Industrial equipment intelligent operation and maintenance system based on multi-source information fusion

Abstract

The invention belongs to the technical field of industrial operation and maintenance, and particularly relates to an industrial equipment intelligent operation and maintenance system based on multi-source information fusion, and the system comprises a multi-source information sensing module which is used for collecting physical quantity data and technological parameters of related equipment of a petrochemical engineering site in real time; the local data processing module is used for preprocessing the multi-source data acquired by the multi-source information sensing module and uploading the processed multi-source data; wherein the preprocessing operation comprises missing data completion, data normalization and abnormal data detection; the cloud server is provided with a data fusion module and a fault diagnosis module; the data fusion module is used for carrying out space-time synchronization on all data so as to associate data with consistent time features and / or space features, carrying out weighted summation on the associated data, and fusing to generate a high-dimensional feature vector; and the fault diagnosis module is used for diagnosing the fault type of the equipment according to the high-dimensional feature vector, and triggering a corresponding alarm strategy according to a diagnosis result and an anti-explosion rule of a petrochemical engineering site.

Description

Technical Field

[0001] This invention belongs to the field of industrial operation and maintenance technology, specifically relating to an intelligent operation and maintenance system for industrial equipment based on multi-source information fusion. Background Technology

[0002] As industrial equipment becomes larger, more complex, and more clustered, its operation and maintenance level directly determines industrial production efficiency and safety margins. Especially in the petrochemical industry, related equipment is often in flammable and explosive hazardous environments. Traditional operation and maintenance methods, in addition to having fundamental limitations, also face many special challenges in monitoring equipment in explosion-proof areas, such as fragmented data perception, extensive data fusion, and delayed diagnosis and early warning.

[0003] Therefore, there is an urgent need to design an intelligent operation and maintenance system that can be applied to the entire closed-loop process of "sensing-fusion-diagnosis-decision" in the petrochemical field, and to achieve in-depth fusion and analysis of sensing data while meeting the explosion-proof safety requirements of petrochemical sites. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide an intelligent operation and maintenance system for industrial equipment based on multimodal feature coupling and evidence theory reasoning, which achieves full-link intelligence through a three-level collaborative architecture of perception layer, edge layer and cloud layer, and makes adaptive design for the special requirements of explosion-proof areas in petrochemical sites.

[0005] This invention provides an intelligent operation and maintenance system for industrial equipment based on multi-source information fusion, comprising: a multi-source information sensing module for real-time acquisition of physical quantity data and process parameters of relevant equipment at the petrochemical site; a local data processing module for preprocessing the multi-source data acquired by the multi-source information sensing module and uploading the processed multi-source data; wherein, the preprocessing operations include missing data completion, data normalization, and abnormal data detection; and a cloud server equipped with a data fusion module and a fault diagnosis module; the data fusion module is used to perform spatiotemporal synchronization of all data to associate data with consistent temporal and / or spatial characteristics, and to perform weighted summation on the associated data to generate a high-dimensional feature vector; the fault diagnosis module is used to diagnose the fault type of the equipment based on the high-dimensional feature vector, and to trigger corresponding alarm strategies based on the diagnosis results and the explosion-proof rules of the petrochemical site.

[0006] In one embodiment of the present invention, the multi-source information sensing module includes: a mechanical condition monitoring unit for collecting triaxial acceleration of equipment vibration and transient stress waves released from crack tips of important components within the equipment under application; an electrical parameter monitoring unit for collecting three-phase current harmonics of motor windings within the equipment and shaft voltage and shaft current on the motor within the equipment; a thermal condition monitoring unit for collecting temperature field data of the equipment; and a data interface unit for connecting to the DCS system at the petrochemical site to obtain process parameters of the equipment, including motor speed, motor load rate, and medium flow rate.

[0007] In one embodiment of the present invention, the local data processing module uses a time series prediction model based on a bidirectional LSTM neural network to dynamically complete the multi-source data transmitted by the multi-source information sensing module at each moment.

[0008] In one embodiment of the present invention, the local data processing module uses a hybrid Min-Max and Z-Score algorithm to normalize the completed multi-source data.

[0009] In one embodiment of the present invention, the local data processing module uses a device normal state hypersphere model built based on One-Class SVM to denoise the normalized multi-source data, and uses the isolated forest algorithm to identify and filter abnormal data in the data.

[0010] In one embodiment of the present invention, the local data processing module is further configured to periodically upload locally accumulated fault sample data to the cloud server for training of the model of the fault diagnosis module; wherein, the local data processing module is further configured to perform sample augmentation processing on the locally accumulated fault sample data.

[0011] In one embodiment of the present invention, the local data processing module uses a conditional generation network to expand the number of fault sample data.

[0012] In one embodiment of the present invention, the data fusion module is used to associate data with consistent time characteristics based on the GPS timestamp of the data, and associate data with consistent spatial characteristics based on the coordinates of the data mapped to the three-dimensional model of the device, and to establish an index for the associated data.

[0013] In one embodiment of the present invention, the data fusion module extracts corresponding features from the associated data one by one through a multi-branch convolutional neural network model and SEBlock attention mechanism, and fuses multiple features in a weighted sum manner to generate a high-dimensional feature vector.

[0014] In one embodiment of the present invention, the fault diagnosis module is configured with multiple different diagnostic models, and the fault diagnosis module uses DS evidence theory to fuse the confidence levels of the diagnostic results output by multiple diagnostic models for the same high-dimensional feature vector, and triggers the corresponding alarm strategy based on the fused confidence level and the explosion-proof rules of the petrochemical site.

[0015] The beneficial effects of this invention are as follows: This invention provides an intelligent operation and maintenance system for industrial equipment based on a "perception layer-edge layer-cloud layer" system architecture. It performs in-depth fusion analysis of multi-source data information based on a three-level fusion mechanism of "spatiotemporal-feature-decision" to diagnose the fault conditions of relevant equipment in petrochemical sites. Combined with explosion-proof rules, it triggers corresponding alarm strategies to achieve early warning and greatly improves the accuracy and reliability of fault diagnosis, dynamically maintaining the operational safety of relevant equipment.

[0016] This invention supports rapid deployment of different equipment (such as motors, transformers, and compressors). Through model transfer learning, it can quickly adapt to the equipment monitoring needs of different explosion-proof areas in the petrochemical industry. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0018] Figure 1 This is a schematic diagram of the structure of an intelligent operation and maintenance system for industrial equipment based on multi-source information fusion, provided in one embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the acquisition of multi-source data from petrochemical field-related equipment according to an embodiment of the present invention. Detailed Implementation

[0019] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0021] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0022] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0023] Example 1 Please see Figure 1 As shown, an intelligent operation and maintenance system for industrial equipment based on multi-source information fusion is particularly suitable for the intelligent operation and maintenance of the entire life cycle of equipment such as high-voltage motors, transformers, compressors, fans, and generator sets deployed in explosion-proof and non-explosion-proof areas in the petrochemical industry. It can accurately adapt to the management scenarios of key equipment involving flammable and explosive media in refining, petrochemical, and other industrial processes, meeting the special safety monitoring needs of explosion-proof areas. The system specifically includes: The multi-source information sensing module 10, serving as the sensing layer of the system architecture, is constructed as a multi-dimensional, three-dimensional sensing network capable of real-time acquisition of physical quantity data and process parameters of relevant equipment at the petrochemical site. Specifically, in practical applications, the multi-source information sensing module 10 can be configured with other data detection devices such as the mechanical condition monitoring unit 11, electrical parameter monitoring unit 12, and thermal condition monitoring unit 12, and deployed at the petrochemical site to monitor the operating status of equipment from multiple dimensions. The mechanical condition monitoring unit 11 can be configured accordingly: The triaxial accelerometer can adopt an Exia intrinsically safe design and be installed in a three-dimensional orthogonal layout on the vibration-sensitive points of the equipment, such as the bearing housing or coupling of the equipment, to collect the triaxial acceleration of the equipment vibration, thereby obtaining the vibration characteristics of the equipment in three dimensions (X / Y / Z axes, corresponding to radial, axial and tangential directions).

[0024] Correspondingly, time-domain features (such as peak value and kurtosis) and frequency-domain features (such as first harmonic amplitude and harmonic components) can be further extracted from the data signals collected by the triaxial accelerometer to identify fault conditions such as the inner / outer ring / rolling element of the equipment bearing, or mechanical abnormalities such as the imbalance of the equipment motor.

[0025] Acoustic emission sensors, by capturing transient stress waves from the propagation of material cracks, can be used to sense the damage status of critical components within equipment. They can monitor in real time the initiation and propagation of microcracks in components made of metallic materials (such as bearings, rotors, and pipes) under stress, thereby capturing the transient stress waves released from the crack tips. Examples include bearing cracks (by identifying micron-level cracks in bearing raceways and rolling elements, capturing early fatigue cracks that are difficult for vibration sensors to detect), rotor / shaft cracks (by diagnosing crack propagation in rotating components such as motor rotors and compressor shafts, preventing sudden fracture accidents caused by cracks), equipment shell / pipeline cracks (by monitoring crack initiation in equipment pressure vessels and conveying pipelines, preventing safety accidents caused by leaks of flammable and explosive media), and material fatigue damage (by identifying fatigue damage caused by long-term alternating loads in transmission components such as gears and couplings, and combining stress wave energy characteristics to determine the degree of damage, including crack length and propagation rate).

[0026] Understandably, by sensing the macroscopic manifestations of microscopic damage such as lattice slip and dislocation movement within materials through acoustic emission sensors, dynamic monitoring of early equipment damage can be achieved. In petrochemical explosion-proof scenarios, it is possible to provide early warning of crack propagation trends in equipment materials more than 72 hours in advance. Furthermore, it is possible to combine the vibration characteristics of the equipment for in-depth fusion analysis, thereby further improving the accuracy of fault diagnosis.

[0027] The electrical parameter monitoring unit 12 can be configured accordingly: Rogowski coil current sensors are mainly used to collect the three-phase current harmonics of motor windings in equipment. For example, when a short circuit occurs between turns or between phases in the motor windings, the current harmonic distortion rate increases significantly, especially the 3rd and 5th harmonic components.

[0028] Correspondingly, the harmonic characteristics of the three-phase current signals (such as the 50Hz fundamental wave and the 100Hz second harmonic) can be further extracted from the Rogowski coil current sensor to construct a harmonic spectrum, which can then be compared with the reference of the normal state of the equipment to identify abnormal components.

[0029] Specifically, in practical applications, the ExnA sparkless Rogowski coil sensor can be used to meet the safety requirements of explosion-proof areas in petrochemical sites. It can monitor the current harmonics of high-voltage motors (such as 10kV motor groups) in real time, avoiding the risk of explosion caused by electrical faults.

[0030] It is also understandable that the characteristics of current harmonics can be combined with the vibration characteristics of the equipment for in-depth analysis to accurately identify the faults or abnormal states of the equipment.

[0031] Hall effect sensors are used to monitor shaft voltage and shaft current on the motor of equipment. It's important to note that abnormal shaft voltage can cause the bearing oil film to break down, creating shaft current and leading to bearing galvanic corrosion. Furthermore, specific frequency components (such as 100Hz and 200Hz) will appear in the current harmonics at this time.

[0032] Correspondingly, by analyzing the shaft voltage and shaft current collected by the Hall sensor in conjunction with the current harmonic characteristics, the electrical corrosion of the bearing can be diagnosed.

[0033] The thermal condition monitoring unit 13 can be configured accordingly: Distributed fiber optic temperature measurement systems and infrared thermal imagers diagnose overheating faults, insulation degradation, and some mechanical anomalies in equipment by monitoring the temperature field of distributed monitoring devices. For example, when a short circuit occurs between turns or the insulation ages in the motor windings, the increased local resistance leads to an abnormal temperature rise (e.g., exceeding 155°C). Alternatively, when the silicon steel sheets in the iron core are poorly stacked or the hysteresis loss is abnormal, the temperature field of the iron core will show a non-uniform distribution (e.g., local temperature difference > 10°C). By extracting temperature gradient features from the collected temperature field data, such faults can be identified. Further, in-depth fusion analysis can be performed by combining vibration features to improve the accuracy of diagnosis. Or, due to insufficient bearing lubrication or raceway wear, frictional heat is generated, causing the bearing housing temperature to rise continuously. Combining vibration features (e.g., large kurtosis) can diagnose early bearing faults (e.g., microcracks). And when uneven air gap between the rotor and stator causes friction, the corresponding temperature field shows local high temperature. Infrared thermal imagers can capture abnormal surface temperature distribution, and fusion analysis can be performed by combining current harmonic features (e.g., 100Hz component) to identify rotor eccentricity.

[0034] Therefore, the distributed fiber optic temperature measurement system uses explosion-proof armored optical cables laid along the windings and iron cores to accurately locate hot spots and monitor the temperature field distribution in real time. Meanwhile, the infrared thermal imager is used to acquire the surface temperature spectrum of the equipment. By combining the data collected by both, the temperature gradient characteristics of the equipment can be further extracted.

[0035] The multi-source information sensing module 10 also includes a data interface unit 14, which connects to the DCS system via the OPCUA protocol to obtain the equipment's process parameters. It is understood that the DCS (Distributed Control System) is the core system of industrial process control. Its data originates from various process sensors deployed on-site (such as flow transmitters, speed encoders, load sensors, etc.), and it is responsible for real-time acquisition, control, and storage of the equipment's process parameters. Accordingly, it acts as a relay hub, providing processed real-time data to the system provided in this embodiment via the OPCUA protocol.

[0036] Specifically, in practical applications, the data interface unit 14 can acquire parameters such as the motor speed, motor load rate, and medium flow rate of the equipment through the DCS. Then, it can perform cross-modal correlation with the physical quantity data of the equipment, such as vibration, temperature, and current, to deeply integrate and analyze whether the equipment has a fault and the type of fault. For example, the load rate (0-100%) can be correlated with the distortion rate of current harmonics to identify electrical faults of the motor under different operating conditions (such as the harmonic characteristics of winding short circuits are more obvious under high load); the medium flow rate (0-1000m³ / h) is coupled with the temperature field of the bearing to diagnose mechanical overload faults of the compressor caused by abnormal flow, and so on.

[0037] Therefore, the process parameters of the equipment can be used to construct operating condition labels for the equipment (such as "high load-low flow" condition), and combined with the physical quantity data of the equipment for comprehensive diagnosis.

[0038] It should be noted that most equipment in petrochemical sites includes motors. Therefore, this embodiment focuses on the fault status of the motors within the equipment. The monitoring units configured in the multi-source information sensing module 10 are all arranged around the motors of the equipment, as shown in the following figure. Figure 2 As shown, this is not intended to limit the application scope of the system architecture of this application. Adaptive adjustments can be made for different industrial equipment in petrochemical sites. No further restrictions are imposed on this. Modifications and refinements made by those skilled in the art to the embodiments of this invention without departing from the spirit of this invention still fall within the scope of the invention application patent of this invention.

[0039] The local data processing module 20, as the edge layer of the system architecture, is used to preprocess the multi-source data collected by the multi-source information sensing module 10, including missing data completion, data normalization, and abnormal data detection, and is configured on the terminal of the device (such as the control cabinet).

[0040] Specifically, for missing data completion, a time series prediction model based on a bidirectional LSTM neural network (e.g., 128 hidden layer nodes and 500 iterations) can be used to dynamically complete the multi-source data transmitted by the multi-source information perception module 10 at each moment, with a particular optimization of the data recovery algorithm under network fluctuations.

[0041] For data normalization, a hybrid algorithm combining Min-Max (suitable for normalization of bounded data such as temperature and load) and Z-Score (suitable for normalization of unbounded data such as vibration and current) can be used to map multi-source data to the interval [-1, 1], eliminating the dimensional differences between different data types and ensuring data format uniformity.

[0042] Regarding abnormal data detection, a hyperspherical model of the normal state of the equipment can be constructed based on One-Class SVM (such as RBF kernel function, penalty coefficient C=0.8) to denoise the multi-source data, and the isolated forest algorithm (100 trees) can be used to identify and filter the abnormal data, effectively identifying the abnormal data generated by the multi-source information sensing module 10 due to environmental interference.

[0043] It should be noted that the data preprocessing operations provided by the local data processing module 20 include, but are not limited to, those described above. Modifications and refinements made by those skilled in the art to the embodiments of the present invention without departing from the spirit of the present invention still fall within the scope of the invention application patent of the present invention.

[0044] The cloud server 30, serving as the cloud layer of the system architecture, can be deployed on a private enterprise cloud platform, acting as an intelligent analysis and decision-making hub, and communicating securely and isolated from the edge layer via a gateway. The cloud server 30 innovatively performs a three-level fusion of "spatiotemporal-feature-decision" data uploaded by the local data processing module 20, thereby deeply analyzing the operational status of relevant equipment in the petrochemical explosion-proof area to comprehensively diagnose whether equipment malfunctions and the type of malfunction.

[0045] Specifically, the cloud server 30 is equipped with a data fusion module 31 and a fault diagnosis module 32. The data fusion module 31 can be used to perform spatiotemporal synchronization on all data, thereby associating data with consistent temporal and / or spatial characteristics. After extracting the corresponding features from the associated data, it performs weighted summation and fusion to generate a high-dimensional feature vector, so as to build a data foundation through a two-level fusion mechanism of "spatiotemporal-feature".

[0046] In one specific embodiment, spatiotemporal fusion involves registering GPS timestamps (±1ms accuracy) with the coordinates of the device's 3D model. This maps multi-source sensor data (such as vibration, temperature, current, etc.) into a "time-space-physical quantity" correlation matrix. For example, the vibration characteristics of the device's bearing housing, i.e., the X / Y / Z axis acceleration data collected by a triaxial accelerometer, are mapped to the coordinate points of the bearing housing's 3D model (which can be constructed using SolidWorks). A spatial index is then established with the temperature characteristics collected by the temperature measurement points of the distributed fiber optic temperature measurement system, forming a "location-physical quantity-time" cubic data structure. This allows for precise location of the fault, solving the spatiotemporal misalignment problem of multi-source data and providing standardized data input for spatiotemporal alignment for feature fusion.

[0047] This can be understood as the spatiotemporal fusion mechanism meaning that all different types of data follow the underlying logic of "GPS timestamp synchronization + 3D model coordinate registration". For example, the X / Y / Z axis data collected by the triaxial sensor for vibration characteristics corresponds to the radial, axial, and tangential spatial directions of the bearing housing. The temperature measurement points arranged by the distributed fiber optic temperature measurement system cover the winding, iron core, and other locations with a spatial resolution of 0.1m. The spatial index construction of the two is the standard case of a "position-physical quantity-time" cube. The spatial positioning of the acoustic emission sensor needs to be mapped to the 3D coordinates of the equipment housing to form cross-validation with the spatial points of the triaxial accelerometer for vibration characteristics. The data collected by the current sensor needs to be mapped to the electrical connection position of the motor stator through the 3D model of the equipment to establish a spatial correlation with the winding temperature characteristics. Although process parameters (such as load rate) do not have physical spatial coordinates, they can also form an implicit spatiotemporal mapping of "time-condition-physical quantity" by associating them with the spatial position of the equipment's operating status (such as the bearing position under high load conditions) through timestamps, ensuring that multi-source data are fused under a unified spatiotemporal reference.

[0048] Secondly, feature fusion is achieved by using a multi-branch convolutional neural network model and the SEBlock attention mechanism to further extract core features from cross-modal data such as vibration features, electrical features, and temperature features, and then coupling them to generate a high-dimensional feature vector. This allows the spatiotemporally fused data to be abstracted into the core features of the equipment state, providing a reasonable "knowledge carrier" for subsequent decision fusion.

[0049] Specifically, for data characterized by vibration, time-frequency domain features can be extracted, for example, through 5-layer convolution; for data characterized by electrical features, current harmonic features can be extracted, for example, through wavelet transform and convolutional layers; for data characterized by temperature features, spatial gradient features can be extracted, for example, through spatial convolution. Furthermore, industrial parameters can be mapped to operating conditions through fully connected layers, such as load rate and normalized rotational speed.

[0050] Correspondingly, a weighted summation can be performed based on the aforementioned multiple features. That is, the features of each branch are fused through the modal interaction layer to obtain the final high-dimensional feature vector, achieving deep coupling of cross-modal data and significantly improving the data coupling degree compared to traditional data concatenation methods. Specifically, a SE Block channel attention mechanism is introduced to achieve dynamic weight allocation for each branch feature, and the calculation formula is as follows: , This represents the weight of the feature in the i-th branch. This represents the Sigmoid activation function. and Indicates the parameters of the fully connected layer. This represents the global average pooling result for the i-th branch feature.

[0051] It should be noted that in this embodiment, only the above-mentioned monitoring units are provided for reference regarding the sensing layer, and the corresponding branch features are only taken as examples of vibration, electrical, temperature, and process parameters. However, this is not intended to limit the number of branches of feature data, and they can all be fused according to the above method. Modifications and refinements made by those skilled in the art to the embodiments of the present invention without departing from the spirit of the present invention still fall within the scope of the invention application patent of the present invention.

[0052] Finally, the fault diagnosis module 32 constructs a confidence propagation network based on DS evidence theory. It inputs the high-dimensional feature vectors output by the data fusion module 31 into various diagnostic models (such as ResNet-50, DenseNet, and XGBoost), and uses the diagnostic results of each model as evidence. The fusion confidence of each diagnostic result is calculated using the Basic Probability Assignment (BPA) function. Simultaneously, to improve interpretability, it combines explosion-proof rules from petrochemical industry standards to trigger corresponding alarm strategies, forming a dual-insurance mechanism of "data-driven + rule-driven."

[0053] The alarm strategies are as follows: Level I (Emergency Shutdown): When a fatal fault is detected (such as the temperature of the equipment motor winding ≥155℃, or the triaxial acceleration of vibration ≥50g), the ESD system is triggered (response time ≤100ms), which in turn activates the emergency shut-off valve and inert gas protection device in the petrochemical explosion-proof area, cuts off the equipment power supply and triggers the production system switchover.

[0054] Level II (Immediate Repair): Emergency work orders can be pushed via mobile terminal, including fault location (accuracy ±0.5m), cause analysis (e.g., "bearing outer ring fault, confidence level 96%)", repair steps (including tool list), and specific safety measures and hot work permit requirements for repairs in explosion-proof areas.

[0055] Level III (Planned Maintenance): Based on the equipment's current health index HI (Health Index) The system automatically generates maintenance window suggestions (such as "4 hours of downtime within the next 72 hours") in conjunction with production plans, and optimizes maintenance windows based on equipment start-up and shutdown cycles.

[0056] Level IV (Continuous Monitoring): Include the equipment in the key list, increase the monitoring frequency, such as once per minute, and generate corresponding status trend curves (e.g., a daily temperature increase rate ≥ 0.5℃ requires special attention).

[0057] Based on the above alarm strategies, single-type data can directly determine the warning level based on the fault diagnosis results. For example, a bearing inner ring fault with a confidence level of 97.2% triggers a Level II warning, requiring emergency repairs. Multi-source data is analyzed through fusion to determine the warning level. For example, in the joint diagnosis of vibration, temperature, and electrical faults, if the vibration diagnosis indicates a bearing fault with a confidence level of 92%, but the temperature and current are normal, it may be downgraded to a Level III warning (planned maintenance). If all three indicate the same fault location, it is upgraded to a Level II warning (immediate repair). Furthermore, explosion-proof rules from the petrochemical industry can be embedded to determine the warning level. For example, even if the diagnostic confidence level is <80%, a temperature >155℃ (GB50058 threshold) still triggers a Level I warning.

[0058] In one specific embodiment, when early cracks appear in the motor bearings of equipment in the explosion-proof area of ​​a petrochemical site: The data fusion module 31, based on spatiotemporal fusion, locates the vibration features (X-axis acceleration data collected by a triaxial accelerometer) and the temperature features collected by the temperature measurement points of the distributed fiber optic temperature measurement system at the bearing seat position in the 3D model, achieving spatial synchronization. Simultaneously, it uses GPS timestamps to ensure that the vibration features (10kHz sampling) and temperature features (second-level sampling) are aligned on the time axis, achieving time synchronization. Next, it uses MB-CNN to extract kurtosis features (bearing fault-sensitive features) from the vibration features and gradient features (friction heating features) from the temperature features, dynamically assigning corresponding weights through SEBlock (e.g., vibration feature weight 0.6, temperature feature weight 0.3), and fusing them to generate a high-dimensional feature vector.

[0059] The fault diagnosis module 32 uses the high-dimensional feature vector to diagnose equipment faults through its configured diagnostic models ResNet-50 and DenseNet. It assumes that the confidence level of the ResNet-50 diagnostic results is 92% and that of DenseNet is 91%. Based on the DS evidence theory, it calculates the fusion confidence level (≥0.85) and then superimposes the rule in GB 50156 that "bearing temperature > 90℃ requires emergency monitoring" to trigger a Level III warning and generate a dynamic maintenance plan (such as shutdown for maintenance within 48 hours).

[0060] In another specific embodiment, taking "temperature > 155℃ triggering Level I warning" as an example, the fault diagnosis module 32 identifies winding overheating through the DenseNet model with a confidence level of 85%. The fault diagnosis module 32 then integrates the results of other models based on DS evidence theory, such as ResNet-50 diagnosing insulation degradation with a confidence level of 88%, and the corresponding calculated fusion confidence level is 90%. At the same time, since the temperature = 158℃ > 155℃, it triggers the Level I warning threshold in the GB50058 rule base. The fault diagnosis module 32 can directly respond to the Level I warning to trigger the ESD system (response time ≤ 100ms), link up to cut off the equipment power supply, and generate an emergency work order, including the fault location (±0.5m), safety protection measures (such as the hot work permit process in explosion-proof areas), etc.

[0061] Therefore, all types of data (physical quantities, process parameters) need to undergo three-level fusion processing: spatiotemporal synchronization, feature extraction, and decision fusion, to ensure the complete transformation from data to knowledge, which greatly improves the accuracy and reliability of equipment diagnosis.

[0062] Furthermore, the local data processing module 20 is also used to periodically upload locally accumulated fault sample data to the cloud server 30 for training of the diagnostic model configured by the fault diagnosis module 32. The local data processing module 20 can also perform sample augmentation processing on the locally accumulated fault sample data.

[0063] Understandably, offline expansion can solve the problem of sample scarcity by expanding historically accumulated fault samples (such as bearing fault data from the past year), for example, expanding the original 50 samples to 500.

[0064] Specifically, in practical applications, the perception layer connects to the SCADA system via the OPCUA protocol (historical data storage period ≥ 1 year) to obtain the equipment's operation logs and alarm records, which are then transmitted to the local data processing module 20. The local data processing module 20 can then filter out labeled fault samples (such as vibration and current data corresponding to historical alarms) and, in conjunction with equipment maintenance records (such as fault type labels in maintenance work orders), push the data to on-site maintenance personnel for manual labeling of historical data, forming a correspondence between "fault type" and "feature data" (such as a sample pair of "bearing outer ring fault - vibration kurtosis = 9.2"). Therefore, the diagnostic model configured in the fault diagnosis module 32, such as the ResNet-50 network, can be pre-trained using this fault sample data, enabling the model to better handle anomalies in real-time data. For example, when real-time data exhibits features similar to historical faults (such as kurtosis = 8.5), the model can improve its recognition accuracy based on the expanded sample library.

[0065] To address this, the local data processing module 20 utilizes the Conditional Generative Network (CGAN) for sample augmentation to expand the quantity of fault sample data. The prerequisite for CGAN to generate new samples is the existence of a "small number of real fault samples," which must come from historical data accumulation. This necessitates the local data processing module 20 storing the corresponding fault samples.

[0066] In summary, this invention provides an intelligent operation and maintenance system for industrial equipment based on a "perception layer-edge layer-cloud layer" system architecture. It performs deep fusion analysis of multi-source data information based on a three-level fusion mechanism of "spatiotemporal-feature-decision" to diagnose the fault conditions of relevant equipment in petrochemical sites. Combined with explosion-proof rules, it triggers corresponding alarm strategies to achieve early warning and greatly improves the accuracy and reliability of fault diagnosis, dynamically maintaining the operational safety of relevant equipment.

[0067] This invention supports rapid deployment of different equipment (such as motors, transformers, and compressors). Through model transfer learning, it can quickly adapt to the equipment monitoring needs of different explosion-proof areas in the petrochemical industry.

[0068] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

[0069] Throughout this specification, the terms "an embodiment," "embodiment," or "specific embodiment" refer to a particular feature, structure, or characteristic described in connection with an embodiment that is included in at least one embodiment of the invention, but not necessarily in all embodiments. Therefore, the various representations of the phrases "in one embodiment," "in an embodiment," or "in a specific embodiment" in different places throughout the specification do not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic of any specific embodiment of the invention can be combined with one or more other embodiments in any suitable manner. It should be understood that other variations and modifications of the embodiments of the invention described and illustrated herein may be based on the teachings herein and will be considered part of the spirit and scope of the invention.

[0070] Furthermore, unless otherwise expressly stated, any arrows in the accompanying drawings should be considered illustrative only and not limiting. Additionally, unless otherwise stated, the term "or" as used herein is generally intended to mean "and / or". Where a term is anticipated to provide a separation or combination capability that is unclear, a combination of components or steps will also be considered as indicated.

[0071] As used herein and throughout the claims below, unless otherwise specified, “a” and “the” include the plural references. Similarly, as used herein and throughout the claims below, unless otherwise specified, “in” means “in” and “on”.

[0072] The above description of the embodiments shown in this invention (including the content set forth in the abstract of the specification) is not intended to be an exhaustive enumeration or to limit the invention to the precise forms disclosed herein. Although specific embodiments and examples of the invention have been described herein for illustrative purposes only, various equivalent modifications are possible within the spirit and scope of the invention, as will be recognized and understood by those skilled in the art. As indicated, these modifications can be made to the invention in accordance with the above description of the embodiments described herein, and such modifications will be within the spirit and scope of the invention.

[0073] This document has generally described the systems and methods in detail to aid in understanding the invention. Furthermore, various specific details have been set forth to provide a general understanding of embodiments of the invention. However, those skilled in the art will recognize that embodiments of the invention can be practiced without one or more specific details, or using other means, systems, accessories, methods, components, materials, parts, etc. In other instances, well-known structures, materials, and / or operations have not been specifically shown or described in detail to avoid obscuring aspects of embodiments of the invention.

[0074] Therefore, although the invention has been described herein with reference to specific embodiments thereof, freedom of modification, various changes and substitutions are also within the scope of the foregoing disclosure, and it should be understood that in some cases, certain features of the invention may be adopted without departing from the scope and spirit of the invention and without corresponding use of other features. Thus, many modifications can be made to adapt a particular environment or material to the essential scope and spirit of the invention. The invention is not intended to be limited to the specific terminology used in the following claims and / or the specific embodiments disclosed as the best mode for carrying out the invention, but the invention will include any and all embodiments and equivalents falling within the scope of the appended claims. Therefore, the scope of the invention will be defined only by the appended claims.

Claims

1. An industrial equipment intelligent operation and maintenance system based on multi-source information fusion, characterized in that, include: The multi-source information sensing module is used to collect physical quantity data and process parameters of relevant equipment at the petrochemical site in real time; The local data processing module is used to preprocess the multi-source data collected by the multi-source information sensing module and upload the processed multi-source data; wherein, the preprocessing operations include missing data completion, data normalization, and abnormal data detection; The cloud server is equipped with a data fusion module and a fault diagnosis module. The data fusion module is used to perform spatiotemporal synchronization of all data, to associate data with consistent temporal and / or spatial characteristics, and to perform weighted summation on the associated data to generate a high-dimensional feature vector. The fault diagnosis module is used to diagnose the fault type of the equipment based on the high-dimensional feature vector, and to trigger the corresponding alarm strategy based on the diagnosis results and the explosion-proof rules of the petrochemical site. 2.The industrial equipment intelligent operation and maintenance system based on multi-source information fusion according to claim 1, characterized in that, The multi-source information sensing module includes: The mechanical condition monitoring unit is used to collect triaxial acceleration of equipment vibration and transient stress waves released from crack tips of important components inside the equipment under application. The electrical parameter monitoring unit is used to collect the three-phase current harmonics of the motor windings inside the equipment, as well as the shaft voltage and shaft current of the motor inside the equipment. Thermal condition monitoring unit, used to collect temperature field data of the equipment; The data interface unit is used to connect to the DCS system at the petrochemical site to obtain the process parameters of the equipment, including the motor speed, motor load rate, and medium flow rate. 3.The industrial equipment intelligent operation and maintenance system based on multi-source information fusion of claim 1, characterized in that, The local data processing module uses a time series prediction model based on a bidirectional LSTM neural network to dynamically complete the multi-source data transmitted by the multi-source information perception module at each moment. 4.The industrial equipment intelligent operation and maintenance system based on multi-source information fusion of claim 1, characterized in that, The local data processing module uses a hybrid Min-Max and Z-Score algorithm to normalize the completed multi-source data. 5.The industrial equipment intelligent operation and maintenance system based on multi-source information fusion according to claim 1, characterized in that, The local data processing module uses a device normal state hypersphere model built on One-Class SVM to denoise the normalized multi-source data, and uses the isolated forest algorithm to identify and filter abnormal data. 6.The industrial equipment intelligent operation and maintenance system based on multi-source information fusion according to claim 1, characterized in that, The local data processing module is also used to periodically upload the locally accumulated fault sample data to the cloud server for training of the diagnostic model of the fault diagnosis module. The local data processing module is also used to perform sample augmentation processing on the locally accumulated fault sample data.

7. The industrial equipment intelligent operation and maintenance system based on multi-source information fusion according to claim 6, characterized in that, The local data processing module uses a conditional generation network to expand the number of fault sample data. 8.The industrial equipment intelligent operation and maintenance system based on multi-source information fusion of claim 1, characterized in that, The data fusion module is used to associate data with consistent time characteristics based on the GPS timestamp of the data, and to associate data with consistent spatial characteristics based on the coordinates mapped to the device's 3D model of the data, and to create an index for the associated data. 9.The industrial equipment intelligent operation and maintenance system based on multi-source information fusion of claim 1, characterized in that, The data fusion module extracts corresponding features from the associated data one by one through a multi-branch convolutional neural network model and SEBlock attention mechanism, and fuses multiple features in a weighted sum manner to generate a high-dimensional feature vector. 10.The industrial equipment intelligent operation and maintenance system based on multi-source information fusion of claim 1, characterized in that, The fault diagnosis module is configured with multiple different diagnosis models, and the fault diagnosis module fuses the diagnosis result confidence output by the multiple diagnosis models for the same high-dimensional feature vector by using D-S evidence theory, and triggers a corresponding alarm strategy according to the fused confidence and the explosion-proof rules of the petrochemical field.

Images