point of significant change within the abnormal fluctuation range as the potential trend deviation point This time point indicates that the equipment's operating status begins to deviate from the normal trajectory, which may be an important early warning signal before a failure occurs.

[0033] Step 5: Calculate a prediction window based on the trend deviation point. The prediction window is generated by comparing the difference between the current state and a preset threshold. Specifically, the following steps are included: Get potential trend deviation points , and extract the state change rate E within the set range before and after the time point, the expression is ,in, Indicates the state value at the current moment (such as deformation rate or spectral change intensity), represents the state value corresponding to the previous moment, and Δt is the time interval between the two acquisition moments. This formula is used to measure the rate of change of the device state over time and reflects the dynamic stability during operation.

[0034] The state change rate E is compared with a preset threshold θ. If E > θ, the current state change rate exceeds the normal fluctuation range, indicating that the device is experiencing an accelerating change trend. This judgment mechanism helps identify rapidly deteriorating conditions that may cause failures and avoids missing critical warning signals.

[0035] Determine the starting point of the forecast window based on the duration of the acceleration trend and end point , forming the prediction window interval This window covers the critical time period from when the device status begins to change significantly to when it may develop into a fault, providing a time basis for subsequent refined monitoring and response.

[0036] Initiate high-frequency image acquisition and analysis within the forecast window to increase data acquisition density per unit time, thereby enhancing the ability to perceive and respond to abnormal development trends.

[0037] Step 6: Combine the prediction window with the equipment operating environment parameters to generate abnormal pattern matching rules; specifically including: Collect multi-dimensional parameters of the equipment's environment within the prediction window, including temperature T, humidity H, and vibration intensity V. These parameters reflect the equipment's external operating conditions within a specific time period and help identify the impact of environmental factors on equipment status changes.

[0038] Classify and label the image features in the prediction window based on the environmental parameters to form a state sequence with environmental labels , where each A description of the state at a specific point in time; it includes dynamic features extracted from image data (such as deformation rate and spectral changes) and their corresponding environmental parameter combinations. This sequence combines image analysis results with the operating environment, providing a foundation for establishing an environmental relevance judgment mechanism.

[0039] Count the recurring patterns of state sequences under the same environmental conditions and define similarity indicators ,in, and Respectively represent the state description values ​​at two time points, represents the sum of environmental parameters at the i-th time point. This formula measures the consistency of states between different time points through normalization. Smaller values ​​indicate more similar states, and can be used to identify whether similar behavior patterns occur in the same environment.

[0040] The state combination that appears frequently and meets the similarity threshold γ is summarized as an abnormal pattern, and the corresponding matching rules are generated. , which is used for subsequent real-time status comparison and identification. These rules can be used for status comparison in subsequent real-time monitoring. Once the newly collected data matches a rule, the early warning mechanism can be triggered.

[0041] Step 7: Adjusting the response threshold of the abnormal pattern matching rule to adapt to the device performance degradation characteristics; specifically including the following steps: Record the state change curve of the equipment during multiple operation cycles and extract the basic response value at the end of each cycle This value represents the device's ability to respond to input conditions (such as load, temperature, etc.) at the end of the cycle and is used to reflect the performance trend over time.

[0042] Based on the basic response value Calculating the long-term decay factor , the expression is ,in, Indicates the response value of the initial operation cycle of the device. represents the response value at the end of the nth operating cycle, where n represents the total number of operating cycles. This formula quantifies the degree of degradation in overall device performance by measuring the decrease in the response value over time. Larger values ​​indicate more significant performance degradation.

[0043] The attenuation factor Introduced the threshold adjustment of the current matching rules and updated the similarity judgment benchmark , where k is the influence weight coefficient, which is used to adjust the influence of the attenuation factor on the threshold adjustment. In this way, the system can dynamically adjust the sensitivity of anomaly detection based on the degree of device aging or performance degradation, avoiding false positives or false negatives caused by natural device degradation.

[0044] The updated benchmark γ' is used to replace the original threshold to dynamically evaluate the matching degree between the new acquisition state and the abnormal pattern.

[0045] Step 8: Generate an early warning signal through cross-validation of the abnormal pattern matching rule and the prediction window; specifically including: In the current prediction window, the newly collected device status data is compared with the high-frequency status combination in the abnormal pattern matching rules summarized in the early stage in real time, and the number of matches M between the two is counted. The matching density is calculated based on the number of matches M. ,in Indicates the length of the prediction window; this formula is used to measure the consistency between the device state and the abnormal pattern within a unit of time. A higher value indicates that the current state is closer to the known abnormal evolution path.

[0046] Matching density Compared with the updated benchmark γ', if , the current device operating status is determined to be deviating from the normal trajectory and is confirmed to have an abnormal development trend. γ' is a threshold that is dynamically adjusted based on the device's performance degradation. This ensures that the judgment criteria adapts to device aging, avoiding misjudgments and missed judgments.

[0047] Trigger graded warning signals based on abnormal development trends The strength of the warning signal is linearly related to the magnitude of the deviation from the baseline γ'; that is, the greater the deviation, the higher the warning level. This mechanism implements closed-loop control from trend identification to risk response, enabling the system to output corresponding warning levels based on the severity of risks, facilitating timely action by operations and maintenance personnel.

[0048] On the other hand, the present invention proposes a fault monitoring system using AI intelligent image recognition, such as Figure 2 Shown, including: A dynamic feature extraction module is used to collect image data of the spatiotemporal information of the operating status of the device and extract dynamic features of the image data, wherein the dynamic features include motion trajectory, deformation rate and spectral change; A trend identification module is used to construct a dynamic correlation matrix based on the mapping relationship between the dynamic features and the historical status of the equipment, analyze the time series of the dynamic correlation matrix, and identify potential trend deviation points; a matching rule generation module, configured to calculate a prediction window based on the trend deviation point, the prediction window being generated by comparing the difference between the current state and a preset threshold, and combining the prediction window with equipment operating environment parameters to generate an abnormal pattern matching rule; The early warning signal generation module is used to adjust the response threshold of the abnormal pattern matching rule to adapt to the equipment performance attenuation characteristics, and generate an early warning signal through cross-validation of the abnormal pattern matching rule and the prediction window.

[0049] In addition, when executed, each of the above modules is also used to implement other steps of the fault monitoring method using AI intelligent image recognition as described above, as follows: Consider a large rotating device (such as a fan) used at an industrial site. Its core components include bearings, rotors, and casings. After long-term operation, this device may develop structural loosening, localized temperature rise, or deformation. Failure to detect these problems in a timely manner can lead to serious failures. To improve safety and maintenance efficiency, the AI ​​intelligent image recognition fault monitoring method described in this paper is deployed.

[0050] Step 1: Collect spatiotemporal information image data of the equipment's operating status Multiple high-definition cameras and infrared thermal imagers are placed at key locations of the fan (such as the bearing seat and shaft connection) to continuously capture visible light images and temperature distribution images during equipment operation.

[0051] Assume that 10 frames of image are collected per second; after pixel-level time alignment of the image sequence, the displacement change between adjacent frames is extracted using the difference method: .

[0052] Assume that in a certain area, the image intensity value of the tth frame is 120 and that of the t+1th frame is 135, then: D(x,y) = |135-120| = 15. Set the reference value δ = 10. Since 15>10, the area is marked as the area of ​​interest.

[0053] Infrared images were superimposed in the area of ​​interest to generate a fused image, which showed that the temperature had risen from the normal 45°C to 68°C, and it was preliminarily determined that an abnormality existed.

[0054] Step 2: Extract dynamic features of image data Select 5 points of interest in the area of ​​interest (such as the edge points of the bearing outer ring) and use the optical flow method to calculate the motion vector of each point of interest: .

[0055] Assume that the coordinates of a point of interest in the two frames are (100,120) and (105,125), respectively. Then: .

[0056] Record the trajectory line Li of each point of interest and calculate the deformation rate: ; Assume that the calculation =10mm, =12mm, Δt=1s, then: =(12-10) / 1=2mm / s.

[0057] Simultaneously analyze the brightness changes of infrared images: ; Assume that the average brightness of the current time period =210, previous period =190, then: S=|210-190|=20.

[0058] Step 3: Construct a dynamic correlation matrix Collect deformation rate, spectral changes and operating parameters (such as load and speed) in historical operating cycles to form basic data pairs: ;For example: =(2mm / s,20,rotation speed 1500rpm); Calculate the similarity score between the current state and the historical state: ; If the historical status =(1.5mm / s,15,rotation speed 1500rpm), then =(0.5,5,0) → the sum of the differences is 5.5: C=1 / (1+5.5)=0.154.

[0059] Generate the matching matrix M, and then normalize it to get the dynamic association matrix A: ; Assumptions =100, =15, then =15 / 100=0.15.

[0060] Step 4: Identify potential trend deviation points Sort the dynamic correlation matrix by time to form a time series ; The average correlation strength is calculated using a sliding window method: ; Assume that the window length m=5 and the value of A in the window is [0.15, 0.18, 0.16, 0.14, 0.10], then: =(0.15+0.18+0.16+0.14+0.10) / 5=0.146.

[0061] Compare Current Window With the previous window =0.16, set ε=0.02: |0.146-0.16|=0.014<ε, no abnormal fluctuation range is triggered; Continue monitoring until the difference exceeds ε at a certain time, and locate the earliest change point as the potential trend deviation point .

[0062] Step 5: Calculate the prediction window Get State change rate before and after a time point: ; Assumptions =7.07, =5.00, Δt=1s, then: E=(7.07-5.00) / 1=2.07; The threshold θ is set to 1.5. Since 2.07>1.5, it is determined that there is an accelerating trend.

[0063] Determine the prediction window based on duration , for example, the next 30 minutes is the prediction window.

[0064] During this period, high-frequency image acquisition is started, collecting 20 frames per second to improve response sensitivity.

[0065] Step 6: Generate exception pattern matching rules Environmental parameters within the acquisition prediction window: T = 68°C, H = 60%, V = 0.12 mm / s²; Create a state sequence with environment labels , and count the recurring patterns.

[0066] Define the similarity metric: ; Assumptions =120, =110, =68, =60, =0.12: Sim=|120-110| / (68+60+0.12)=10 / 128.12=0.078.

[0067] Set γ=0.1. Since Sim<γ, the states are considered consistent and included in the abnormal pattern library to generate matching rule R_u.

[0068] Step 7: Adjust the response threshold to adapt to performance degradation Record the basic response values ​​of the past 10 operating cycles ,like: =100, =85; Calculate the attenuation factor: =(100-85) / 10=1.5.

[0069] Update similarity benchmark: , set k=0.2, then: γ'=0.1+0.2*1.5=0.4.

[0070] The new benchmark γ' = 0.4 was used to replace the original value for subsequent evaluation.

[0071] Step 8: Generate early warning signals Compare the real-time status with the exception pattern rules within the current prediction window: Number of matches M = 8; window time length L_w = 30 minutes; Calculate the matching density: =8 / 30=0.267.

[0072] Compared with the updated benchmark γ'=0.4, 0.267<0.4, and no warning is triggered.

[0073] If the subsequent Increases to 0.45, then: , confirm abnormal development trends and trigger graded warning signals: , where K is the scaling factor (e.g. K=10) =(0.45-0.4)×10=0.5, corresponding to the "medium level" warning.

[0074] This example demonstrates that the present invention provides a complete fault monitoring method based on AI-powered intelligent image recognition, encompassing the entire process from image acquisition, dynamic feature extraction, trend modeling, anomaly identification, and early warning output. By introducing a mathematical model to quantify changing trends and adaptively adjusting the system based on environmental parameters and equipment aging characteristics, the system possesses highly intelligent fault prediction capabilities, enabling early warnings before significant equipment failures occur, effectively improving operational safety and maintenance efficiency.

[0075] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A fault monitoring method using AI intelligent image recognition, characterized in that: The following steps are involved: Collecting image data of spatiotemporal information of the operating status of the device, and extracting dynamic features of the image data, wherein the dynamic features include motion trajectory, deformation rate, and spectral change; Constructing a dynamic correlation matrix through the mapping relationship between the dynamic features and the historical status of the equipment, analyzing the time series of the dynamic correlation matrix, and identifying potential trend deviation points; Calculating a prediction window based on the trend deviation point, the prediction window is generated by comparing the difference between the current state and a preset threshold, and combining the prediction window with the device operating environment parameters to generate an abnormal pattern matching rule; The response threshold of the abnormal pattern matching rule is adjusted to adapt to the device performance attenuation characteristics, and an early warning signal is generated through cross-validation of the abnormal pattern matching rule and the prediction window.

2. A fault monitoring method using AI intelligent image recognition according to claim 1, characterized in that: Image data that collects spatiotemporal information about the equipment's operating status, including: Arrange multi-angle imaging devices in the equipment area to obtain continuous frame image sequences at different viewing angles; Performing pixel-level time alignment on the image sequence to extract displacement changes between adjacent frames; Divide the dynamic area based on the displacement change, and mark the area where the change amplitude exceeds the set reference as the area of ​​interest; Infrared thermal imaging data is superimposed in the area of ​​interest to generate a fused image containing temperature distribution and deformation information.

3. The fault monitoring method using AI intelligent image recognition according to claim 2, characterized in that: Extracting dynamic features of the image data, including: Selecting a number of interest points in the fused image, analyzing the position offset of each interest point in consecutive frames, and calculating a motion vector; Determine the motion trajectory of each point of interest using the motion vector, form a trajectory line by connecting the positions in consecutive frames, and analyze the trajectory line to obtain the deformation rate; Spectral changes are detected for the infrared thermal imaging data superimposed within the area of ​​interest to identify temperature fluctuation trends.

4. The fault monitoring method using AI intelligent image recognition according to claim 3 is characterized in that: A dynamic association matrix is ​​constructed by mapping the dynamic features to the historical status of the device, including: Collecting data on the deformation rate and spectral change, and recording the equipment operating parameters at the corresponding time, to form a basic data pair; Calculating a similarity score between each data point and the historical state based on the basic data pair, and matching the current dynamic feature with the historical state using the similarity score to generate a matching matrix; The matching matrix is ​​normalized to obtain a final dynamic correlation matrix.

5. The fault monitoring method using AI intelligent image recognition according to claim 1, characterized in that: Analyze the time series of the dynamic correlation matrix to identify potential trend deviation points, including: Arranging the dynamic correlation matrix in chronological order to generate a time series set, and calculating an average correlation strength within a sliding window based on the time series set; Compare the average correlation strength of the current window with that of the previous window. If the difference exceeds the set range, mark the window as an abnormal fluctuation range. The earliest time point at which a significant change occurs within the abnormal fluctuation range is located as a potential trend deviation point.

6. The fault monitoring method using AI intelligent image recognition according to claim 1, characterized in that: Calculating a prediction window according to the trend deviation point includes: Obtain the potential trend deviation point and extract the state change rate within a set range before and after the time point; Comparing the state change rate with a preset threshold to determine whether there is an accelerating change trend; The starting point and the ending point of the prediction window are determined based on the duration of the accelerating change trend to form a prediction window interval.

7. The fault monitoring method using AI intelligent image recognition according to claim 1, characterized in that: Combining the prediction window with the equipment operating environment parameters, anomaly pattern matching rules are generated, including: Collecting multi-dimensional parameters of the environment in which the device is located within the prediction window, including temperature, humidity, and vibration intensity; Classify and label the image features in the prediction window based on the environmental parameters to form a state sequence with environmental labels; Count the recurring patterns of state sequences under the same environmental conditions and define similarity indicators to measure the consistency of states between different time points; The state combinations that appear frequently and meet the similarity threshold are summarized as abnormal patterns, and the corresponding matching rules are generated.

8. The fault monitoring method using AI intelligent image recognition according to claim 1, characterized in that: Adjusting the response threshold of the abnormal pattern matching rule includes the following steps: Record the state change curve of the equipment during multiple operation cycles and extract the basic response level at the end of each cycle; Calculating a long-term attenuation factor based on the base response level to quantify the trend of device performance degradation over time; Introducing the attenuation factor into the threshold adjustment of the current matching rule to update the similarity judgment benchmark; The updated benchmark is used to replace the original threshold and dynamically evaluate the degree of match between the new acquisition state and the abnormal pattern.

9. The fault monitoring method using AI intelligent image recognition according to claim 1, characterized in that: Generate an early warning signal through cross-validation of the abnormal pattern matching rule and the prediction window, including: Compare the acquisition state with the high-frequency state combination in the abnormal pattern matching rule in real time within the current prediction window, and count the number of matches; Calculating a matching density based on the number of matches, comparing the matching density with an updated benchmark, and confirming the presence of an abnormal development trend if the matching density exceeds the benchmark; A graded warning signal is triggered according to the abnormal development trend, and the signal strength increases with the matching difference to achieve warning output.

10. A fault monitoring system using AI intelligent image recognition for implementing the method according to any one of claims 1 to 9, characterized in that: include: A dynamic feature extraction module is used to collect image data of the spatiotemporal information of the operating status of the device and extract dynamic features of the image data, wherein the dynamic features include motion trajectory, deformation rate and spectral change; A trend identification module is used to construct a dynamic correlation matrix based on the mapping relationship between the dynamic features and the historical status of the equipment, analyze the time series of the dynamic correlation matrix, and identify potential trend deviation points; a matching rule generation module, configured to calculate a prediction window based on the trend deviation point, the prediction window being generated by comparing the difference between the current state and a preset threshold, and combining the prediction window with equipment operating environment parameters to generate an abnormal pattern matching rule; The early warning signal generation module is used to adjust the response threshold of the abnormal pattern matching rule to adapt to the equipment performance attenuation characteristics, and generate an early warning signal through cross-validation of the abnormal pattern matching rule and the prediction window.

Images