Pressure monitoring system

Abstract

The invention relates to the technical field of pressure monitoring, and discloses a pressure monitoring system, which comprises a multi-mode sensing module used for acquiring acoustic resonance information and an optical disturbance field image sequence of a medium; the multi-dimensional feature extraction module is used for constructing acoustic resonance spectrum features and analyzing and generating medium disturbance field fluctuation entropy features; the environment parameter compensation module is used for acquiring environment parameters; the fusion intelligent analysis engine is used for fusing acoustic resonance spectrum features, medium disturbance field fluctuation entropy features and environmental parameters; and the distributed intelligent early warning module is used for predicting the result or the identified abnormal information according to the fusion intelligent analysis engine. According to the invention, acoustic resonance information and an optical disturbance field image sequence are acquired by using the multi-modal sensing module, and acoustic resonance spectrum features and medium disturbance field fluctuation entropy features are constructed by using the multi-dimensional feature extraction module, so that subtle precursor information of medium pressure change is conveniently captured, and a pressure prediction result is constructed.

Description

Technical Field

[0001] This application relates to the field of pressure monitoring technology, specifically a pressure monitoring system. Background Technology

[0002] Pressure monitoring is increasingly important as a key technology in many fields, including industrial process control, infrastructure safety operation, and environmental monitoring. Accurate and continuous pressure data is fundamental for assessing system health, optimizing operating parameters, predicting potential failures, and ensuring the safety of personnel and equipment. For example, in the petrochemical industry, pressure monitoring of pipelines and storage tanks is directly related to production efficiency and leakage risk. With the increasing level of industrial automation and intelligence, the performance requirements for pressure monitoring systems, especially their accuracy, real-time performance, and ability to identify early anomalies, are continuously increasing.

[0003] Existing pressure monitoring technologies primarily rely on single-physical-quantity sensors, such as piezoresistive, capacitive, strain gauge, or piezoelectric sensors. These sensors convert the physical pressure of the medium into an electrical signal through direct contact or indirect sensing. After signal conditioning, amplification, and digitization, the signal is read by operators or input into the control system. These technologies are widely used in industrial settings, for example, by monitoring fluid pressure through pressure transmitters installed on pipelines or by directly displaying the pressure value inside a storage tank using pressure gauges.

[0004] However, existing pressure monitoring technologies are susceptible to interference from environmental factors such as temperature, humidity, and external vibrations, which can affect the measurement accuracy of sensors, leading to data drift or distortion and reducing the reliability of monitoring results. Furthermore, data processing methods often focus on threshold alarms, making it difficult to extract deeper correlation patterns and trend information from massive amounts of data. This limits the system's ability to make proactive predictions. Traditional sensors lack the ability to detect microscopic changes within the medium, and a single pressure sensor cannot effectively capture these early warning signs, resulting in delayed warnings and increasing the risk of accidents and maintenance costs. Therefore, this invention provides a pressure monitoring system to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this application is to provide a pressure monitoring system that solves the problems of insufficient monitoring accuracy, timely early warning, and intelligent fault diagnosis in existing pressure monitoring and early warning technologies.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The first aspect of the present invention provides a pressure monitoring system, including a multimodal sensing module, a multidimensional feature extraction module, an environmental parameter compensation module, a fusion intelligent analysis engine, and a distributed intelligent early warning module.

[0008] The multimodal sensing module is used to acquire acoustic resonance information and optical disturbance field image sequences of the monitored medium in a non-contact manner. Specifically, the module includes an acoustic resonance excitation and reception unit and an optical medium disturbance field imaging unit.

[0009] The acoustic resonance excitation and reception unit includes an array of acoustic wave exciters and an array of non-contact acoustic sensors. The acoustic wave exciter array is used to emit swept-frequency acoustic waves or composite excitation waveforms into the medium. The non-contact acoustic sensor array is used to capture the resonant response signal of the medium induced by the acoustic wave exciter array.

[0010] The optical medium perturbation field imaging unit includes a beam emitter and an imaging array. The beam emitter emits a structured beam or vortex beam modulated by a spatial light modulator. The imaging array captures the changes in the optical field generated by the interaction between the beam and the medium after the beam penetrates the medium, and generates a perturbation field image sequence based on these changes.

[0011] The multi-dimensional feature extraction module is electrically connected to the multi-modal sensing module. This module is used to construct acoustic resonance spectrum features based on the acquired acoustic resonance information, and to generate medium perturbation field wave entropy features based on the analysis of the acquired optical perturbation field image sequence.

[0012] The construction of acoustic resonance spectrum features involves time-frequency analysis of the response signals acquired by acoustic sensors. Specifically, the center frequencies, peak intensities, and quality factors of the resonance modes of acoustic resonance peaks can be identified and extracted. For example, the center frequencies of each resonance peak can be identified by performing a Fourier transform on the digitized acoustic signal. and peak intensity Quality factor of resonance mode It can be calculated using the following formula:

[0013] ;

[0014] In the formula, Indicates the first The observed frequency of each resonance mode; Indicates the first Half-power bandwidth of each resonant mode.

[0015] The analysis of the wave entropy characteristics of a medium perturbation field involves processing a sequence of optical perturbation field images. Specifically, multi-scale wavelet decomposition can be performed on each image frame in the image sequence, and the wave entropy can be calculated based on the decomposition results. Wave entropy The calculation process is as follows: For an image frame At each scale Calculate the energy of wavelet coefficients ,Right now In the formula It is a scale The wavelet coefficients. Calculate the normalized energy probability at each scale. Final fluctuation entropy Defined as:

[0016] ;

[0017] In the formula, This represents the normalized energy probability at each scale; Indicated in scale The energy of the wavelet coefficients calculated above; This represents the sum of the energy of the wavelet coefficients across all scales; It represents the fluctuation entropy.

[0018] Fluctuation entropy reflects the degree of disorder in the pixel intensity distribution of an image. Furthermore, the area, shape, and pixel gradient distribution features of perturbed regions can be extracted from image sequences.

[0019] The environmental parameter compensation module is used to acquire environmental parameters, including ambient temperature, ambient humidity, and ambient pressure. These environmental parameters are used to construct a parameter compensation model to correct the acoustic resonance spectrum features constructed by the multi-dimensional feature extraction module and the wave entropy features of the media disturbance field generated by analysis in real time, thereby eliminating the influence of environmental factors on the accuracy of feature measurement and calculation.

[0020] The fusion intelligent analysis engine is electrically connected to the multi-dimensional feature extraction module and the environmental parameter compensation module. This engine fuses compensated acoustic resonance spectrum features, medium disturbance field wave entropy features, and environmental parameters. The fused data is input into a multimodal fusion deep learning model for deep learning and pattern recognition. The multimodal fusion deep learning model includes a biomimetic auditory neural network (BANN) architecture, a graph neural network (GNN), and a Transformer structure. The model is trained based on the fused data to learn and construct the nonlinear dynamic correlations and complementary information between various features under different operating conditions.

[0021] The fusion intelligent analysis engine, based on learned nonlinear dynamic correlations, analyzes subtle changing trends of the fused multi-dimensional features to construct predictions of future pressure values ​​and pressure drop rates, enabling early warning. These predictions can be further used to generate visual charts or trend curves to intuitively display pressure change trends.

[0022] The fusion intelligent analysis engine identifies abnormal patterns related to microscopic disturbances in the medium based on learned multi-dimensional feature patterns, thereby constructing anomaly identification information. These abnormal patterns include initial microcracks, minor valve leaks, and changes in medium composition. The anomaly identification information may further include the anomaly type, probability of occurrence, and recommended maintenance measures.

[0023] The distributed intelligent early warning module is electrically connected to the fusion intelligent analysis engine. This module is used to provide rapid early warnings and transmit information locally based on the prediction results constructed by the fusion intelligent analysis engine or the anomaly information identified. The distributed intelligent early warning module includes an edge computing unit and a wireless communication module.

[0024] The edge computing unit is used to process and intelligently judge the prediction results or anomaly information identified by the fusion intelligent analysis engine in real time, and filter out key early warning events that need to be reported immediately. The wireless communication module is used to trigger event-driven data transmission based on the key early warning events, sending the refined early warning information to a local display, industrial human-machine interface (HMI) or central monitoring platform, and can trigger audible and visual alarms or SMS notifications.

[0025] A second aspect of the present invention provides a pressure monitoring method, comprising the following steps:

[0026] Acquiring multimodal sensing data: Non-contact acquisition of acoustic resonance information and optical disturbance field image sequences of the medium under test.

[0027] Constructing and generating multi-dimensional features: Time-frequency analysis is performed based on the acquired acoustic resonance information to construct acoustic resonance spectrum features; at the same time, multi-scale wavelet decomposition is performed based on the acquired optical perturbation field image sequence, and the medium perturbation field wave entropy features are generated based on the decomposition results.

[0028] Environmental parameter compensation: Obtain environmental parameters and use them to construct a parameter compensation model to correct the constructed acoustic resonance spectrum characteristics and the generated medium disturbance field wave entropy characteristics, thereby eliminating the influence of environmental factors on the accuracy of the characteristics.

[0029] Multimodal data fusion and intelligent analysis are performed: the constructed acoustic resonance spectrum features, the generated medium disturbance field wave entropy features, and the acquired environmental parameters are fused, and the fused data is input into a fusion intelligent analysis model for deep learning and pattern recognition. This model learns and constructs nonlinear dynamic correlations and complementary information between various features under different operating conditions through training.

[0030] Generating prediction results or anomaly identification information: Based on the nonlinear dynamic correlation learned by the fusion intelligent analytical model, the subtle changing trends of the fused multi-dimensional features are analyzed to construct prediction results for the future pressure value or pressure drop rate of the medium. Simultaneously, based on the complementary information learned by the fusion intelligent analytical model, abnormal patterns of micro-disturbances in the medium are identified, and anomaly identification information is constructed.

[0031] Local intelligent early warning and information transmission: Based on the constructed prediction results or anomaly identification information, the edge computing unit performs real-time processing and intelligent judgment to filter out key early warning events that require immediate reporting. Based on these key early warning events, event-driven data transmission is triggered, sending the refined early warning information to a local display, industrial HMI, or central monitoring platform, and may trigger audible and visual alarms or SMS notifications.

[0032] In summary, this application includes at least one of the following beneficial technical effects:

[0033] 1. This invention employs a multimodal sensing module to acquire acoustic resonance information and optical disturbance field image sequences, and a multi-dimensional feature extraction module to construct acoustic resonance spectrum features and medium disturbance field wave entropy features. Combined with an environmental parameter compensation module to correct for environmental factors, and a fusion intelligent analysis engine to perform deep fusion and nonlinear correlation learning of multi-dimensional features, the system can capture subtle precursor information of medium pressure changes, thereby constructing pressure prediction results.

[0034] 2. This invention utilizes a multimodal fusion deep learning model in a fusion intelligent analysis engine to fully learn and utilize the complementary information between acoustic and optical data, enabling the system to effectively identify abnormal patterns caused by micro-disturbances in the medium, such as initial micro-cracks, minor valve leaks, or changes in medium composition, thereby improving the sensitivity of early fault diagnosis.

[0035] 3. The present invention employs a distributed intelligent early warning module, including an edge computing unit and a low-power wireless communication module. The edge computing unit can process and intelligently judge the prediction results or abnormal information output by the fusion intelligent analysis engine in real time locally, and filter out key early warning events that need to be reported immediately. The low-power wireless communication module can trigger event-driven data transmission based on these key events to ensure that the early warning information is sent to the relevant terminals in a timely manner. Attached Figure Description

[0036] Figure 1 This is a system architecture diagram for this application;

[0037] Figure 2 This is a schematic diagram of the multimodal sensing module structure of this application;

[0038] Figure 3 This is a schematic diagram of the multi-dimensional feature extraction module structure in this application;

[0039] Figure 4 This is a schematic diagram of the environmental parameter compensation module structure in this application;

[0040] Figure 5 This is a schematic diagram of the fusion intelligent parsing engine structure of this application;

[0041] Figure 6 This is a schematic diagram of the distributed intelligent early warning module structure of this application;

[0042] Figure 7 This is a schematic diagram of the pressure monitoring method of this application. Detailed Implementation

[0043] The following is in conjunction with the appendix Figure 1 - Appendix Figure 7 This application will be described in further detail below.

[0044] See attached document Figure 1 , Figure 1 This is an architecture diagram of a pressure monitoring system according to an embodiment of the present invention. The present invention provides a pressure monitoring system, including a multimodal perception module, a multi-dimensional feature extraction module, an environmental parameter compensation module, a fusion intelligent analysis engine, and a distributed intelligent early warning module.

[0045] The multimodal sensing module is used to acquire acoustic resonance information and optical disturbance field image sequences of the monitored medium in a non-contact manner. This module uses physical sensing methods to collect the physical response of the medium under specific excitations, generating raw data as the basis for subsequent data processing.

[0046] The multi-dimensional feature extraction module is electrically connected to the multi-modal sensing module. It is used to construct acoustic resonance spectrum features based on the acquired acoustic resonance information and to generate medium perturbation field wave entropy features based on the acquired optical perturbation field image sequence. This module is responsible for transforming the raw, high-dimensional sensing data into quantitative features with physical meaning and recognition value.

[0047] The environmental parameter compensation module is used to acquire environmental parameters and use them as compensation inputs for acoustic resonance spectrum features or medium disturbance field wave entropy features. This module aims to quantify and provide the potential impact of external environmental factors (such as temperature, humidity, and pressure) on the sensed data, providing a basis for subsequent feature correction.

[0048] The integrated intelligent analysis engine is electrically connected to the multi-dimensional feature extraction module and the environmental parameter compensation module. This engine fuses compensated acoustic resonance spectrum features, medium disturbance field wave entropy features, and environmental parameters. The fused data is input into a deep learning model for deep learning and pattern recognition, thereby predicting pressure trends in the medium or identifying anomalies within the medium. This module is the core intelligent processing unit of the system, responsible for mining potential correlation patterns from multi-source heterogeneous data.

[0049] The distributed intelligent early warning module is electrically connected to the fusion intelligent analysis engine, and is used to quickly issue early warnings and transmit information locally based on the prediction results or anomaly information identified by the fusion intelligent analysis engine. This module is responsible for converting the results of intelligent analysis into actionable early warning signals and information transmission, realizing closed-loop feedback of the monitoring system.

[0050] The multimodal perception module generates raw perception data and transmits it to the multidimensional feature extraction module for feature construction. Then, the environmental parameter compensation module obtains environmental parameters. The constructed features and the obtained environmental parameters are transmitted to the fusion intelligent analysis engine for fusion and intelligent analysis to generate prediction results or abnormal information. These prediction results or abnormal information are transmitted to the distributed intelligent early warning module for local early warning and information transmission.

[0051] See attached document Figure 2 , Figure 2 This is a schematic diagram of a multimodal sensing module according to an embodiment of the present invention. The multimodal sensing module of the present invention is used to acquire acoustic resonance information and optical disturbance field image sequences of the medium to be monitored in a non-contact manner. This module achieves comprehensive sensing of the medium's state by integrating sensor units based on different physical principles.

[0052] The multimodal sensing module includes an acoustic resonance excitation and reception unit and an optical medium perturbation field imaging unit.

[0053] The acoustic resonance excitation and reception unit is used to actively excite the acoustic resonance modes of the medium and acquire its response signals. This unit includes an array of acoustic wave exciters and an array of non-contact acoustic sensors. The acoustic wave exciter array, for example, consists of multiple broadband, tunable miniature piezoelectric or pneumatic exciters, designed to ensure the emission of sweep-frequency acoustic waves or composite excitation waveforms into the medium being monitored. The emitted acoustic waves propagate within the medium and interact with its structure, exciting the medium's inherent resonance.

[0054] Non-contact acoustic sensor arrays, such as those consisting of multiple high-sensitivity, wide-bandwidth laser vibrometers and microelectromechanical systems (MEMS) microphone arrays, are used to capture the resonant response signals of a medium induced by an array of acoustic wave exciters. These sensors are positioned externally to the medium, non-invasively acquiring acoustic signals generated by vibrations on or within the medium's surface. The acquired acoustic response signals are then digitally processed to form acoustic resonance information.

[0055] The optical medium perturbation field imaging unit is used to emit a light beam that penetrates the medium and acquires a sequence of perturbation field images. This unit includes a beam emitter and an imaging array. The beam emitter emits a structured beam or vortex beam modulated by a spatial light modulator (SLM). The SLM generates an optical field with a specific spatial structure, such as a beam with a helical phase, by modulating the phase or amplitude of the incident beam. These specialized beams penetrate the medium being monitored and interact with minute perturbations within the medium (such as density variations or refractive index inhomogeneities), causing changes in the optical field.

[0056] Imaging arrays, such as those composed of ultra-high resolution complementary metal-oxide-semiconductor (CMOS) camera arrays and charge-coupled device (CCD) camera arrays, are used to capture changes in the optical field resulting from the interaction of a specific light beam with a medium. The imaging array converts the captured optical field changes into a series of two-dimensional image frames, forming a perturbation field image sequence. The grayscale value of each pixel in the image sequence... Indicates at time Spatial location The light field intensity at that point, i.e.:

[0057] ;

[0058] In the formula, Indicates pixel intensity; This represents the light intensity captured by the imaging array at the corresponding location and time. These image sequences contain spatial and temporal information about the perturbations within the medium.

[0059] See attached document Figure 3 , Figure 3 This is a schematic diagram of a multi-dimensional feature extraction module according to an embodiment of the present invention. The multi-dimensional feature extraction module of the present invention is electrically connected to a multi-modal sensing module, and is used to construct acoustic resonance spectrum features based on the acquired acoustic resonance information, and to generate medium disturbance field wave entropy features based on the analysis of the acquired optical disturbance field image sequence.

[0060] The construction of acoustic resonance spectrum features involves time-frequency analysis of the acoustic resonance information acquired by the acoustic resonance excitation and receiving units. Specifically, time-frequency analysis methods such as Fourier transform or wavelet transform can be used to convert the time-domain acoustic response signal into a frequency domain or time-frequency domain representation, thereby identifying the inherent resonance modes of the medium.

[0061] Resonance frequency drift: Monitoring the change in the center frequency of a specific resonance mode over time. Changes in medium pressure or internal structure can alter its acoustic properties, thereby causing a shift in the resonance frequency.

[0062] Resonance peak intensity variation: Tracking the change in the peak intensity of the resonance peak. Changes in the damping properties or energy dissipation of the medium will be reflected in the increase or decrease of the resonance peak intensity.

[0063] Quality factor of resonance mode ( (Value): The quality factor reflects the energy dissipation characteristics of a resonant system. For the th... Each resonance mode has a quality factor It can be calculated using the following formula:

[0064] ;

[0065] In the formula, Indicates the first The center frequency of each resonance mode; This represents the 3dB bandwidth (i.e., half-power bandwidth) of the resonance mode. The lower the value, the greater the energy loss. These features are extracted by analyzing the spectrum of the acoustic signal, identifying resonance peaks, and calculating their corresponding parameters.

[0066] The process of analyzing and generating the wave entropy characteristics of the medium perturbation field involves multi-scale analysis of the optical perturbation field image sequence acquired by the optical medium perturbation field imaging unit. This process aims to quantify the degree and distribution characteristics of light field changes caused by microscopic perturbations within the medium in the image sequence.

[0067] Specifically, multi-scale wavelet decomposition is performed on each frame of the optical perturbation field image sequence. Wavelet decomposition can decompose the image into sub-bands of different frequencies and directions, thereby capturing the detailed information and global trends of the image at different scales. Based on the results of wavelet decomposition, the wave entropy characteristics of the medium perturbation field are calculated. Wave entropy is an indicator that measures the complexity and disorder of image information; its value reflects the randomness or irregularity of the perturbation field inside the medium.

[0068] In addition to fluctuation entropy, other auxiliary features can be extracted from image sequences, such as the area, shape, and pixel gradient distribution characteristics of the perturbation region. The area and shape of the perturbation region can characterize the macroscopic size and geometry of the internal anomaly. Pixel gradient distribution characteristics can reflect the sharpness of the perturbation field boundary or the severity of internal changes.

[0069] See attached document Figure 4 , Figure 4 This is a schematic diagram of an environmental parameter compensation module according to an embodiment of the present invention. The environmental parameter compensation module of the present invention is used to acquire environmental parameters and use them as compensation inputs for the acoustic resonance spectrum features constructed by the multi-dimensional feature extraction module and the wave entropy features of the media disturbance field generated by analysis.

[0070] The environmental parameter compensation module acquires real-time data on the environment in which the monitored medium exists through integrated environmental sensors. The acquired environmental parameters include, but are not limited to, ambient temperature, ambient humidity, and ambient pressure. These sensors can be installed directly near the monitoring system or receive data wirelessly from external environmental monitoring stations.

[0071] The environmental parameter compensation module constructs a parameter compensation model based on the acquired environmental parameters. This model is used to correct the acoustic resonance spectrum features constructed by the multi-dimensional feature extraction module and the media disturbance field wave entropy features generated analytically in real time. Environmental factors, especially temperature and pressure, affect the sound velocity, medium density, and the refractive index of light propagating in the medium, thus causing drift in the measured values ​​of the acoustic resonance spectrum features and optical disturbance field wave entropy features. The parameter compensation model can eliminate the influence of these environmental factors on the accuracy of the features.

[0072] For example, for compensation of acoustic resonant frequencies, the velocity of sound in the medium With temperature The resonant frequency changes with the frequency of the resonant energy, thus affecting the resonant frequency. The compensated resonant frequency... It can be represented as:

[0073] ;

[0074] In the formula, It is the measured resonant frequency; It is the compensation coefficient for the effect of temperature on the resonant frequency; This is the current ambient temperature; This is a reference temperature.

[0075] The wave entropy characteristics of optical perturbation fields can also be affected by environmental parameters. For example, changes in humidity may cause microscopic water vapor condensation or diffusion in the medium, affecting the propagation characteristics of the light field. (Compensated wave entropy) It can be represented as:

[0076] ;

[0077] In the formula, It is the measured fluctuation entropy; It is the compensation coefficient for the effect of humidity on fluctuation entropy; This refers to the current ambient humidity. This is a reference humidity level. These compensation coefficients and model structures can be obtained through experimental data calibration, ensuring the accuracy and comparability of characteristic data under different environmental conditions.

[0078] See attached document Figure 5 , Figure 5 This is a schematic diagram of a fusion intelligent analysis engine according to an embodiment of the present invention. The fusion intelligent analysis engine of the present invention is electrically connected to a multi-dimensional feature extraction module and an environmental parameter compensation module. The engine is used to fuse compensated acoustic resonance spectrum features, medium disturbance field wave entropy features, and environmental parameters, and performs deep learning and pattern recognition based on the fused data, thereby predicting the pressure trend of the medium or identifying anomalies in the medium.

[0079] Multimodal data fusion is the primary component of the integrated intelligent analysis engine. This mechanism aims to effectively integrate heterogeneous feature data from acoustic, optical sensors, and environmental parameters to form a unified, high-dimensional feature representation. Specifically, this involves the acoustic resonance spectrum features corrected by the environmental parameter compensation module. Characteristics of wave entropy in medium disturbance field and related environmental parameters ( (Such as temperature, humidity, and pressure) are synchronously transmitted to this engine.

[0080] These feature data from different sources are first standardized or normalized to eliminate the influence of differences in units and numerical ranges. The standardized feature vectors are then concatenated or fused at the feature layer using methods such as weighted summation to form a comprehensive fused feature vector. Fusion of feature vectors It can be represented as:

[0081] ;

[0082] In the formula, It is the first A characteristic of the acoustic resonance spectrum; It is the first The wave entropy characteristics of a medium disturbance field; It is the first One environmental parameter; , , These represent the quantity of each type of feature. This fused feature vector serves as the input to the deep learning model.

[0083] The fusion intelligent parsing engine employs a multimodal fusion deep learning model. This model includes a biomimetic auditory neural network architecture, a graph neural network, and a Transformer structure, the choice of which depends on the specific data characteristics and task requirements.

[0084] Bionic auditory neural network architecture excels at processing complex patterns in temporal signals, simulating the extraction and classification process of sound features by biological auditory systems, and is suitable for analyzing temporal changes in acoustic resonance spectra. BANN extracts deep patterns related to pressure changes or anomalies from acoustic features through multi-layer nonlinear transformations.

[0085] Graph Neural Networks (GNNs) can effectively handle nonlinear dynamic relationships and complementary information between data. In pressure monitoring, complex interactions exist between different acoustic features, optical features, and environmental parameters. GNNs capture these complex multidimensional relationships by constructing graph structures between features and utilizing operations such as graph convolution to learn information on nodes (features) and edges (associations). For example, a simple graph convolutional layer operation can be represented as:

[0086] ;

[0087] In the formula, It is a node In the Layer feature representation; It is a node The set of neighboring nodes; It is the first Layer weight matrix; It is the bias vector; It is the normalization coefficient.

[0088] The Transformer architecture is well-suited for capturing long-range dependencies and global information in sequence data. In stress monitoring scenarios, historical multi-dimensional feature sequences are crucial for predicting future stress trends or identifying anomalies. Through its self-attention mechanism, the Transformer assigns different weights to different features and their temporal relationships within the fused feature vector, thereby more effectively extracting the key information needed for prediction and identification.

[0089] The multimodal fusion deep learning model is trained based on the fused acoustic resonance spectrum features, medium disturbance field wave entropy features, and environmental parameters. The training process involves inputting a large amount of historical monitoring data (including data under normal and abnormal conditions) and using actual pressure values ​​or anomaly labels as supervisory signals. The backpropagation algorithm is used to adjust the model parameters, enabling the model to learn and construct nonlinear dynamic correlations and complementary information between various features under different operating conditions. The training objective is to minimize the error between the model's predicted results and the actual values, or to achieve a preset standard for the accuracy and recall of anomaly identification.

[0090] The fusion intelligent analysis engine analyzes the subtle changing trends of the fused multi-dimensional features based on the learned nonlinear dynamic correlation, thereby constructing prediction results for the future pressure value and the pressure drop rate of the medium.

[0091] The prediction results are constructed by using a deep learning model to infer from the fused feature vectors of the current and historical data. The predicted value output by the model directly represents the pressure value of the medium at a future moment, or the rate of pressure drop within a specific time window. For example, the model can predict the pressure value in the next 1 hour, 6 hours, or 24 hours, or predict the pressure drop per minute.

[0092] The forecast results can be further used to generate visualizations or trend curves. For example, the predicted pressure values ​​can be correlated with a time axis to generate a real-time pressure trend curve, or the predicted rate of pressure decrease can be displayed as a bar chart or line chart. These visualizations are designed to visually represent the pressure change trends of the medium, making it easier for operators to interpret the data.

[0093] The integrated intelligent analysis engine identifies abnormal patterns related to microscopic disturbances in the medium based on learned multi-dimensional feature patterns, thereby constructing anomaly identification information.

[0094] Anomaly pattern identification is based on the classification or clustering results of the input fused feature vectors using a deep learning model. The model analyzes the degree of matching between the feature patterns and a known anomaly pattern library to determine whether the current medium state deviates from the normal range and, more specifically, the type of deviation. Anomaly patterns include initial microcracks, minor valve leaks, and changes in medium composition. For example, a specific combination of changes in the acoustic resonance spectrum and the specific morphology of the optical perturbation field wave entropy may indicate the appearance of initial microcracks; while enhancement of acoustic signals at certain frequencies combined with blurring of specific regions in an optical image may indicate minor valve leaks.

[0095] Anomaly identification information can further include the anomaly type, probability of occurrence, and suggested maintenance measures. The anomaly type specifies the specific anomaly detected (e.g., initial microcracks). The probability of occurrence is the model's confidence level in the existence of the anomaly, given in numerical form. The suggested maintenance measures are preliminary suggestions based on the identified anomaly type, either preset by the system or generated by a rule engine, such as suggesting non-destructive testing, checking valve leaks, or sampling and analyzing the composition of the medium.

[0096] See attached document Figure 6 , Figure 6 This is a schematic diagram of a distributed intelligent early warning module according to an embodiment of the present invention. The distributed intelligent early warning module of the present invention is electrically connected to a fusion intelligent analysis engine and is used to perform rapid early warning and information transmission locally based on the prediction results constructed by the fusion intelligent analysis engine or the abnormal information identified.

[0097] The distributed intelligent early warning module includes an edge computing unit. This edge computing unit is a low-power computing device integrating a processor, memory, and communication interface, deployed in the field close to the medium being monitored. The edge computing unit is used to process and intelligently judge the prediction results or identified anomalies generated by the fusion intelligent analysis engine in real time. This localized processing avoids the latency and bandwidth consumption of sending all raw data back to a central server for processing.

[0098] The edge computing unit receives stress prediction results (e.g., future stress value, stress decrease rate) and anomaly identification information (e.g., anomaly type, probability of occurrence) from the fusion intelligent analysis engine. Based on preset warning rules and thresholds, the edge computing unit performs logical judgments on this information. For example, if the predicted stress value is lower than a safety threshold, or the stress decrease rate exceeds a preset limit, an alert is triggered. Similarly, if the probability of an anomaly pattern occurring is higher than a certain confidence level, it is identified as a critical warning event. Through this intelligent judgment, the edge computing unit can filter out critical warning events that require immediate reporting, avoiding the transmission of irrelevant information.

[0099] The distributed intelligent early warning module also includes a low-power wireless communication module. This wireless communication module is electrically connected to the edge computing unit and is used to trigger event-driven data transmission based on key early warning events. It sends refined early warning information to a local display, industrial human-machine interface (HMI), or central monitoring platform, and can trigger audible and visual alarms or SMS / App notifications.

[0100] Event-driven data transmission means that the wireless communication module only initiates data transmission when the edge computing unit determines that a critical warning event has occurred, rather than periodically transmitting all data. This transmission mechanism reduces energy consumption and communication load. The wireless communication module can employ various low-power wireless communication protocols.

[0101] Warning information is sent to local displays and industrial human-machine interfaces (HMIs) to provide field operators with real-time visual alerts and detailed information. Simultaneously, information can also be transmitted to a central monitoring platform for remote monitoring, data archiving, and advanced analytics. Furthermore, the system can be configured to issue audible and visual alarms upon detecting critical warning events and send notifications to designated personnel via SMS or a mobile application (App), ensuring timely delivery of critical information for appropriate action.

[0102] See attached document Figure 7 , Figure 7 This is a schematic flowchart of a pressure monitoring method according to an embodiment of the present invention. The pressure monitoring method of the present invention is implemented based on the above-described pressure monitoring system and includes the following steps:

[0103] Multimodal sensing data is acquired non-contactly through the multimodal sensing module to obtain acoustic resonance information and optical disturbance field image sequences of the monitored medium. The acoustic resonance excitation and reception unit actively excites the acoustic resonance mode of the medium and collects the response signal to obtain acoustic resonance information; the optical medium disturbance field imaging unit emits a special beam of light to penetrate the medium and capture changes in the light field, generating an optical disturbance field image sequence.

[0104] Multi-dimensional features are constructed and generated by a multi-dimensional feature extraction module. Based on the acquired acoustic resonance information, time-frequency analysis is performed to construct acoustic resonance spectrum features, including resonance frequency, peak intensity, and quality factor. Simultaneously, multi-scale wavelet decomposition and wave entropy calculation are performed based on the acquired optical perturbation field image sequence to generate medium perturbation field wave entropy features, as well as other image features, such as the area, shape, and pixel gradient distribution of the perturbation region.

[0105] Environmental parameter compensation is performed by the environmental parameter compensation module, which acquires environmental parameters such as ambient temperature, humidity, and pressure. These parameters are used to construct a parameter compensation model to correct the constructed acoustic resonance spectrum characteristics and the generated medium disturbance field wave entropy characteristics in real time. The correction process aims to eliminate the influence of environmental factors on the accuracy of feature measurement and calculation, ensuring the purity of the feature data.

[0106] Multimodal data fusion and intelligent analysis are performed by a fusion intelligent analysis engine. This engine fuses compensated acoustic resonance spectrum features, medium disturbance field wave entropy features, and acquired environmental parameters to form a unified fusion feature vector. This fusion feature vector is then input into a pre-trained multimodal fusion deep learning model (e.g., a model incorporating biomimetic auditory neural networks, graph neural networks, and Transformer structures) for deep learning and pattern recognition. Through learning and training, this model constructs nonlinear dynamic correlations and complementary information among various features under different operating conditions.

[0107] The generation of prediction results or anomaly identification information is executed by the fusion intelligent analysis engine. Based on the nonlinear dynamic correlation learned by the model, it analyzes the subtle changing trends of the fused multi-dimensional features to construct prediction results for the future pressure value or pressure drop rate of the medium. Simultaneously, based on the complementary information learned by the model, it identifies anomaly patterns related to microscopic disturbances in the medium (such as initial microcracks, minor valve leaks, and changes in medium composition) and constructs corresponding anomaly identification information, which may include the anomaly type, probability of occurrence, and suggested maintenance measures.

[0108] Local intelligent early warning and information transmission are performed by a distributed intelligent early warning module. Based on the constructed prediction results or anomaly identification information, the edge computing unit performs real-time processing and intelligent judgment to filter out key early warning events that require immediate reporting. Once a key early warning event is identified, the low-power wireless communication module will trigger event-driven data transmission based on the event, sending the refined early warning information to a local display, industrial human-machine interface, or central monitoring platform. It can also trigger audible and visual alarms or SMS notifications to ensure that the early warning information is delivered to relevant parties in a timely and accurate manner.

Claims

1. A pressure monitoring system, characterized in that, include: A multimodal sensing module is used to acquire acoustic resonance information and optical disturbance field image sequences of the medium under test in a non-contact manner. The multi-dimensional feature extraction module is used to construct acoustic resonance spectrum features based on the acquired acoustic resonance information, and to generate medium disturbance field wave entropy features based on the acquired optical disturbance field image sequence analysis. The environmental parameter compensation module is used to acquire environmental parameters and use them as compensation for acoustic resonance spectrum characteristics or medium disturbance field wave entropy characteristics. The integrated intelligent analysis engine is used to fuse acoustic resonance spectrum features, medium disturbance field wave entropy features, and environmental parameters, and performs deep learning and pattern recognition based on the fused data to predict the pressure trend of the medium or identify anomalies in the medium. The distributed intelligent early warning module is used to quickly issue early warnings and transmit information locally based on the prediction results or anomaly information identified by the fusion intelligent analysis engine.

2. The pressure monitoring system according to claim 1, characterized in that, The multimodal sensing module includes: Acoustic resonance excitation unit, used to actively excite the acoustic resonance mode of the medium and acquire the response signal; An optical medium perturbation field imaging unit is used to emit a light beam that penetrates the medium and acquires a perturbation field image sequence.

3. The pressure monitoring system according to claim 2, characterized in that, The acoustic resonance excitation unit includes an acoustic wave exciter array and a non-contact acoustic sensor array. The acoustic wave exciter array emits swept-frequency acoustic waves or composite excitation waveforms into the medium, and works with the non-contact acoustic sensor array to capture the resonance response signal of the medium caused by the acoustic wave exciter array.

4. A pressure monitoring system according to claim 2, characterized in that, The optical medium perturbation field imaging unit includes a beam emitter and an imaging array. The beam emitter emits a structured beam or vortex beam modulated by a spatial light modulator. The imaging array captures the changes in the light field generated after the beam interacts with the medium, thus generating a perturbation field image sequence.

5. A pressure monitoring system according to claim 1, characterized in that, The environmental parameters acquired by the environmental parameter compensation module include ambient temperature, ambient humidity, and ambient pressure. The environmental parameters are used to construct a parameter compensation model to correct the acoustic resonance spectrum features constructed by the multi-dimensional feature extraction module and the generated medium disturbance field wave entropy features in real time.

6. A pressure monitoring system according to claim 1, characterized in that, The fusion intelligent analysis engine uses deep learning to learn based on the fused acoustic resonance spectrum features, medium disturbance field wave entropy features, and environmental parameters, and constructs nonlinear dynamic correlations and complementary information between the features under different working conditions.

7. A pressure monitoring system according to claim 6, characterized in that, The fusion intelligent analysis engine, based on the constructed nonlinear dynamic correlation, analyzes the subtle changing trends of the fused multi-dimensional features, constructs prediction results of the future pressure value of the medium and the pressure drop rate, and the prediction results are used to generate visual charts or trend curves to display the pressure change trend. The fusion intelligent analysis engine identifies abnormal patterns of microscopic disturbances in the medium based on the constructed complementary information, and constructs abnormal identification information. The abnormal patterns include initial microcracks, slight valve leakage, and changes in medium composition.

8. A pressure monitoring system according to claim 1, characterized in that, The distributed intelligent early warning module includes: Edge computing units are used to process and intelligently judge the prediction results or anomalies identified by the fusion intelligent analysis engine in real time, and filter out key early warning events that need to be reported immediately. The wireless communication module is used to trigger event-driven data transmission based on the key early warning event, sending the early warning information to a local display, industrial human-machine interface or central monitoring platform, triggering audible and visual alarms or SMS notifications.

9. A pressure monitoring method, applied to a pressure monitoring system according to any one of claims 1-8, characterized in that, Includes the following steps: The acoustic resonance information and optical disturbance field image sequence of the medium under test are acquired non-contactly through the multimodal sensing module. Based on the acquired acoustic resonance information, acoustic resonance spectrum features are constructed. Based on the acquired optical disturbance field image sequence, wave entropy features of the medium disturbance field are generated. Environmental parameters are acquired to construct a parameter compensation model, and the constructed acoustic resonance spectrum characteristics and the generated medium disturbance field wave entropy characteristics are corrected. The acoustic resonance spectrum features, medium disturbance field wave entropy features, and environmental parameters are fused and input into the fusion intelligent analytical model for deep learning and pattern recognition, thereby constructing nonlinear dynamic correlation and complementary information between the features under different working conditions. Based on the nonlinear dynamic correlation and complementary information learned by the fusion intelligent analytical model, the subtle change trend of the fused multi-dimensional features is analyzed, the prediction result of the future pressure value of the medium is constructed, and the abnormal patterns of micro-disturbances in the medium are identified to construct anomaly identification information. Based on the constructed prediction results or anomaly identification information, local intelligent early warning and information transmission are carried out.

Images