A multi-source pressure equipment safety monitoring system based on image fusion analysis

Abstract

This invention relates to the field of industrial equipment safety monitoring technology, and discloses a multi-source pressure equipment safety monitoring system based on image fusion analysis. The system includes: acquiring baseline and real-time image sequences during the equipment's operating cycle through an image acquisition module; dividing the field of view into multiple blocks and extracting the average brightness time series of each block; calculating the mismatch between the real-time sequence and the baseline sequence and generating a spatial distribution map; and combining temperature data to achieve anomaly warning. This invention establishes an adaptive dynamic visual baseline for the equipment, enabling the perception of local temporal pattern mismatches that are difficult to quantify in traditional monitoring, and identifying early anomalies before physical deformation manifests. Simultaneously, utilizing a logical arbitration mechanism between visual and thermodynamic data, a high-confidence warning is triggered only when the two modal anomalies reach spatial consensus, effectively solving the problem of false alarms in complex industrial environments.

Description

Technical Field

[0001] This invention relates to a multi-source pressure equipment safety monitoring system based on image fusion analysis, belonging to the field of industrial equipment safety monitoring technology. Background Technology

[0002] Currently, the field generally adopts physical quantitative monitoring methods based on image feature extraction. This involves using high-precision image sensors to capture geometric deformation or texture changes on the surface of equipment and converting them into physical parameters such as micron-level displacement or crack length for threshold alarms. However, when deployed in scenarios such as chemical storage and transportation and energy pipelines, this type of method often faces fundamental limitations: its technical logic relies on the identification of explicit physical defects, while the stress concentration in the early stages of equipment failure often only manifests as statistical drift of the dynamic response mode of surface pixel clusters. Such weak signals are difficult to effectively separate under the noise of complex industrial environments. Multi-source data fusion schemes introduced to improve the signal-to-noise ratio, such as infrared thermometry and visual superposition, often use mathematical weighted averaging or feature-level stitching. Under extreme conditions such as arc flashes and dust interference, they are prone to misjudgment due to spatiotemporal inaccuracies in the data.

[0003] It is worth noting that as the demand for early warning of hidden risks in industrial sites increases, the existing technology system has exposed deeper contradictions: when trying to capture early pattern anomalies by increasing image resolution or algorithm complexity, the system loses its real-time performance due to the surge in computational load; while simply increasing the accuracy of infrared sensors can locate temperature rise areas, it cannot explain false anomalies where visual dynamics are unstable but the temperature distribution is normal. Although the acoustic vibration-assisted solution explored by the industry in recent years can trigger event monitoring, the physical correlation model between its vibration signal and visual dynamics has not yet been established, resulting in a break in the decision-making chain.

[0004] Specifically, existing technologies suffer from the following bottlenecks: 1. They lack the ability to model the dynamic steady state formed by pixel clusters on the device surface during pressure cycling, making it impossible to quantify early failure symptoms such as statistical stability drift; 2. The logical coupling of visual, thermodynamic, and acoustic vibration data is limited to data superposition, without constructing a cross-modal mutual verification mechanism based on physical common sense; 3. Common interferences such as sudden changes in global illumination directly contaminate the visual analysis results, and the system lacks continuity of decision-making guarantees in the event of instantaneous sensor failure. Therefore, how to achieve online evaluation of the dynamic mode stability of the device surface under strong noise environments and establish a high-confidence early warning mechanism through multimodal logical arbitration has become the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a multi-source pressure equipment safety monitoring system based on image fusion analysis. Its main purpose is to solve the problem that current pressure equipment safety monitoring technology cannot provide high-confidence early warning of hidden risks in high-noise industrial environments.

[0006] To achieve the above objectives, the present invention provides a multi-source pressure equipment safety monitoring system based on image fusion analysis, characterized in that it includes:

[0007] The image acquisition module is configured to acquire a baseline image sequence of the pressure equipment during a complete working cycle of the pressure equipment; and to acquire a real-time image sequence of the pressure equipment during continuous monitoring.

[0008] The baseline analysis module is configured to divide the field of view of the baseline image sequence into multiple image blocks, and for each image block, extract the baseline time series of the mean pixel brightness over time.

[0009] The real-time analysis module is configured to divide the field of view of the real-time image sequence into multiple image blocks in the same way as the baseline image sequence, and extract the real-time time series of the mean pixel brightness change over time for each image block.

[0010] The mismatch calculation module is configured to calculate a mismatch value for each image block based on the time series similarity between the real-time time series and the baseline time series, using the following method. , in, This indicates the real-time time series at time 10:00. The average pixel brightness Indicates the baseline time series at time 10:00. The average pixel brightness This represents the average value of a real-time time series. This represents the average value of the baseline time series. It represents the number of sampling points in the time series; and generates a spatial distribution map of mismatch based on the mismatch values ​​of all image blocks;

[0011] The temperature acquisition module is configured to acquire low-resolution temperature distribution data from the pressure equipment.

[0012] The early warning judgment module is configured to output a high-confidence early warning message when an abnormal region appears in the mismatch spatial distribution map that is spatially continuous and temporally persistent for a duration equal to or exceeding a preset duration, and the abnormal region spatially overlaps with a temperature rise region in the temperature distribution data that is persistent and has a temperature higher than its neighboring regions. The early warning judgment module is also configured to determine that the system has entered an optical interference state and temporarily suspend the early warning logic based on the mismatch spatial distribution map when the image acquisition module detects that the difference between the global brightness statistical features of the real-time image sequence and the global brightness statistical features of the baseline image sequence exceeds a first preset brightness difference threshold, and the temperature distribution data acquired by the temperature acquisition module at the same time does not change by more than a second preset temperature difference threshold compared with the temperature distribution data at the previous time or under normal working conditions.

[0013] Preferably, the system further includes: an acoustic vibration acquisition module, configured to continuously monitor the acoustic vibration signal intensity of the pressure equipment; when the acoustic vibration signal intensity reaches or exceeds a preset acoustic vibration intensity threshold, a wake-up command is sent to the image acquisition module and the mismatch calculation module to enable the system to enter the working mode from the sleep mode; and when the acoustic vibration signal intensity is lower than the preset acoustic vibration intensity threshold for a preset duration, the system automatically returns to the sleep mode.

[0014] Preferably, the mismatch calculation module is configured to use a dynamic time warping algorithm or a cross-correlation function to calculate the similarity of time series.

[0015] Preferably, the baseline analysis module is configured to periodically and automatically update the baseline time series to adapt to long-term, slow changes on the surface of the pressure equipment.

[0016] Preferably, the early warning judgment module is further configured to perform secondary thermodynamic arbitration on the abnormal area on the mismatch spatial distribution map before outputting high confidence early warning information, by judging whether its temperature continues to rise within a preset time period and meets the judgment conditions of preset temperature rise rate or preset temperature rise amount. Only when it is confirmed that the judgment condition of continuous temperature rise is met will the high confidence early warning information be output.

[0017] Preferably, the image acquisition module is configured as an industrial camera to record changes in pixel brightness, and the temperature acquisition module is configured as an infrared array sensor to provide logical confidence weighting for visual anomalies.

[0018] Preferably, the mismatch calculation module is configured as an edge computing unit for one-dimensional time series comparison.

[0019] Preferably, the early warning judgment module is configured to pause the early warning logic based on the mismatch degree spatial distribution map when the system is determined to enter the optical interference state, and activate a temporary acoustic vibration-thermal correlation model. Under this model, when the acoustic vibration acquisition module detects that the amplitude of the acoustic vibration signal intensity fluctuation exceeds the third preset acoustic vibration fluctuation threshold, and the temperature acquisition module detects that the temperature distribution change exceeds the fourth preset temperature anomaly threshold, an alarm is issued.

[0020] Preferably, the system further includes: a thermal excitation module configured to periodically control a thermal excitation source to apply a standardized thermal pulse to a predetermined area of ​​the pressure equipment; and a temperature acquisition module further configured to monitor the temperature decay characteristics of the predetermined area after the thermal pulse is applied, so as to infer the material fatigue state of the area; and an early warning judgment module further configured to weight the risk level in the mismatch spatial distribution map according to the material fatigue state.

[0021] Preferably, the temperature acquisition module characterizes the thermal diffusion rate of the material by analyzing the thermal decay time constant of the temperature decay characteristic curve; wherein, the extension of the thermal decay time constant indicates the increased fatigue of the material, and the fatigue state of the material is quantified by comparing it with a preset healthy baseline time constant.

[0022] Compared with the prior art, the beneficial effects of the present invention are:

[0023] 1. The system captures the temporal changes in pixel brightness during the complete working cycle of the pressure equipment to form a breathing pattern baseline that characterizes the health status of the equipment. This unsupervised learning mechanism, built on the dynamic characteristics of the equipment itself, enables the system to perceive minute mismatches in the temporal patterns of local areas. It transforms the dynamic stability drift, which is difficult to quantify in traditional monitoring, into a calculable spatial distribution of mismatch, thereby identifying early abnormalities before physical deformation characteristics appear. When visual dynamic analysis detects a spatiotemporally continuous abnormal area, the system does not directly trigger an alarm. Instead, it requires that the area must spatially coincide with the area of ​​continuous temperature rise in the infrared temperature field. This multimodal mutual verification mechanism based on physical common sense ensures that visual artifacts caused by illumination interference or local environmental temperature changes cannot trigger false alarms on their own. Only when visual and thermodynamic anomalies reach a logical consensus will the system output a high-confidence warning, thus solving the problem of false alarms in complex industrial scenarios.

[0024] 2. The acoustic and vibration sensors act as ultra-low-power state sentinels, waking up the high-energy-consuming visual analysis module only when changes in equipment operating conditions are detected. This event-driven, on-demand monitoring mechanism avoids invalid data processing under static conditions and concentrates core computing power on the window period when the equipment state is most likely to evolve. While reducing energy consumption by more than 95%, it also improves the signal-to-noise ratio of anomaly identification by focusing on critical periods. When the global brightness changes abruptly while the temperature field remains stable, the system automatically determines it as optical transient pollution and temporarily suspends the visual analysis logic. At this time, by activating the acoustic-vibration-thermal correlation model, basic monitoring capabilities can still be maintained during the brief failure of the camera. This self-diagnostic mechanism based on the logical proof of physical quantity changes makes the system resilient to strong interference such as electric arc flashes, ensuring the continuity of monitoring logic.

[0025] 3. The system periodically applies standardized thermal pulses to key areas, and inverts the material's thermal diffusion rate by analyzing its temperature decay characteristics. The obtained fatigue state data is then incorporated into the visual mismatch analysis as a weighting factor, which significantly increases the risk level of the same visual anomaly in the material fatigue area. This closed-loop coupling of active detection and passive monitoring enables the assessment of hidden risks that are close to failure without visual signs. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the architecture of a multi-source pressure equipment safety monitoring system based on image fusion analysis according to the present invention;

[0027] Figure 2 This is a comparison of temperature decay characteristic curves of different materials under fatigue states according to the present invention;

[0028] Figure 3 This is a flowchart of the sleep and wake-up process driven by the acoustic vibration signal of the system of the present invention.

[0029] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0030] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0031] The present invention discloses a multi-source pressure equipment safety monitoring system based on image fusion analysis. Its overall architecture includes an image acquisition module, a baseline analysis module, a real-time analysis module, a mismatch calculation module, a temperature acquisition module, and an early warning judgment module. Through the collaboration of auxiliary units such as an acoustic vibration acquisition module and a thermal excitation module, it forms a closed-loop monitoring system with adaptive and self-diagnostic capabilities. Its core operation process begins with establishing a dynamic visual baseline of the equipment's health status, and then achieves high-confidence early warning of early anomalies through real-time comparison and multimodal logic arbitration. At the same time, through event-driven and active detection mechanisms, it takes into account both system energy efficiency and the depth of risk prediction.

[0032] In a typical application scenario, such as safety monitoring of a chemical reactor, the specific deployment and operation process of the system is described as follows: During the initialization phase after system deployment, the core task is to establish a dynamic visual baseline characterizing the healthy operating status of the equipment. Given that pressure equipment exhibits periodic optical responses on its surface due to internal pressure fluctuations during operation, which are imperceptible to the human eye but have a fixed pattern, accurately capturing this breathing pattern becomes crucial for identifying early, subtle anomalies. Therefore, once the reactor is confirmed to be in a complete and fault-free standard operating cycle, the operator will initiate the baseline acquisition program. The image acquisition module within the system, specifically configured as a high-frame-rate industrial camera, is used to continuously record images of key monitoring areas of the reactor, forming a baseline image sequence containing thousands of frames. Next, the baseline analysis module virtually divides the camera's field of view into a fine grid, such as a 128×128 matrix, thus obtaining multiple independent image blocks. For each block, the module calculates the average pixel brightness of each frame in the baseline image sequence, thereby generating a unique baseline time series for each block representing its brightness variation over time under healthy operating conditions. To address the long-term, slow changes on the equipment surface caused by coating aging or trace deposits, the baseline analysis module is also configured to periodically and automatically update this baseline time series. For example, after the equipment has run continuously without failure for a certain number of work cycles, a new baseline sequence is automatically recorded to ensure the timeliness and adaptability of the benchmark. Once the continuous monitoring phase begins, the system executes the core logic of real-time analysis and anomaly identification. The image acquisition module captures real-time image sequences of the pressure equipment at the same frame rate as the baseline acquisition, while the real-time analysis module uses the same grid division method to extract the real-time time series of the average pixel brightness over time for each image block. To achieve low-latency online computation, the system preferably employs an edge computing unit deployed on-site as the mismatch calculation module. This module, for each image block, extracts the mismatch degree in real time. With stored baseline time series The two are compared and their similarity is quantified using a normalized cross-correlation function, thereby calculating a mismatch value. The specific calculation formula is as follows: ,in, This indicates the real-time time series at time 10:00. The average pixel brightness Indicates the baseline time series at time 10:00. The average pixel brightness and These represent the average values ​​of the corresponding time series. The number of sampling points for the time series; in scenarios where the rhythm of the device's working cycle fluctuates slightly, the mismatch calculation module can also be configured to use a dynamic time warping algorithm to more robustly measure the morphological similarity of two time series, unaffected by slight time axis scaling; mismatch values ​​for all image blocks. The data is integrated in real time to generate an intuitive spatial distribution map of the mismatch, which dynamically shows the degree of deviation of each area on the device surface from its healthy baseline pattern in the form of a color spectrum or grayscale.

[0033] To address the technical challenge of false alarms caused by interference in visual analysis within complex industrial environments, the system introduces a multimodal logic arbitration mechanism based on physical common sense. A temperature acquisition module, specifically a low-resolution infrared array sensor, is integrated with the industrial camera to synchronously collect temperature distribution data from the equipment. The core procedure of the early warning judgment module does not simply trigger an alarm for high mismatch areas, but rather executes a strict cross-modal mutual verification logic: a mismatch value is determined only if an abnormal area appears in the mismatch spatial distribution map that is spatially continuous and temporally persistent (e.g., lasting for more than a preset duration of 5 seconds). The system will only determine a high-confidence anomaly if the temperature consistently exceeds a preset mismatch threshold, such as 0.7 (an engineering setting that balances detection sensitivity and false alarm rate). Furthermore, the anomaly must spatially coincide with a persistently existing temperature-rising region in the temperature distribution data, where the temperature is higher than its neighboring regions (e.g., a temperature difference greater than 5°C). To further improve warning accuracy, before outputting a high-confidence warning, the system can initiate a secondary thermodynamic arbitration. This involves analyzing the temperature change trend of the spatiotemporally overlapping anomaly region over a preset time period. Only when the temperature in this region meets a preset continuous temperature rise condition, such as a continuous temperature rise rate greater than 0.2°C / second, will the warning judgment module finally output an alarm message. This design utilizes the phenomenon that physical failure processes are usually accompanied by energy dissipation and localized temperature rise, providing a basis for visual anomalies. Physical confidence weighting is often provided. In the balance design between system energy efficiency and monitoring real-time performance, this embodiment introduces an event monitoring mode driven by acoustic vibration signals. The system has a built-in acoustic vibration acquisition module, which acts as the sole sentinel unit in the standby or static state of the equipment, continuously monitoring the intensity of acoustic vibration signals at the scene with extremely low power consumption. When the pressure equipment starts running, if the intensity of the acoustic vibration generated exceeds a preset acoustic vibration intensity threshold obtained by calibration at the moment of equipment startup, the acoustic vibration acquisition module will immediately send a wake-up command to the image acquisition module and the mismatch calculation module, so that the entire high-power visual analysis system can quickly enter the working mode from the sleep mode. Correspondingly, when the equipment stops running, if the intensity of the acoustic vibration signal is lower than the threshold and continues for a preset period of time, the system will automatically return to the sleep mode. This on-demand working mechanism reduces the energy consumption and data processing burden of the system during non-critical periods.

[0034] To ensure the continuity of system decision-making under specific interference conditions, the system also constructs a self-identification and adaptation mechanism for optical interference. When an electric arc flash or a sudden change in ambient light occurs, the global brightness statistics of the real-time image sequence captured by the image acquisition module will momentarily deviate from the corresponding characteristics of the baseline sequence and exceed a first preset brightness difference threshold. At this time, the early warning judgment module will immediately query the temperature data at the same moment. If it finds that the temperature distribution data has not changed by more than a second preset temperature difference threshold compared with the previous moment, the system will determine that it has entered an optical interference state, rather than a physical abnormality of the equipment. In this state, the system temporarily suspends the early warning logic based on the mismatch spatial distribution map and simultaneously activates a temporary acoustic-vibration-thermal correlation model as a backup monitoring channel. Under this model, the system will only issue an alarm when the acoustic acquisition module detects severe acoustic signal fluctuations and the temperature acquisition module also detects significant temperature distribution changes. This design ensures that even during moments of temporary visual sensor malfunction, the system can still maintain basic safety monitoring capabilities through correlation analysis of other modalities. Finally, to achieve a combination of passive monitoring and proactive early warning, this system also integrates an optional thermal excitation module and related components. The corresponding material fatigue state analysis function utilizes a thermal excitation module, such as a controllable miniature thermal radiation source, configured to periodically apply a standardized thermal pulse to the critical pressure-bearing area of ​​the equipment. The temperature acquisition module then precisely monitors the temperature decay characteristic curve of this area after the thermal pulse. Since fatigue damage within the material hinders heat conduction, slowing the heat diffusion rate, the fatigue state of the material can be inferred by analyzing this temperature decay characteristic curve and calculating its thermal decay time constant. An extended thermal decay time constant directly indicates the escalation of material fatigue. By comparing it with a health baseline time constant calibrated at the time of equipment delivery, the system can quantify the degree of accumulated fatigue. This quantified fatigue state data is then used as a dynamic risk weighting factor by the early warning judgment module to adjust the risk level of the mismatch spatial distribution map. For example, for an area assessed as highly fatigued, the mismatch threshold triggering a visual anomaly alarm will be dynamically lowered. This means that a slight visual pattern mismatch that might otherwise be ignored in this area will be judged as a high-risk event because it occurs in a fragile material, thus enabling early warning of potential failure risks.

[0035] Example 1: In a high-pressure hydrogenation reactor that has been operating continuously for many years, the critical pressure-bearing weld area faces the potential risk of sudden failure due to material fatigue and hydrogen embrittlement caused by long-term exposure to harsh conditions of high temperature, high pressure, and hydrogen permeation. The system of this invention is deployed in this scenario. Its operation begins during a planned shutdown maintenance window, where a standardized thermal pulse is applied to the critical weld area by the thermal excitation module. The temperature acquisition module then records and analyzes the subsequent temperature decay characteristics. The calculated thermal decay time constant is compared with the initial health baseline time constant of the equipment, showing a significant extension. Based on this, the system identifies the weld area as a fatigue region with a high probability of failure in its internal model and stores this assessment result as a static risk weighting factor. Within several months after the unit resumes operation, the system's acoustic and vibration acquisition module, based on the reactor... Based on the operating characteristics of the core pump unit, the system automatically wakes up and puts the image acquisition and mismatch calculation modules into sleep mode, avoiding invalid data processing when the equipment is in a quiet state. During this period, a safety valve on an adjacent pipeline momentarily opened, causing a large amount of high-temperature steam to gush out, which severely obscured the monitoring field of view. The global brightness statistics of the real-time image sequence changed drastically, and a large area of ​​high mismatch values ​​appeared instantaneously on the mismatch spatial distribution map. The built-in procedure of the early warning judgment module was immediately triggered. After comparison and confirmation, although the global brightness change exceeded the first preset brightness difference threshold, the temperature distribution data fed back by the temperature acquisition module did not change beyond the second preset temperature difference threshold that corresponds precisely to the visual anomaly in space. Based on this, the system determined that it had entered an optical interference state and suspended the early warning logic based on the mismatch spatial distribution map, without generating an alarm output for this event.

[0036] After the external interference dissipates, the system automatically resumes its normal monitoring logic. At this point, the mismatch calculation module begins to stably detect a weak but persistent mismatch value in several image blocks of the weld that was previously identified as a fatigue area. The unidirectional deviation, though not yet reaching the system's default general warning threshold, was identified as a potential anomaly requiring escalation when the warning judgment module processed data in this area by invoking a specific judgment threshold adjusted by the aforementioned risk weighting factor. Instead of directly outputting an alarm, the system initiated multimodal logic arbitration, guiding the temperature acquisition module to perform focused data analysis on this specific area. In a subsequent work cycle, it confirmed a small temperature rise area that completely overlapped with the visually abnormal area and whose temperature remained consistently higher than its neighboring areas. Thus, the risk zone established by active detection was... The prior knowledge, dynamic behavioral instability captured by passive monitoring, and energy dissipation characteristics verified by thermodynamic data together constitute a judgment criterion. The system ultimately outputs a clear, high-confidence early warning message with historical fatigue state assessment. This implementation process shows that the system shifts the focus of monitoring from isolated judgment of whether physical quantities exceed limits to assessing the logical consistency between the dynamic behavior stability of the equipment and its physical environment prior conditions. By establishing a cross-modal, multi-level decision-making mechanism based on physical correlation, the ability to identify early signs of real faults no longer compromises between the pursuit of sensitivity and the robustness of suppressing environmental interference.

[0037] Example 2: To verify the effectiveness of the collaborative mechanism in this embodiment—which uses active thermal excitation to weight the risk level of subsequent passive visual monitoring—a dedicated test platform was constructed. This platform uses a standard-sized pressure vessel steel plate as its core. A hydraulic loading device applies periodic loads simulating the working cycle of the equipment. A high-frequency mechanical vibration device applies prolonged fatigue loading to a predetermined area of ​​the steel plate, region A. Meanwhile, region B, with identical physical properties, is selected on the other side of the steel plate as a control group representing the healthy state. The platform integrates the features described in the aforementioned specific embodiments. The complete monitoring system includes a thermal excitation module, an image acquisition module, a temperature acquisition module, and an early warning judgment module. The standardized thermal pulse applied by the thermal excitation module is set with the energy value determined by an engineering trade-off that aims to obtain a temperature decay characteristic curve with a sufficient signal-to-noise ratio while ensuring no thermal damage to the tested material. The setting procedure is based on the material properties and thickness of the steel plate to calculate the theoretical energy required to obtain an effective temperature rise, such as 5°C to 10°C. Based on this, the system is modified in conjunction with a heat dissipation model to finally determine an energy value that ensures the accuracy of subsequent thermal decay time constant calculations while being far below the critical value for changes in the metallographic structure of the material.

[0038] The experimental procedure first enters the active detection phase. The thermal excitation module applies identical standardized thermal pulses to regions A and B sequentially, while the temperature acquisition module simultaneously records the temperature decay characteristics of the two regions. By analyzing these characteristic curves, the system calculates the thermal decay time constant, which characterizes the material's thermal diffusion rate. Region A, having undergone fatigue loading, exhibits a significantly longer thermal decay time constant than region B, serving as the control group. This establishes a quantified and significantly different material fatigue state benchmark for the two regions, and this benchmark data is stored in the weighted database of the early warning judgment module. Subsequently, the experiment enters the passive monitoring and fault injection phase. Under a continuously applied background load cycle by the hydraulic device, two identical micro-heat sources apply a continuous thermal disturbance with extremely low power and identical heating characteristics to regions A and B respectively. This simulates a very weak early fault thermal signal generated by microcrack propagation or stress concentration in the real world. During the fault injection phase, the system's real-time analysis module and mismatch calculation module continuously operate. With the introduction of the weak thermal signal, regions A and B both exhibit a slowly increasing mismatch value with almost identical amplitude on the mismatch spatial distribution map. However, the early warning judgment module gave completely different internal risk assessments and final outputs for the same visual anomaly data reported by the two areas. Table 1 is a selection of data records of the key states of the system at this stage.

[0039] Table 1: Comparison of system responses to the same fault signal in different fatigue states.

[0040]

[0041] Referring to Table 1, the difference in system response stems from the built-in procedure of the early warning judgment module: when assessing the risk of mismatch data in region A, the module invokes its pre-stored high fatigue state weight quantified by the thermal decay time constant. This weight factor causes the system's internal judgment logic to upgrade a mismatch value that is normally considered low-risk to a high-risk event requiring attention, and outputs a level-two early warning. In contrast, for region B in a healthy state, the same mismatch value is judged as a normal fluctuation insufficient to trigger any early warning. This experimental data shows that the present invention can use the material physical health status obtained by active detection as prior knowledge to effectively guide the passive monitoring system in interpreting the risk of abnormal signals, thereby achieving early identification of potential faults in high-risk areas without relying on increasing the intensity of the abnormal signals themselves.

[0042] Example 3: This example combines Figures 1 to 3 This paper describes the implementation of a multi-source pressure equipment safety monitoring system based on image fusion analysis, such as... Figure 1As shown in the figure, this diagram illustrates the overall architecture of a multi-source pressure equipment safety monitoring system based on image fusion analysis. It includes a core module and auxiliary modules. The image acquisition module (industrial camera) is responsible for acquiring baseline and real-time image sequences of the pressure equipment (reactors / storage tanks, etc.). The baseline analysis module generates a baseline time series based on the baseline image sequences, while the real-time analysis module generates a real-time time series based on the real-time image sequences. The mismatch calculation module (edge ​​computing unit) uses these two time series to calculate the mismatch. The system generates a spatial distribution map of the mismatch degree. Simultaneously, the temperature acquisition module (infrared array sensor) collects temperature distribution data of the pressure equipment. This data is then aggregated into the early warning judgment module (multimodal logic arbitration). Combined with the wake-up / sleep control signal from the acoustic and vibration acquisition module (vibration sensor) and the material fatigue state information provided by the thermal excitation module (active detection), a comprehensive judgment is made. Under specific circumstances, the system will enter an optical interference state (acoustic-vibration-thermal model). At this time, the early warning logic will be adjusted. Finally, the system outputs high-confidence early warning information based on these analyses, thereby achieving safety monitoring of the pressure equipment.

[0043] like Figure 2 As shown in the figure, this graph illustrates the temperature decay characteristics over time (seconds) after a thermal pulse is applied to different material states. The vertical axis represents temperature (°C), and the horizontal axis represents time (seconds). The legend clearly indicates the three material states: the solid line corresponds to the healthy material, and its thermal decay time constant is... Approximately 3.8 seconds; mild fatigue corresponds to the dashed line, with its thermal decay time constant. Approximately 5.5 seconds; the point corresponding to severe fatigue is indicated by the dashed line, and its thermal decay time constant is... The difference between these curves, approximately 8.2 seconds, indicates that increased material fatigue leads to a longer thermal decay time constant, thus quantitatively characterizing the fatigue state of the material.

[0044] like Figure 3 As shown in the figure, this diagram details the system's workflow of controlling sleep and wake-up via acoustic vibration signals. It involves the interaction between the acoustic vibration acquisition module, image acquisition module, mismatch calculation module, and system controller. In the initial state, the acoustic vibration acquisition module is in a low-power sentry mode and continuously monitors the acoustic vibration signal. When the acoustic vibration intensity is detected, if the intensity exceeds a threshold, the acoustic vibration acquisition module sends a wake-up command to the image acquisition module and the mismatch calculation module, activating their working modes and putting them into a high-power working state, thus initiating visual analysis. Conversely, when the acoustic vibration intensity remains below the threshold for a preset duration, the acoustic vibration acquisition module sends a sleep command, causing the image acquisition module and the mismatch calculation module to enter sleep mode successively, ultimately leading to the system entering an energy-saving sleep state.

[0045] Example 4: On a specific pressure vessel where the monitoring system of this invention is to be deployed, to ensure the accuracy and reliability of subsequent online monitoring, a complete system initialization and parameter calibration procedure needs to be executed. This procedure aims to transform the system's built-in general algorithm model and logical thresholds into a dedicated configuration for the specific equipment structure, materials, and operating characteristics. The initial state definition of this process requires clarifying the key monitoring areas of the pressure vessel, such as high-stress welds, and confirming their material grade and design wall thickness. At the same time, it ensures that the field of view, resolution, and sampling rate of the deployed image acquisition module and temperature acquisition module are sufficient to support the accuracy requirements of subsequent data analysis. The execution of the procedure begins with an offline baseline data acquisition and modeling phase. The system drives the pressure vessel to run multiple standard working cycles under fault-free conditions. During this period, the image acquisition module, temperature acquisition module, and acoustic vibration acquisition module synchronously record a high-fidelity multimodal data stream covering the entire process, forming a comprehensive dynamic baseline database. The analysis program within the system traverses the mismatch values ​​of all image blocks in the baseline database. The time series is used to calculate its overall mean under normal fluctuations. with standard deviation and the basic mismatch threshold Set as Similarly, temperature-related thresholds, such as the temperature difference for determining temperature rise zones, are also determined by analyzing the statistical characteristics of the temperature field between different regions in the baseline data.

[0046] The early warning judgment module's logic for determining abnormal areas is deterministically algorithmized. Spatial continuity is determined using a connected component labeling algorithm, meaning that only when the mismatch value within the field of view is... Image blocks exceeding their corresponding thresholds are considered candidate anomalous regions only when they spatially form a connected cluster with a number greater than a preset lower limit for the number of blocks. Temporal persistence is determined using a sliding time window counter; that is, the aforementioned candidate anomalous regions must exist within a time window of length... Within the sliding window of frames, the number of frames that appear continuously reaches or exceeds a preset frame count threshold. Only then was it finally confirmed as a valid anomaly with continuity in space and time; and The value of is determined based on the maximum duration of normal disturbances exhibited by the device during baseline operation.

[0047] To link the results of active thermal excitation analysis with passive monitoring thresholds, the system establishes a clear mathematical model for the aforementioned risk weighting mechanism. When the thermal excitation module completes detection of a specific area and obtains its thermal decay time constant... Subsequently, the early warning judgment module will call the following linear decay function to calculate the mismatch threshold for this region, which has been adjusted for risk weighting. : In this model, It is the baseline time constant under the healthy state of the device. This is a positive calibration coefficient, which is set to maximize the expected value. Bias, mapped to a pair The system establishes a definite linear mapping relationship by setting a preset maximum adjustment range. At the same time, to define the backup monitoring logic under optical interference conditions, during the baseline data acquisition phase, the system projects all synchronized acoustic and vibration signal intensities and global temperature change rate data as data points into a two-dimensional phase space and constructs a minimum convex hull that can enclose all normal operating condition data points. The boundary of this convex hull constitutes the normal state space of the acoustic-vibration-thermal correlation model. In subsequent monitoring, once the system switches to this backup model, any real-time data point falling outside the boundary of this convex hull will be judged as a correlation anomaly. After this procedure is completed, all core judgment thresholds, logical rules, and backup models within the system have been transformed from general configurations to dedicated states that match the operating characteristics of this specific equipment.

[0048] Example 5: To ensure the consistency of deployment and monitoring reliability of the monitoring system of the present invention across different individual devices and in varying operating environments, a standardized on-site pre-deployment calibration procedure must be performed before the system is put into continuous monitoring. This procedure first requires the acquisition of an initial dynamic baseline database in one or more confirmed fault-free complete working cycles of the target device. The system then performs a baseline quality self-checking procedure on the database. By analyzing the dispersion and stationarity of the baseline time series Bi of all image blocks, if its statistical characteristics do not meet the preset benchmark model, it is determined that the initial baseline may have been affected by potential anomalies or noise contamination and prompts that it needs to be re-acquired. At the same time, in order to establish a robust event-driven wake-up mechanism, the acoustic and vibration acquisition module records the acoustic and vibration signal characteristics of the device under multiple different operating conditions such as startup, stable operation, and shutdown during this stage, and constructs an operating condition identification model containing multiple states and corresponding transition thresholds to replace the single acoustic and vibration intensity threshold. This allows the system to adaptively adjust its sleep and wake-up judgment conditions according to the specific operating stage of the device.

[0049] After baseline calibration is completed, the system will further perform adaptive fine-tuning of core parameters, especially for the aforementioned risk-weighted model. calibration coefficients in Its value is not determined by on-site debugging, but by conducting gradient fatigue loading tests on standard specimens of the same material as the target equipment in an offline materials laboratory. The test establishes the thermal decay time constant under different fatigue levels. The measured value and the minimum mismatch that can be detected when the corresponding material develops initial microcracks. Functional relationship between them, calibration coefficients This is the key parameter in this functional relationship, which is fixed as a physical property constant of the specific material and stored in the system. In addition, the procedure also clearly defines the exit mechanism for the judgment logic of optical interference state. That is, when the system enters the optical interference state, it will continuously monitor the global brightness statistical characteristics of the real-time image sequence. Only when the characteristic value returns to its baseline statistical range and remains stable for a preset period of time will the warning judgment module automatically exit the optical interference state and restore the warning logic based on the mismatch degree spatial distribution map.

[0050] Example 6: When the monitoring system is first deployed on a specific pressure device or after major maintenance, a standardized system health self-check and monitoring parameter adaptive configuration procedure must be executed. The first step of this procedure is the self-state verification of the image acquisition module. Under the condition that the target device is completely stationary and the surface illumination is uniform, the system continuously acquires a set of static image sequences and generates a baseline noise map characterizing the inherent noise level of the sensor by calculating the time standard deviation of the brightness value of each pixel in the sequence. This noise map is stored as a benchmark for subsequent sensor health assessment. During the long-term operation of the system, if the statistical characteristics of the new noise map obtained by periodically repeating this verification procedure drift beyond the preset range compared with the baseline noise map, the system determines that the image acquisition module itself may have deteriorated in performance and triggers a maintenance warning.

[0051] The next step in this procedure is to determine the size of the most critical image blocks in visual analysis. This determination is based on a balance between the detection sensitivity for the smallest expected anomaly and the overall computational load of the system. The operator first inputs the estimated engineering value of the smallest physical defect size to be detected for that specific device. The system, combining known camera resolution and object distance information, automatically calculates the pixel area corresponding to that physical size on the imaging plane. The block size setting procedure involves setting the pixel area of ​​a single image block within a predetermined proportional range of the calculated pixel area. This allows the average brightness variation of a single block to effectively reflect the physical size of the target defect. The emergence and evolution of defects; simultaneously, regarding the determination of key parameters in the secondary thermodynamic arbitration logic, namely the preset temperature rise rate and preset temperature rise amount, the system will analyze the statistical distribution of the temperature change rate and instantaneous temperature rise amplitude of all blocks during the initial fault-free working cycle data acquisition phase, and set the above two preset thresholds outside the upper limit of the confidence interval of the statistical distribution, so that the arbitration logic can effectively filter out normal thermodynamic fluctuations; to cope with the non-faulty slow surface changes that may occur after long-term operation of the equipment, the system also has a built-in adaptive update logic for the baseline time series; this logic continuously monitors the mismatch value of all image blocks. The long-term average drift, when most blocks When the value exhibits a consistent, slow growth trend without any thermodynamic anomalies over a relatively long period of time, and exceeds a preset global drift threshold, the system will determine that this phenomenon is caused by global surface variability. At this time, the system will automatically trigger a new baseline image sequence acquisition and analysis process, and replace the old baseline with the newly generated baseline time series. This mechanism enables the monitoring system to maintain the long-term ability to identify real faults without human intervention.

[0052] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0053] Finally, 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.

Claims

1. A multi-source pressure equipment safety monitoring system based on image fusion analysis, characterized in that, include: The image acquisition module is configured to acquire a baseline image sequence of the pressure equipment during one complete working cycle of the pressure equipment; And during continuous monitoring, real-time image sequences of the pressure equipment are acquired. The baseline analysis module is configured to divide the field of view of the baseline image sequence into multiple image blocks, and for each image block, extract the baseline time series of the mean pixel brightness over time. The real-time analysis module is configured to divide the field of view of the real-time image sequence into multiple image blocks in the same way as the baseline image sequence, and extract the real-time time series of the mean pixel brightness change over time for each image block. The mismatch calculation module is configured to calculate a mismatch value for each image block based on the time series similarity between the real-time time series and the baseline time series, using the following method. , in, This indicates the real-time time series at time 10:

00. The average pixel brightness Indicates the baseline time series at time 10:

00. The average pixel brightness This represents the average value of a real-time time series. This represents the average value of the baseline time series. It represents the number of sampling points in the time series; and generates a spatial distribution map of mismatch based on the mismatch values ​​of all image blocks; The temperature acquisition module is configured to acquire low-resolution temperature distribution data from the pressure equipment. The early warning judgment module is configured to output a high-confidence early warning message when an abnormal region appears in the mismatch spatial distribution map that is spatially continuous and temporally persistent for a duration equal to or exceeding a preset duration, and the abnormal region spatially overlaps with a temperature rise region in the temperature distribution data that is persistent and has a temperature higher than its neighboring regions. The early warning judgment module is also configured to determine that the system has entered an optical interference state and temporarily suspend the early warning logic based on the mismatch spatial distribution map when the image acquisition module detects that the difference between the global brightness statistical features of the real-time image sequence and the global brightness statistical features of the baseline image sequence exceeds a first preset brightness difference threshold, and the temperature distribution data acquired by the temperature acquisition module at the same time does not change by more than a second preset temperature difference threshold compared with the temperature distribution data at the previous time or under normal working conditions.

2. The multi-source pressure equipment safety monitoring system based on image fusion analysis according to claim 1, characterized in that, The system also includes an acoustic vibration acquisition module, configured to continuously monitor the acoustic vibration signal intensity of the pressure equipment. When the acoustic vibration signal intensity reaches or exceeds the preset acoustic vibration intensity threshold, a wake-up command is sent to the image acquisition module and the mismatch calculation module to enable the system to enter the working mode from the sleep mode. When the acoustic vibration signal intensity is lower than the preset acoustic vibration intensity threshold for a preset duration, the system automatically returns to the sleep mode.

3. The multi-source pressure equipment safety monitoring system based on image fusion analysis according to claim 1, characterized in that, The mismatch calculation module is configured to use a dynamic time warping algorithm or a cross-correlation function to calculate the similarity of time series.

4. The multi-source pressure equipment safety monitoring system based on image fusion analysis according to claim 1, characterized in that, The baseline analysis module is configured to periodically and automatically update the baseline time series.

5. A multi-source pressure equipment safety monitoring system based on image fusion analysis according to claim 1, characterized in that, The early warning judgment module is also configured to perform secondary thermodynamic arbitration on abnormal areas on the mismatch spatial distribution map before outputting high confidence early warning information. This is done by judging whether the temperature of abnormal areas on the mismatch spatial distribution map continues to rise within a preset time period and meets the judgment conditions of preset temperature rise rate or preset temperature rise amount. Only when it is confirmed that the judgment condition of continuous temperature rise is met will the high confidence early warning information be output.

6. The multi-source pressure equipment safety monitoring system based on image fusion analysis according to claim 1, characterized in that, The image acquisition module is configured as an industrial camera to record changes in pixel brightness, and the temperature acquisition module is configured as an infrared array sensor to provide logical confidence weighting for visual anomalies.

7. A multi-source pressure equipment safety monitoring system based on image fusion analysis according to claim 1, characterized in that, The mismatch calculation module is configured as an edge computing unit for one-dimensional time series comparison.

8. A multi-source pressure equipment safety monitoring system based on image fusion analysis according to claim 1, characterized in that, The early warning judgment module is configured to pause the early warning logic based on the mismatch degree spatial distribution map when the system is determined to enter the optical interference state, and activate a temporary acoustic vibration-thermal correlation model. Under this model, when the acoustic vibration acquisition module detects that the amplitude of the acoustic vibration signal intensity fluctuation exceeds the third preset acoustic vibration fluctuation threshold, and the temperature acquisition module detects that the temperature distribution change exceeds the fourth preset temperature anomaly threshold, an alarm is issued.

9. A multi-source pressure equipment safety monitoring system based on image fusion analysis according to claim 1, characterized in that, The system also includes: a thermal excitation module configured to periodically control a thermal excitation source to apply a standardized thermal pulse to a predetermined area of ​​the pressure equipment; a temperature acquisition module configured to monitor the temperature decay characteristics of the predetermined area after the thermal pulse is applied, so as to infer the material fatigue state of the area; and an early warning judgment module configured to weight the risk level in the mismatch spatial distribution map according to the material fatigue state.

10. A multi-source pressure equipment safety monitoring system based on image fusion analysis according to claim 9, characterized in that, The temperature acquisition module characterizes the thermal diffusion rate of a material by analyzing the thermal decay time constant of the temperature decay characteristic curve. The extension of the thermal decay time constant indicates an increase in the degree of fatigue of the material, and the fatigue state of the material is quantified by comparing it with a preset healthy baseline time constant.

Images