Edge-computing-based industrial equipment remote monitoring and fault early warning method and system

Abstract

This invention proposes a method and system for remote monitoring and fault early warning of industrial equipment based on edge computing. It belongs to the fields of industrial Internet of Things (IIoT), edge computing, and intelligent operation and maintenance technology. The method includes: performing multi-dimensional physical field modeling of the operating area of ​​the industrial equipment to generate spatiotemporal distribution data of the equipment's physical field; deploying a physical information neural network on edge computing nodes based on the spatiotemporal distribution data of the equipment's physical field to construct a real-time inference framework on the edge side; and introducing causal decoupling technology to accurately separate the dynamic response of equipment operating conditions from physical degradation characteristics, effectively avoiding false alarms caused by changes in operating conditions and greatly improving the accuracy of fault early warning.

Description

Technical Field

[0001] This invention proposes a method and system for remote monitoring and fault early warning of industrial equipment based on edge computing, belonging to the fields of industrial Internet of Things, edge computing and intelligent operation and maintenance technology. Background Technology

[0002] In modern complex industrial production systems, the stable operation of core equipment such as large CNC machine tools, wind turbines, and heavy extrusion presses is crucial. Once a machine fails and stops, it often results in catastrophic economic losses.

[0003] Traditional edge computing-based remote monitoring and early warning systems for industrial equipment have many problems. On the one hand, in fault diagnosis, existing edge early warning algorithms are mostly data-driven pattern matching, which makes it difficult to distinguish between "sudden changes in equipment status" and "physical degradation". When equipment changes its operating conditions due to process requirements, such as increased speed or sudden load changes, vibration and current signals change drastically. This normal dynamic response is often misjudged as an early fault, resulting in a large number of false alarms and keeping maintenance personnel busy.

[0004] On the other hand, in terms of computing power utilization, existing edge computing nodes mostly act as "data gateways," performing only simple noise reduction and downsampling before transmitting feature data or raw waveforms to the cloud for deep learning inference. This not only consumes a large amount of uplink bandwidth, but also causes warning delays or even failures when the network fluctuates, failing to leverage the real-time blocking capabilities of the edge side.

[0005] In addition, regarding remote monitoring, the monitoring center mainly receives monotonous time-series curves or red and green alarm lights, lacking an intuitive presentation of the evolution process of the internal physical field of the equipment. Maintenance personnel find it difficult to understand the cause of the alarm and can only respond passively. There is a lack of explainable tracing of the fault evolution mechanism, which makes it difficult to meet the needs of modern industry for efficient and precise operation and maintenance. Summary of the Invention

[0006] This invention provides a method and system for remote monitoring and fault early warning of industrial equipment based on edge computing, in order to solve the problems mentioned in the background art above:

[0007] The present invention proposes a method for remote monitoring and fault early warning of industrial equipment based on edge computing, the method comprising:

[0008] S1. Perform multi-dimensional physical field modeling on the operating area of ​​industrial equipment to generate spatiotemporal distribution data of the equipment's physical field; based on the spatiotemporal distribution data of the equipment's physical field, deploy a physical information neural network on edge computing nodes to build a real-time inference framework on the edge side.

[0009] S2. The multi-source sensor data during equipment operation is decoupled causally using the edge-side real-time inference framework to separate the dynamic response features of the operating conditions from the physical degradation features, generating decoupled operating condition feature data and degradation feature data; the decoupled degradation feature data is then subjected to edge-side lightweight feature enhancement processing to generate enhanced degradation feature data.

[0010] S3. Based on the enhanced degradation feature data, perform semantic encoding of equipment health status to generate equipment health semantic tokens; dynamically map the equipment health semantic tokens through the edge-side digital twin model to generate equipment physical field evolution visualization data; perform fault mode matching on the equipment physical field evolution visualization data to generate potential fault mode data.

[0011] S4. Drive the edge-side digital twin model with potential failure mode data to perform fault propagation simulation and generate fault propagation path data; render risk heatmaps of equipment physical field evolution visualization data based on fault propagation path data to generate equipment fault risk heatmap data; predict the probability of falling risk from the equipment fault risk heatmap data to generate fault falling risk prediction data.

[0012] S5. Perform multimodal fusion decision-making on the edge side based on the fault drop risk prediction data to generate a comprehensive equipment risk index; trigger a hierarchical early warning mechanism on the edge computing node based on the comprehensive equipment risk index to generate equipment fault early warning response data.

[0013] The present invention proposes an edge computing-based remote monitoring and fault early warning system for industrial equipment, the system comprising:

[0014] One or more processors;

[0015] Memory, used to store one or more programs;

[0016] Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.

[0017] The beneficial effects of this invention are as follows: By introducing causal decoupling technology, it can accurately separate the dynamic response of equipment operating conditions from physical degradation characteristics, effectively avoiding false alarms caused by changes in operating conditions and greatly improving the accuracy of fault early warning. Utilizing edge-side physical information neural networks and digital twin semantic tokenization, complex reasoning and visualization mapping can be completed at edge computing nodes, reducing dependence on cloud computing power, reducing uplink bandwidth consumption, avoiding early warning delays or even failures caused by network fluctuations, and enhancing the real-time performance and stability of the system. Simultaneously, this technology can intuitively present the evolution process of the internal physical field of the equipment, providing interpretable fault tracing evidence for maintenance personnel, and can generate hierarchical early warning responses through multimodal fusion decision-making, reducing the blindness of passive responses. When facing complex equipment operating conditions and potential faults, it effectively avoids operational decision-making errors caused by information black-boxing, significantly improving the overall efficiency of remote monitoring and fault early warning of industrial equipment. Attached Figure Description

[0018] Figure 1 This is a diagram illustrating the steps of the method described in this invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] One embodiment of the present invention, such as Figure 1 As shown, a method for remote monitoring and fault early warning of industrial equipment based on edge computing is described, the method comprising:

[0021] S1. Perform multi-dimensional physical field modeling on the operating area of ​​industrial equipment to generate spatiotemporal distribution data of the equipment's physical field; based on the spatiotemporal distribution data of the equipment's physical field, deploy a physical information neural network on edge computing nodes to build a real-time inference framework on the edge side.

[0022] S2. The multi-source sensor data during equipment operation is decoupled causally using the edge-side real-time inference framework to separate the dynamic response features of the operating conditions from the physical degradation features, generating decoupled operating condition feature data and degradation feature data; the decoupled degradation feature data is then subjected to edge-side lightweight feature enhancement processing to generate enhanced degradation feature data.

[0023] S3. Based on the enhanced degradation feature data, perform semantic encoding of equipment health status to generate equipment health semantic tokens; dynamically map the equipment health semantic tokens through the edge-side digital twin model to generate equipment physical field evolution visualization data; perform fault mode matching on the equipment physical field evolution visualization data to generate potential fault mode data.

[0024] S4. Drive the edge-side digital twin model with potential failure mode data to perform fault propagation simulation and generate fault propagation path data; render risk heatmaps of equipment physical field evolution visualization data based on fault propagation path data to generate equipment fault risk heatmap data; predict the probability of falling risk from the equipment fault risk heatmap data to generate fault falling risk prediction data.

[0025] S5. Perform multimodal fusion decision-making on the edge side based on the fault drop risk prediction data to generate a comprehensive equipment risk index; trigger a hierarchical early warning mechanism on the edge computing node based on the comprehensive equipment risk index to generate equipment fault early warning response data.

[0026] The working principle of the above technical solution is as follows: First, a comprehensive modeling of the operating area of ​​the industrial equipment is completed from spatial, temporal, and multi-type physical field levels. Parameters related to the site environment, equipment structure, and actual operating conditions are collected and integrated. After standardization and calibration, these parameters are broken down into multi-dimensional related information and converted into spatiotemporal distribution data of the equipment's physical field. Based on the volume of this data and computational consumption, edge computing nodes are divided into computing power levels and business permission levels. Then, the physical information neural network is decomposed into modules and deployed layer by layer to the corresponding edge nodes, forming a real-time edge-side inference framework capable of performing data computation and analysis locally. Second, relying on the established inference framework, multi-source sensor data streams such as equipment vibration, temperature, pressure, and rotational speed are continuously collected. After time-series alignment and noise filtering and purification, the internal feature association logic of the data is decomposed, separating the dynamic response characteristics brought by normal operating conditions from the physical degradation characteristics brought by the equipment's own aging. The degraded features obtained from the decomposition are simplified in dimensions and amplified in detail. Feature dimensions are then added and semantic expression is strengthened to complete lightweight feature enhancement on the edge side, adapting to the low-load computational needs of edge nodes. Subsequently, state thresholds, fluctuation trends, and decay gradients are extracted from the enhanced degradation features to decompose multi-level health characterization information of the equipment. Discretized semantic transformation and encoding encapsulation are then performed to uniformly generate standardized equipment health semantic tokens. An edge-side digital twin model that conforms to the physical structure and operational rules of the equipment is built. The semantic tokens are connected to the model's temporal evolution link to reconstruct the true state of the equipment's physical field over time, producing visualized evolution data. This data is then compared and matched with historical fault sample features in multiple dimensions to identify various potential fault patterns. Based on these potential fault patterns, the initial fault initiation points are located. Various core parameters of these points are extracted, standardized, and then fed into the digital twin model to deduce the transmission sequence, propagation speed, and impact range of the fault among different components of the equipment, forming complete fault propagation path data. Combining the evolution rate and hazard range of the propagation path, risk level labels are overlaid on the visualized physical field data to generate a global fault risk heatmap. The proportion and evolution cycle of high-risk areas in the heatmap are statistically analyzed to calculate the probability of sudden faults at different times, obtaining fault drop risk prediction data. Finally, information from multiple aspects, including fault risk prediction, real-time equipment operating conditions, and on-site environmental interference, is summarized and integrated into a complete set of risk assessment elements. Through weighted assignment and quantitative conversion, multimodal fusion calculation is completed at the edge side to obtain a comprehensive risk index that can intuitively measure the equipment status. Multiple risk threshold ranges are defined, and the corresponding early warning logic and on-site handling procedures are determined by matching the index level. Early warning prompts, risk details, and practical handling guidelines are synchronously pushed at the edge nodes, thereby realizing a complete closed-loop operation of all-weather remote status monitoring, early fault identification, and graded early warning for industrial equipment.

[0027] The above technical solution achieves the following results: By leveraging real-time inference and multi-dimensional modeling at the edge, it improves the real-time performance and accuracy of industrial equipment operation status monitoring, enhances the effectiveness of fault feature identification, and enables rapid detection of equipment anomalies. Simultaneously, it reduces network transmission pressure and edge node computing power consumption for remote equipment monitoring, minimizing missed or false fault diagnoses and preventing equipment downtime or safety incidents caused by the escalation of faults. It enables dynamic and visual monitoring of equipment health status and provides early fault warnings through a tiered early warning mechanism, reducing equipment maintenance costs and unplanned downtime, ensuring continuous and stable industrial production, and also reducing the workload and human error associated with manual monitoring, making equipment management more efficient and reliable.

[0028] In one embodiment of the present invention, S1 includes:

[0029] S11. Collect environmental parameters, structural parameters, and basic operating parameters of the industrial equipment operating site, and aggregate them to form a set of basic parameters for the entire equipment domain.

[0030] S12. Perform redundant data filtering and abnormal data smoothing calibration on the equipment's overall basic parameter set to generate a normalized equipment basic parameter set.

[0031] S13. Perform multi-level correlation decomposition of the standardized equipment basic parameter set in spatial, temporal and physical dimensions, complete the multi-dimensional physical field modeling of the industrial equipment operating area, and generate the spatiotemporal distribution data of the equipment physical field.

[0032] S14. Divide the computing power levels and data processing permission ranges of edge computing nodes to adapt to the computational requirements of the spatiotemporal distribution data of the physical field of the device; deploy the physical information neural network to the corresponding edge computing nodes in layers according to the computing power range, and build a real-time inference framework on the edge side.

[0033] The working principle and effects of the above technical solution are as follows: By comprehensively collecting and integrating various parameters related to equipment operation, the integrity and comprehensiveness of basic equipment data are improved, reducing modeling deviations caused by data omissions. Redundancy filtering and anomaly calibration of basic parameters reduce interference from invalid data in subsequent processing, enhance data regularity and reliability, and avoid modeling distortion and inference bias caused by data clutter. Multi-level correlation decomposition enables accurate multi-dimensional physical field modeling, improving the accuracy of physical field spatiotemporal distribution data. Layered deployment of neural networks adapts to computing power requirements, improving the operating efficiency of the edge-side inference framework and reducing computing power waste. This provides solid data support for subsequent equipment monitoring and fault early warning, ensures the real-time performance of edge-side inference, avoids processing delays caused by computing power mismatch, and lays a solid foundation for subsequent full-process operations.

[0034] In one embodiment of the present invention, S13 includes:

[0035] Extract the basic parameters of the standardized equipment to centrally represent the point information and structural layout information of the equipment spatial layout, and form the equipment spatial reference elements;

[0036] Continuous temporal sampling of equipment spatial reference elements captures dynamic changes in equipment operating status over time, generating equipment temporal evolution elements.

[0037] Integrate spatial reference elements of equipment with temporal evolution elements of equipment into the laws of physical field change, and fuse them to form multi-dimensional related decomposition basic data;

[0038] Cross-dimensional correlation and coupling analysis is carried out on the basic data of multi-dimensional correlation decomposition to complete the multi-dimensional physical field modeling of the industrial equipment operating area and generate the spatiotemporal distribution data of the equipment physical field.

[0039] The working principle and effects of the above technical solution are as follows: By accurately extracting relevant information about the spatial layout of equipment, standardized spatial benchmark elements are formed, improving the accuracy of capturing spatial features of equipment and reducing the deviation of spatial dimension data. Continuous temporal sampling of the spatial benchmark elements comprehensively captures the dynamic changes in equipment operation, enhancing the integrity of temporal data and avoiding the omission of details of equipment state changes due to discontinuous sampling. Integrating spatial and temporal elements and incorporating physical field laws reduces the fragmentation of multi-dimensional data and improves the reliability of the associated decomposition of basic data. Cross-dimensional correlation and coupling analysis makes multi-dimensional physical field modeling more closely match the actual operating scenario of the equipment, improving the authenticity of the spatiotemporal distribution data of the physical field. This provides accurate data support for subsequent edge-side inference and avoids subsequent monitoring and early warning errors caused by modeling deviations, further ensuring the overall effectiveness of remote equipment monitoring.

[0040] In one embodiment of the present invention, step S14 includes:

[0041] The volume and computational consumption of the physical field spatiotemporal distribution data of the statistical equipment are used to form statistical results of data computational load; combined with the hardware configuration of the edge nodes and the real-time idle resource status, the gradient computing power level is divided to generate a hierarchical sequence of edge node computing power.

[0042] Match the statistical results of data processing load with the computing power classification sequence of edge nodes, define the data processing permission range corresponding to each node, and complete the adaptation and matching of computing power and business load.

[0043] The network layers and computation modules of the physical information neural network are separated to create independent computation units that are adapted to different computing power levels.

[0044] The split network computing units are sequentially allocated and deployed to the corresponding edge computing nodes, and the data interaction links between nodes are integrated to build a real-time inference framework on the edge side.

[0045] The working principle and effects of the above technical solution are as follows: By statistically analyzing the computational load, data processing needs are accurately grasped. Combining the resource status of edge nodes to classify computing power levels improves the rationality of computing power allocation and reduces computing power waste and resource idleness. Matching computing power with business load and defining processing permission ranges reduces interference between different nodes, enhances the accuracy of computing power adaptation, and avoids processing delays or resource overload caused by mismatch between computing power and load. Splitting the neural network and deploying it in layers allows for precise adaptation between computing units and node computing power, improving the operating efficiency of the edge-side inference framework and accelerating data processing speed. Integrating the data interaction links between nodes ensures smooth data transmission, fully leveraging the computing power advantages of each edge node while avoiding inference lag and data loss due to unreasonable deployment, providing strong support for the real-time nature of subsequent equipment monitoring and fault early warning.

[0046] In one embodiment of the present invention, S2 includes:

[0047] S21. Deploy a full-domain sensing network for industrial equipment, continuously collect raw data streams from multiple sources of sensors, including vibration, temperature, pressure, and rotational speed, and aggregate them to form a raw dataset of multi-source sensing data for the equipment.

[0048] S22. Perform time-series alignment and noise filtering on the original dataset of multi-source sensing of the device to obtain standardized device sensing time-series data.

[0049] S23. Disassemble the internal correlation logic of the sensing time sequence data of standardized equipment, complete the layered separation of dynamic response characteristics and physical degradation characteristics under working conditions, and generate decoupled working condition characteristic data and degradation characteristic data.

[0050] S24. Perform dimensionality reduction and detail feature amplification on the decoupled degenerate feature data, adapt to the lightweight operation specifications of edge nodes, and generate intermediate degenerate feature data.

[0051] S25. Perform feature dimension completion and semantic enhancement on the intermediate state degenerate feature data, complete the edge-side lightweight feature enhancement processing, and generate enhanced degenerate feature data.

[0052] The working principle and effects of the above technical solution are as follows: By deploying a comprehensive sensing network, data from multiple sensors is continuously collected, improving the comprehensiveness and continuity of equipment operation data acquisition and reducing the omission of key operating parameters. Time-series alignment and noise filtering of raw data reduce the impact of invalid interference on data quality, enhance the purity and reliability of sensor time-series data, and avoid deviations in subsequent feature extraction due to data clutter. Layered separation of operating conditions and degraded features allows for clear distinction between the two types of features, improving the accuracy of feature recognition and reducing the risk of misjudgment caused by feature confusion. Dimensionality reduction, amplification, and enhancement of degraded features adapt to the lightweight computational needs of edge nodes, improving data processing efficiency and avoiding computational delays caused by data redundancy. This provides high-quality feature data for equipment health status assessment while also considering edge-side computational performance, providing accurate and efficient support for subsequent fault early warning.

[0053] In one embodiment of the present invention, step S23 includes:

[0054] Analyze the numerical fluctuation patterns and feature correlation strength of standardized equipment sensing time series data, and generate a time series feature correlation map;

[0055] Based on the temporal feature correlation map, the feature response intervals under different operating conditions are split to distinguish between operating condition correlation features and degradation correlation features;

[0056] Time-series trajectory extraction and feature clustering are performed on the working condition-related features to retain the core information of dynamic changes in the working conditions and generate working condition feature data.

[0057] Abnormal fluctuations and trends are filtered and extracted from degradation-related features to capture the core features of equipment performance degradation and generate degradation feature data.

[0058] Verify the independence of operating condition feature data and degradation feature data, eliminate redundant features, complete the layered stripping, and obtain decoupled operating condition feature data and degradation feature data.

[0059] The working principle and effects of the above technical solution are as follows: By analyzing the fluctuation patterns and correlation strength of sensor time-series data, a time-series feature correlation map is generated, improving the clarity of feature correlation analysis and reducing the probability of feature confusion. Separating the response intervals of operating conditions and degradation features accurately distinguishes the differences between the two types of features, enhancing the effectiveness of feature differentiation and avoiding subsequent data processing deviations caused by feature confounding. Extracting the core trajectory of operating conditions and clustering and filtering abnormal fluctuations in degradation features improves the purity of both types of feature data, retains key information, and reduces interference from invalid features. Verifying feature independence and eliminating cross-redundancy further enhances data reliability, ensuring that the decoupled feature data accurately corresponds to the equipment operating conditions and degradation states. This provides clear and pure basic data for subsequent degradation feature enhancement and avoids fault misjudgment caused by feature confusion, laying a solid data foundation for equipment health monitoring and fault early warning.

[0060] In one embodiment of the present invention, S3 includes:

[0061] S31. Extract the state change threshold, feature fluctuation trend and performance degradation gradient from the enhanced degradation feature data, and decompose the multi-level characterization information of the equipment health status.

[0062] S32. Discretize and encapsulate the multi-level representation information of equipment health status to generate equipment health status semantic tokens.

[0063] S33. Build an edge-side digital twin model that fits the physical structure and operating mechanism of the equipment to replicate the operating evolution pattern of the equipment throughout its entire life cycle;

[0064] S34. Map the device health semantic token to the temporal evolution link of the edge-side digital twin model to restore the real-time changes in the device's physical field and generate visualized data on the evolution of the device's physical field.

[0065] S35. Collect a feature library of historical fault samples of industrial equipment, compare the equipment physical field evolution visualization data with the features of historical fault samples, complete multi-dimensional fault mode matching, and generate potential fault mode data.

[0066] The working principle and effects of the above technical solution are as follows: By extracting key information from enhanced degradation features and deconstructing multi-level representations of equipment health, the comprehensiveness of equipment health status identification is improved, and the possibility of misjudgment of health status is reduced. Discretization, transformation, encoding, and encapsulation of representation information enhance the convenience of health information transmission and processing, avoiding transmission errors or parsing deviations caused by messy information formats. Building a digital twin model that closely matches the actual equipment can accurately replicate the equipment's operational evolution patterns, improving the realism of physical field evolution restoration and reducing the deviation between the model and the actual equipment. Mapping semantic tokens to the model and matching them with historical fault samples improves the accuracy of potential fault identification, avoiding the omission of early fault hazards. This solution enables both visualized monitoring of equipment health status and early detection of potential faults, buying time for subsequent fault warnings and handling, and reducing losses caused by fault escalation.

[0067] In one embodiment of the present invention, S32 includes:

[0068] The feature dimensions and numerical ranges of multi-level representation information of equipment health status are analyzed, the representation intervals of different health statuses are divided, and the health representation interval division results are generated.

[0069] The core features of each health characterization interval are discretized, and the continuous state data is converted into discretized semantic labels to generate a discretized semantic feature set.

[0070] Redundancy removal and feature fusion are performed on the discretized semantic feature set to retain the core semantic representation of health status and generate standardized semantic features.

[0071] Standardized semantic features are encoded, assigned a unique identifier, and semantic information is encapsulated and integrated to generate an initial semantic token.

[0072] Verify the integrity and accuracy of the initial semantic token, correct encoding deviations, ensure that the semantics accurately correspond to the device health status, and generate a device health status semantic token.

[0073] The working principle and effects of the above technical solution are as follows: By analyzing the dimensions and ranges of equipment health characterization information and dividing it into state intervals, the precision of health state classification is improved, reducing the discrimination bias caused by ambiguous state classification. Continuous state data is converted into discrete semantic identifiers, enhancing the unified expression capability of various health information types and reducing parsing bias in subsequent model analysis. Redundancy removal and fusion refinement of semantic features reduce the volume of invalid information, improve the purity of core state semantics, and avoid the accumulation of complex features that slows down the processing pace. Initial semantic tokens are formed through encoding and encapsulation, standardizing the format of health information flow and making cross-node data interaction smoother. The integrity of the tokens is verified and encoding biases are corrected, ensuring a high degree of consistency between semantic content and the actual health status of the equipment. This not only adapts to the lightweight flow processing needs of the edge side but also provides a regular and reliable semantic data foundation for subsequent twin model mapping and fault matching.

[0074] In one embodiment of the present invention, step S4 includes:

[0075] S41. Identify the relationships between equipment components and energy transmission links corresponding to potential failure mode data, and identify the starting points of failure induction.

[0076] S42. Import the fault initiation point parameters into the edge-side digital twin model, deduce the transmission and propagation trajectory of the fault between internal components of the equipment, and generate fault propagation path data;

[0077] S43. Analyze the impact range, evolution rate, and severity of fault propagation path data to classify risk coverage areas of different levels;

[0078] S44. Overlay risk level coloring and labeling on the visual data of equipment physical field evolution, complete the rendering of the full-domain regional risk heat map, and generate equipment failure risk heat map data;

[0079] S45. Calculate the proportion of high-risk areas and evolution time cycle in the statistical equipment failure risk heat map data, calculate the probability of failure in different time periods, and generate failure drop risk prediction data.

[0080] The working principle and effects of the above technical solution are as follows: By analyzing the connections between equipment components and the energy transmission chain, the starting point of the fault can be located, improving the accuracy of fault tracing and reducing deviations in fault source identification. Relying on digital twin simulations to reconstruct the fault propagation trajectory, the entire fault transmission process is fully restored, enhancing the realism of fault evolution analysis and avoiding prediction oversights caused by incomplete understanding of fault propagation patterns. Analyzing the propagation path delineates graded risk areas, refining risk level boundaries and reducing blind spots in control caused by general assessments. Overlaying risk labels to generate heat maps provides a visual representation of the overall risk distribution, improving the efficiency of on-site personnel in identifying potential hazard areas. Based on heat map data, the probability of fault occurrence is calculated, quantifying risk development trends. This not only allows for early detection of potential equipment safety hazards but also provides quantitative basis for subsequent graded early warnings, reducing downtime losses and safety risks caused by sudden faults.

[0081] In one embodiment of the present invention, S42 includes:

[0082] Extract the coordinate parameters, performance parameters, and initial fault intensity data of the fault initiation point, and summarize them to form a core parameter set of the initiation point;

[0083] The core parameter set of the starting point is numerically standardized to unify the parameter magnitude and data format, and standardized starting parameters are generated.

[0084] Standardized initial parameters are imported into the edge-side digital twin model to activate the fault evolution simulation module inside the model and simulate the initial fault triggering state.

[0085] Based on the relationships between equipment components and the energy transmission links, the transmission sequence and propagation speed of a fault from its starting point to surrounding components are deduced, generating fault propagation time series data.

[0086] By integrating fault propagation timing data and component spatial location information, a complete fault propagation trajectory is outlined, the rationality of the trajectory is verified, and fault propagation path data is generated.

[0087] The working principle and effects of the above technical solution are as follows: By extracting and summarizing multiple core parameters of the fault initiation point, the comprehensiveness of initial fault information capture is improved, reducing inference deviations caused by missing key parameters. Standardizing the core parameters and unifying their magnitude and format reduces model parsing errors caused by inconsistent data formats, enhances the smoothness of parameter import, and avoids inference failures or distorted results due to inconsistent parameters. Importing standardized parameters into the twin model and activating the inference module accurately simulates the initial fault state, improving the realism of fault inference. Based on component association and energy link inference of the transmission process, integrating temporal and spatial information to delineate and verify the trajectory improves the accuracy of the fault propagation path and reduces trajectory deviation. It can completely reconstruct the entire fault propagation process and provide a reliable basis for subsequent risk classification and heat map rendering, avoiding inadequate risk management due to path judgment errors.

[0088] In one embodiment of the present invention, step S5 includes:

[0089] S51. Integrate fault fall risk prediction data, real-time equipment operating data, and site environmental interference data to combine multi-dimensional risk assessment elements.

[0090] S52. Assign weights and quantify multi-dimensional risk assessment factors, complete edge-side multimodal fusion calculation and decision-making, and generate a comprehensive equipment risk index.

[0091] S53. Define the threshold ranges for four risk levels—minor, moderate, severe, and emergency—corresponding to the comprehensive risk index of the equipment; match the threshold range of the comprehensive risk index of the equipment with the corresponding early warning triggering logic and handling process;

[0092] S54. Simultaneously push early warning signals, risk details and corresponding handling guidelines to edge computing nodes, generate equipment fault early warning response data, and implement the entire process of remote monitoring and fault early warning of industrial equipment.

[0093] The working principle and effects of the above technical solution are as follows: By integrating multiple types of data, including fault prediction, real-time operating conditions, and environmental interference, and combining multi-dimensional risk factors, the comprehensiveness of risk assessment is improved, reducing the bias caused by single-data assessments. Weighting and quantifying various factors generates a comprehensive risk index, enhancing the objectivity and accuracy of risk assessment and avoiding misjudgments caused by subjective judgment. Four-level risk threshold ranges are defined and matched with corresponding early warning logic and handling procedures, refining the early warning classification and reducing over- or under-warning situations, avoiding delays or resource waste due to inaccurate early warnings. Early warning signals, risk details, and handling guidelines are simultaneously pushed to edge nodes, improving the timeliness of early warning response and enabling staff to quickly grasp the situation of potential hazards and take measures. This approach achieves both accurate quantification and graded early warning of risks, and implements full-process early warning and handling, reducing losses caused by the escalation of equipment failures and ensuring the stable operation of industrial production.

[0094] One embodiment of the present invention provides a remote monitoring and fault early warning system for industrial equipment based on edge computing, the system comprising:

[0095] One or more processors;

[0096] Memory, used to store one or more programs;

[0097] Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.

[0098] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for remote monitoring and fault early warning of industrial equipment based on edge computing, characterized in that, The method includes: S1. Perform multi-dimensional physical field modeling on the operating area of ​​industrial equipment to generate spatiotemporal distribution data of the equipment's physical field; based on the spatiotemporal distribution data of the equipment's physical field, deploy a physical information neural network on edge computing nodes to build a real-time inference framework on the edge side. S2. The multi-source sensor data during equipment operation is decoupled causally using the edge-side real-time inference framework to separate the dynamic response features of the operating conditions from the physical degradation features, generating decoupled operating condition feature data and degradation feature data; the decoupled degradation feature data is then subjected to edge-side lightweight feature enhancement processing to generate enhanced degradation feature data. S3. Based on the enhanced degradation feature data, perform semantic encoding of equipment health status to generate equipment health semantic tokens; dynamically map the equipment health semantic tokens through the edge-side digital twin model to generate equipment physical field evolution visualization data; perform fault mode matching on the equipment physical field evolution visualization data to generate potential fault mode data. S4. Drive the edge-side digital twin model with potential failure mode data to perform fault propagation simulation and generate fault propagation path data; render risk heatmaps of equipment physical field evolution visualization data based on fault propagation path data to generate equipment fault risk heatmap data; predict the probability of falling risk from the equipment fault risk heatmap data to generate fault falling risk prediction data. S5. Perform multimodal fusion decision-making on the edge side based on the fault drop risk prediction data to generate a comprehensive equipment risk index; trigger a hierarchical early warning mechanism on the edge computing node based on the comprehensive equipment risk index to generate equipment fault early warning response data.

2. The method for remote monitoring and fault early warning of industrial equipment based on edge computing according to claim 1, characterized in that, S1 includes: S11. Collect environmental parameters, structural parameters, and basic operating parameters of the industrial equipment operating site, and aggregate them to form a set of basic parameters for the entire equipment domain. S12. Perform redundant data filtering and abnormal data smoothing calibration on the equipment's overall basic parameter set to generate a normalized equipment basic parameter set. S13. Perform multi-level correlation decomposition of the standardized equipment basic parameter set in spatial, temporal and physical dimensions, complete the multi-dimensional physical field modeling of the industrial equipment operating area, and generate the spatiotemporal distribution data of the equipment physical field. S14. Divide the computing power levels and data processing permission ranges of edge computing nodes to adapt to the computational requirements of the spatiotemporal distribution data of the physical field of the device; deploy the physical information neural network to the corresponding edge computing nodes in layers according to the computing power range, and build a real-time inference framework on the edge side.

3. The method for remote monitoring and fault early warning of industrial equipment based on edge computing according to claim 2, characterized in that, S13 includes: Extract the basic parameters of the standardized equipment to centrally represent the point information and structural layout information of the equipment spatial layout, and form the equipment spatial reference elements; Continuous temporal sampling of equipment spatial reference elements captures dynamic changes in equipment operating status over time, generating equipment temporal evolution elements. Integrate spatial reference elements of equipment with temporal evolution elements of equipment into the laws of physical field change, and fuse them to form multi-dimensional related decomposition basic data; Cross-dimensional correlation and coupling analysis is carried out on the basic data of multi-dimensional correlation decomposition to complete the multi-dimensional physical field modeling of the industrial equipment operating area and generate the spatiotemporal distribution data of the equipment physical field.

4. The method for remote monitoring and fault early warning of industrial equipment based on edge computing according to claim 2, characterized in that, S14 includes: The volume and computational consumption of the physical field spatiotemporal distribution data of the statistical equipment are used to form statistical results of data computational load; combined with the hardware configuration of the edge nodes and the real-time idle resource status, the gradient computing power level is divided to generate a hierarchical sequence of edge node computing power. Match the statistical results of data processing load with the computing power classification sequence of edge nodes, define the data processing permission range corresponding to each node, and complete the adaptation and matching of computing power and business load. The network layers and computation modules of the physical information neural network are separated to create independent computation units that are adapted to different computing power levels. The split network computing units are sequentially allocated and deployed to the corresponding edge computing nodes, and the data interaction links between nodes are integrated to build a real-time inference framework on the edge side.

5. The method for remote monitoring and fault early warning of industrial equipment based on edge computing according to claim 1, characterized in that, The S2 includes: S21. Deploy a full-domain sensing network for industrial equipment, continuously collect raw data streams from multiple sources of sensors, and aggregate them to form a raw dataset of multi-source sensing data for the equipment. S22. Perform time-series alignment and noise filtering on the original dataset of multi-source sensing of the device to obtain standardized device sensing time-series data. S23. Disassemble the internal correlation logic of the sensing time sequence data of standardized equipment, complete the layered separation of dynamic response characteristics and physical degradation characteristics under working conditions, and generate decoupled working condition characteristic data and degradation characteristic data. S24. Perform dimensionality reduction and detail feature amplification on the decoupled degenerate feature data, adapt to the lightweight operation specifications of edge nodes, and generate intermediate degenerate feature data. S25. Perform feature dimension completion and semantic enhancement on the intermediate state degenerate feature data, complete the edge-side lightweight feature enhancement processing, and generate enhanced degenerate feature data.

6. The method for remote monitoring and fault early warning of industrial equipment based on edge computing according to claim 1, characterized in that, The S3 includes: S31. Extract the state change threshold, feature fluctuation trend and performance degradation gradient from the enhanced degradation feature data, and decompose the multi-level characterization information of the equipment health status. S32. Discretize and encapsulate the multi-level representation information of equipment health status to generate equipment health status semantic tokens. S33. Build an edge-side digital twin model that fits the physical structure and operating mechanism of the equipment to replicate the operating evolution pattern of the equipment throughout its entire life cycle; S34. Map the device health semantic token to the temporal evolution link of the edge-side digital twin model to restore the real-time changes in the device's physical field and generate visualized data on the evolution of the device's physical field. S35. Collect a feature library of historical fault samples of industrial equipment, compare the equipment physical field evolution visualization data with the features of historical fault samples, complete multi-dimensional fault mode matching, and generate potential fault mode data.

7. The method for remote monitoring and fault early warning of industrial equipment based on edge computing according to claim 6, characterized in that, S32 includes: The feature dimensions and numerical ranges of multi-level representation information of equipment health status are analyzed, the representation intervals of different health statuses are divided, and the health representation interval division results are generated. The core features of each health characterization interval are discretized, and the continuous state data is converted into discretized semantic labels to generate a discretized semantic feature set. Redundancy removal and feature fusion are performed on the discretized semantic feature set to retain the core semantic representation of health status and generate standardized semantic features. Standardized semantic features are encoded, assigned a unique identifier, and semantic information is encapsulated and integrated to generate an initial semantic token. Verify the integrity and accuracy of the initial semantic token, correct encoding deviations, and generate a device health status semantic token.

8. The method for remote monitoring and fault early warning of industrial equipment based on edge computing according to claim 1, characterized in that, The S4 includes: S41. Identify the relationships between equipment components and energy transmission links corresponding to potential failure mode data, and identify the starting points of failure induction. S42. Import the fault initiation point parameters into the edge-side digital twin model, deduce the transmission and propagation trajectory of the fault between internal components of the equipment, and generate fault propagation path data; S43. Analyze the impact range, evolution rate, and severity of fault propagation path data to classify risk coverage areas of different levels; S44. Overlay risk level coloring and labeling on the visual data of equipment physical field evolution, complete the rendering of the full-domain regional risk heat map, and generate equipment failure risk heat map data; S45. Calculate the proportion of high-risk areas and evolution time cycle in the statistical equipment failure risk heat map data, calculate the probability of failure in different time periods, and generate failure drop risk prediction data.

9. The method for remote monitoring and fault early warning of industrial equipment based on edge computing according to claim 1, characterized in that, The S5 includes: S51. Integrate fault fall risk prediction data, real-time equipment operating data, and site environmental interference data to combine multi-dimensional risk assessment elements. S52. Assign weights and quantify multi-dimensional risk assessment factors, complete edge-side multimodal fusion calculation and decision-making, and generate a comprehensive equipment risk index. S53. Define the threshold ranges for four risk levels—minor, moderate, severe, and emergency—corresponding to the comprehensive risk index of the equipment; match the threshold range of the comprehensive risk index of the equipment with the corresponding early warning triggering logic and handling process; S54. Synchronously push early warning signals, risk details and corresponding handling guidelines to edge computing nodes, generate equipment fault early warning response data, and implement the entire process of remote monitoring and fault early warning of industrial equipment.

10. A remote monitoring and fault early warning system for industrial equipment based on edge computing, characterized in that, The system includes: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 9.

Images