An industrial equipment anomaly monitoring method, medium and system

Abstract

The application discloses an industrial equipment anomaly monitoring method, medium and system, relates to the field of industrial equipment monitoring, and comprises the following steps: acquiring multi-source heterogeneous operation data of industrial equipment and preprocessing to generate an operation state sequence; inputting the operation state sequence into an anomaly monitoring model to output anomaly data; the anomaly monitoring model comprises monitoring networks of each equipment, each monitoring network comprises a state node, an external working condition node, an observation node and a state transition edge across time slices, the state node and the external working condition node point to the observation node, adjacent time slice state nodes are connected through the state transition edge, adjacent equipment monitoring networks are connected through equipment association nodes, and upstream equipment state nodes point to downstream equipment state nodes through the equipment association nodes; and the application realizes continuous monitoring of an equipment anomaly evolution track and dynamic early warning of overall risks of a linked production line.

Description

Technical Field

[0001] This application relates to the field of industrial equipment monitoring, and in particular to a method, medium, and system for monitoring industrial equipment anomalies. Background Technology

[0002] With the deepening development of intelligent manufacturing and the Industrial Internet, equipment maintenance models are shifting from reactive repair and scheduled maintenance to predictive maintenance. Traditional methods, which struggle to avoid unplanned downtime or over-maintenance, can no longer meet the demands of modern manufacturing for high availability and low-cost operation and maintenance. Fault identification and life prediction methods based on multi-source signals such as vibration, temperature, and current, leveraging sensor, edge computing, and artificial intelligence technologies, have been extensively studied.

[0003] However, in automated production lines such as warehousing and sorting, there are numerous types of equipment, drastic fluctuations in operating conditions, and strong coupling and cycle time dependencies between upstream and downstream equipment. Actual data comes from PLCs (Programmable Logic Controllers), sensors, SCADA (Supervisory Control and Data Acquisition) systems, and business systems, with significant differences in sampling frequency, dimensions, and noise characteristics, constituting a typical multi-source heterogeneous environment. Furthermore, equipment anomalies often evolve from slight drifts to failures, exhibiting significant time-cumulative effects and the risk of cross-equipment propagation. Current mainstream solutions generally adopt a basic framework of collecting multi-sensor data, preprocessing and feature extraction, and then using machine learning or static Bayesian models to determine the current state. This framework fails to establish a dynamic transmission relationship of states over time, lacks a mechanism for isomorphic fusion of multi-source data, and lacks the ability to systematically extract anomaly precursor sequences and evolution paths. Moreover, it fails to incorporate causal relationships between upstream and downstream equipment into the analysis, resulting in difficulty in capturing early anomalies in complex production lines, insufficient reliability of early warnings, and weak interpretability. Summary of the Invention

[0004] The purpose of this application is to provide a method for extracting and monitoring abnormal data of industrial equipment, which can integrate multi-source heterogeneous data to track the time-series evolution of equipment status and analyze the risk propagation across equipment, so as to achieve early warning of anomalies and reliable intervention.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for monitoring anomalies in industrial equipment, including: Acquire multi-source heterogeneous operating data of industrial equipment, wherein the multi-source heterogeneous operating data includes at least sensor-collected data, control system operating status data, and business management data; The multi-source heterogeneous operating data is preprocessed to generate an operating state sequence with a time dimension; The operational status sequence is input into the anomaly monitoring model, which outputs anomaly data. The anomaly monitoring model includes a monitoring network of several devices. Each device's monitoring network includes a state node, an external condition node, and an observation node within a time slice, as well as state transition edges across time slices. The state node characterizes the device's current potential health level and degradation degree. The external condition node characterizes exogenous variables of the device's operating environment and conditions. The observation node characterizes measurable parameters collected by sensors and business systems. The state node points to the observation node, and the external condition node points to the observation node. State nodes in adjacent time slices are connected through the state transition edges. The monitoring networks of adjacent devices are connected through device association nodes, and the state node of an upstream device points to the state node of a downstream device through the device association node.

[0006] Optionally, the preprocessing of the multi-source heterogeneous operating data specifically includes: The multi-source heterogeneous operating data are synchronized and time-aligned according to a unified monitoring time window to generate time-dimension sequence data. The time-dimension sequence data is processed by missing value handling, noise removal, and standardization to generate standard time-dimension sequence data. The standard sequence data with the time dimension is associated with the operating condition information to generate a time-dimensional operating status sequence; the operating condition information includes shift information, order cycle time, equipment start / stop status, load level, and operating mode.

[0007] Optionally, the status nodes include normal state, slightly abnormal state, moderately abnormal state, severely abnormal state, and fault state; the measurable parameters represented by the observation nodes include vibration amplitude, temperature rise rate, current fluctuation coefficient, pressure deviation, start-stop frequency, cycle deviation, speed fluctuation, and number of abnormal stops; the exogenous variables represented by the external operating condition nodes include load rate, ambient temperature, task intensity, shift difference, and control mode.

[0008] Optionally, inputting the running state sequence into the anomaly monitoring model and outputting anomaly data includes: Based on the operating state sequence, the anomaly monitoring model is used to calculate the posterior probability of each state node at the current time, as well as the predicted probability of entering an abnormal state or fault state within a preset time step in the future. When the predicted probability is consistently higher than a preset threshold within a preset time period, or when the growth rate of the predicted probability exceeds a preset growth threshold within multiple consecutive time windows, the data for the corresponding time period is identified as the abnormal data. The abnormal data includes the start time of the abnormality, the end time, the associated device identifier, the main abnormal characteristics, the risk level, the presumed abnormality type, the location of the fault, and the recommended maintenance measures.

[0009] Optionally, a multi-level early warning mechanism can be set up based on the abnormal data; When the predicted probability exceeds the corresponding threshold, a warning output of the corresponding level is triggered. The warning output includes at least one of the following methods: monitoring interface alarm, message push, work order generation, or integration with the enterprise management system.

[0010] Optionally, after inputting the running state sequence into the anomaly monitoring model and outputting anomaly data, the method further includes: The abnormal data and its corresponding state evolution path are structured and stored as an abnormal pattern database; The model parameters of the anomaly monitoring model are updated using the anomaly pattern database.

[0011] Optionally, the contents stored in the anomaly pattern database include: original anomaly data fragments, preprocessed standardized features, state transition paths, corresponding operating conditions, fault labels or maintenance result labels, anomaly precursor features, typical causal chain structures, risk levels and handling suggestions; Updating the model parameters of the anomaly monitoring model using the anomaly pattern database includes: Based on the newly added abnormal data or maintenance confirmation results in the abnormal pattern database, the conditional probability of the abnormal monitoring model is updated using an incremental learning method, and the nodes and dependencies in the abnormal monitoring model are adjusted.

[0012] Optionally, before inputting the operating state sequence into the anomaly monitoring model, the industrial equipment anomaly monitoring method further includes: Based on the aforementioned operating state sequence, suspected abnormal data segments are extracted and input into the anomaly monitoring model; the extraction of suspected abnormal data segments specifically involves: The statistical characteristics of the running state sequence are calculated based on a sliding window. The statistical characteristics include mean, variance, skewness, kurtosis, peak-to-peak value, frequency band energy, and rate of change. Based on the statistical characteristics, calculate the degree of deviation of the operating state sequence relative to the baseline operating condition, calculate the change in feature correlation, and calculate the deviation between the actual value of the observation node and the expected value under normal operating conditions; Based on the degree of offset, the change in feature correlation, and the deviation, suspected abnormal data segments are extracted according to a preset initial screening rule. The initial screening rule includes: when at least one of the degree of offset, the change in feature correlation, or the deviation is greater than its corresponding preset threshold, the corresponding time period of the running state sequence is marked as a suspected abnormal data segment.

[0013] Secondly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the industrial equipment anomaly monitoring method described above.

[0014] Thirdly, this application provides an industrial equipment anomaly monitoring system, the system comprising: The data acquisition module is used to acquire multi-source heterogeneous operating data of industrial equipment. The multi-source heterogeneous operating data includes at least sensor data, control system operating status data, and business management data. The preprocessing module is used to preprocess the multi-source heterogeneous operating data to generate an operating state sequence with a time dimension. The anomaly monitoring module is used to input the running status sequence into the anomaly monitoring model and output anomaly data; The anomaly monitoring model includes a monitoring network of several devices. Each device's monitoring network includes a state node, an external condition node, and an observation node within a time slice, as well as state transition edges across time slices. The state node characterizes the device's current potential health level and degradation degree. The external condition node characterizes exogenous variables of the device's operating environment and conditions. The observation node characterizes measurable parameters collected by sensors and business systems. The state node points to the observation node, and the external condition node points to the observation node. State nodes in adjacent time slices are connected through the state transition edges. The monitoring networks of adjacent devices are connected through device association nodes, and the state node of an upstream device points to the state node of a downstream device through the device association node.

[0015] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method, medium, and system for anomaly monitoring of industrial equipment. By integrating three types of multi-source heterogeneous operational data—sensor-acquired data, control system operating status data, and business management data—it breaks down the data barriers of traditional equipment monitoring that rely solely on single-sensor data. It comprehensively covers core information such as equipment hardware operating parameters, system control status, and business operating conditions. Simultaneously, through data preprocessing, it generates operating status sequences with a time dimension, achieving standardized and time-series unified modeling of heterogeneous data. This solves the problems of chaotic data formats, inconsistent dimensions, and lack of collaborative analysis in industrial scenarios, providing complete, continuous, and reliable data support for subsequent accurate anomaly monitoring and avoiding monitoring blind spots caused by missing data dimensions. The anomaly monitoring model designed in this application builds a dedicated monitoring network for each piece of equipment. Through a hierarchical association architecture of status nodes, external operating condition nodes, and observation nodes, it achieves refined tracing of the root causes of equipment failures. Among them, the status node accurately depicts the potential health status and long-term degradation trend of the equipment, making up for the shortcomings of traditional methods that can only identify surface data anomalies and cannot predict the hidden degradation of the equipment; the external operating condition node introduces exogenous variables such as environment and operating conditions, avoiding the problems of traditional monitoring models ignoring operating condition interference and being prone to false alarms and missed alarms; the observation node connects to real collected parameters, ensuring that the monitoring results are consistent with the actual operating conditions of the equipment. The three types of nodes form a complete "operating condition-status-observation" correlation logic, which significantly reduces the false alarm rate and missed detection rate of anomaly monitoring compared with traditional single algorithm models, while giving the monitoring process a clear physical logic, getting rid of the drawbacks of the black box nature and poor interpretability of traditional AI models.

[0016] Furthermore, this application relies on a time-series state transition mechanism to achieve dynamic and continuous monitoring of equipment degradation trends. The model establishes a temporal correlation between equipment operating states through state transition edges across time slices, continuously capturing the state evolution patterns of equipment at different operating stages. This allows for accurate identification of time-series faults such as slow degradation and periodic anomalies, distinguishing it from traditional fixed-point, static anomaly detection methods. This mechanism can effectively capture gradual faults and intermittent anomalies that occur during the long-term operation of industrial equipment, realizing a shift from "post-fault detection" to "pre-degradation early warning," predicting potential equipment failure risks in advance, and providing a valid basis for preventative equipment maintenance.

[0017] Furthermore, this application adds equipment association nodes to solve the problem of monitoring cascading failures of related equipment on the production line. Addressing the industry pain points of multi-equipment operation in industrial production, the easy propagation of faults between upstream and downstream equipment, and the inability of independent monitoring to identify cascading anomalies, this method achieves interconnection and interoperability of adjacent equipment monitoring networks through equipment association nodes, constructing a status correlation transmission link between upstream and downstream equipment. Based on the association mapping from upstream equipment status nodes to downstream equipment status nodes, cascading failures and coupled anomalies between equipment can be accurately captured, achieving collaborative anomaly monitoring of the entire production line, rather than isolated monitoring of a single device. This effectively adapts to the coordinated operation scenarios of assembly lines and complete sets of industrial equipment, significantly improving the completeness of overall production line anomaly monitoring. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is an application environment diagram of an industrial equipment anomaly monitoring method according to an embodiment of this application; Figure 2 This is a schematic flowchart of an industrial equipment anomaly monitoring method according to an embodiment of this application; Figure 3 This is a schematic diagram of a multi-source heterogeneous data homogenization processing structure according to an embodiment of this application; Figure 4 This is a schematic diagram of the network structure of an anomaly monitoring model according to an embodiment of this application; Figure 5 This is a schematic diagram illustrating the cross-device anomaly propagation relationship in a linked production line according to an embodiment of this application; Figure 6 This is a schematic diagram of an abnormal pattern database structure according to an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] The industrial equipment anomaly monitoring method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on other servers. Terminal 102 can send the collected multi-source heterogeneous operating data of industrial equipment to server 104. After receiving the multi-source heterogeneous operating data, server 104 preprocesses the data to generate a time-dimensional operating state sequence. This operating state sequence is then input into an anomaly monitoring model, which outputs anomaly data and the current status information of the equipment. The anomaly monitoring model includes a monitoring network of several devices. Each device's monitoring network includes a device health layer, a working condition constraint layer, and a feature observation layer within a time slice, as well as state transition edges across time slices. Server 104 can feed back the output anomaly data and early warning information to terminal 102. In addition, in some embodiments, the industrial equipment anomaly monitoring method can also be implemented by the server 104 or the terminal 102 alone. For example, the terminal 102 can directly perform local anomaly monitoring processing on the collected operating data, or the server 104 can obtain historical operating data or real-time collected data from the data storage system and perform anomaly monitoring processing on the data.

[0023] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0024] In one exemplary embodiment, such as Figure 2 As shown, an industrial equipment anomaly monitoring method is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 203. Wherein: Step 201: Obtain multi-source heterogeneous operation data of industrial equipment. The multi-source heterogeneous operation data includes at least sensor acquisition data, control system operation status data, and business management data.

[0025] Specifically, sensor data can include physical quantity data collected by vibration sensors, temperature sensors, pressure sensors, etc.; control system operating status data can include control logic data such as equipment start / stop status and operating speed fed back by programmable logic controllers (PLCs) or supervisory control and data acquisition (SCADA) systems; and business management data can include production management data such as task cycle times and shift arrangements issued by the work order system. By acquiring multi-source heterogeneous operating data, the physical characteristics, control status, and business environment during equipment operation can be comprehensively covered, providing a multi-dimensional data foundation for subsequent anomaly analysis.

[0026] Step 202: Preprocess the multi-source heterogeneous operating data to generate an operating state sequence with a time dimension.

[0027] Specifically, due to significant differences in sampling frequency, data structure, and dimensions among multi-source heterogeneous operational data, the preprocessing process requires mapping data from different sources to the same time coordinate system and performing noise filtering and dimension normalization to generate standard sequence data that characterizes the continuous operational status of equipment over time, facilitating subsequent time-series analysis by the model. Simultaneously, a unique equipment identifier is established for each type of equipment, and all collected data are appended with a unified timestamp, workstation identifier, operating condition identifier, and equipment category identifier to facilitate subsequent data association and traceability. Industrial equipment can include key production line equipment such as conveyors, sorting machines, stacker cranes, elevators, barcode scanners, servo drive mechanisms, motor drive units, pneumatic actuators, and their auxiliary control devices.

[0028] Step 203: Input the running status sequence into the anomaly monitoring model and output the anomaly data.

[0029] The anomaly monitoring model includes a monitoring network of several devices. Each device's monitoring network includes a device health layer, a working condition constraint layer, and a feature observation layer within a time slice, as well as state transition edges across time slices.

[0030] Specifically, the anomaly monitoring model is constructed using a dynamic Bayesian network, which decouples the device's internal health status, external environmental influences, and observable physical quantities through a hierarchical structure. The device health layer, operating condition constraint layer, and feature observation layer constitute the spatial topology within a single time slice, while the state transition edges constitute the evolutionary structure in the time dimension, thus forming a spatiotemporally unified dynamic causal modeling mechanism. When constructing the network topology, dependencies within time slices, transition relationships across time slices, and influence relationships across devices are established based on device mechanism knowledge, process logic, historical fault causal relationships, and statistical correlations. The dynamic Bayesian network can be constructed using a dual-time-slice template structure, that is, first defining the dependencies between two adjacent time slices, and then copying and expanding them to form a full-time-domain model, thereby reducing modeling complexity while ensuring model structural consistency.

[0031] The device health layer has status nodes to characterize the device's current potential health level and degree of degradation.

[0032] Specifically, state nodes are latent variable nodes in the model that cannot be directly measured by sensors but are inferred from observed data. For example, state nodes can characterize the invisible potential health level of equipment, such as normal operation, slight degradation, or failure. The operating condition constraint layer contains external operating condition nodes to characterize exogenous variables of the equipment's operating environment and operating conditions.

[0033] Specifically, the external operating condition node describes the external environmental factors during equipment operation, such as ambient temperature, load rate, or task intensity. These variables are independent of the equipment's own state but significantly affect its operating performance. The feature observation layer has observation nodes to characterize the measurable parameters collected by sensors and business systems.

[0034] Specifically, observation nodes are variables in the model from which data can be directly obtained, such as vibration amplitude, current fluctuation coefficient, or cycle deviation. These parameters are the external manifestations of the combined effect of the equipment's internal health status and external operating conditions.

[0035] In the directed edge connection relationship within the layer, the state node points to the observation node, and the external condition node points to the observation node.

[0036] Specifically, this directed edge connection defines the causal dependency mechanism between variables. The inherent health degradation of equipment directly leads to changes in observable physical quantities; for example, bearing wear causes increased vibration amplitude. Simultaneously, changes in external operating conditions also modulate the performance of observed variables; for example, high load leads to increased temperature. By pointing both the state node and the external operating condition node to the observation node, the model can distinguish between changes caused by physical degradation and changes caused by operating condition fluctuations, avoiding false alarms due to changes in operating conditions and significantly improving the model's robustness under complex operating conditions.

[0037] In the cross-time-slice evolution mechanism, state nodes of adjacent time slices are connected by state transition edges.

[0038] Specifically, equipment aging and failure development is a gradual and dynamic process, with the current health status depending on the status at previous times. State transition edges characterize the probabilistic path of equipment degradation from normal to minor anomalies, and then to severe failures. Through state transition edges across time slices, the model no longer analyzes data at each time point in isolation, but establishes a dynamic correlation between preceding and subsequent time sequences. This allows it to capture the trend characteristics of anomaly evolution, thereby identifying early, subtle signs before failures occur and enabling predictive maintenance.

[0039] In terms of cross-device risk propagation mechanism, the monitoring networks of adjacent devices are connected through device association nodes, and the status nodes of upstream devices point to the status nodes of downstream devices through the device association nodes.

[0040] Specifically, in a connected production line, there are physical and technological couplings between equipment. Anomalies in upstream equipment often propagate to downstream equipment through methods such as cycle time imbalance and load fluctuations. For example, unstable speeds in upstream conveyor equipment can cause frequent acceleration and deceleration in downstream sorting machines, thereby accelerating wear and tear on downstream equipment. By introducing interconnected nodes as bridges for cross-equipment risk propagation, the model can quantify the causal impact of upstream equipment deterioration on the health of downstream equipment. This allows for early assessment of the interconnected risks of the entire production line when a single piece of equipment experiences a minor anomaly, achieving system-level early warning.

[0041] By implementing steps 201 to 203 above, this embodiment simultaneously addresses uncertainties, missing data, and interconnected risks in industrial settings. On one hand, based on the inference mechanism of the probabilistic graphical model, when some sensor data is missing or contains noise, the model can still perform probabilistic inference through other observation nodes and state transition edges, ensuring the continuity of monitoring. On the other hand, by fusing three topological structures—intra-layer causal dependence, cross-time-slice evolution, and cross-device propagation—the model can not only explain the root cause of anomalies but also trace the propagation path of anomalies, providing interpretable technical support for formulating precise maintenance strategies.

[0042] In another exemplary embodiment of this application, the preprocessing process of multi-source heterogeneous operating data and the specific physical quantity mapping of each node in the anomaly monitoring model are described in detail.

[0043] Preprocessing of multi-source heterogeneous operational data specifically includes: Multi-source heterogeneous operational data are synchronized and time-aligned according to a unified monitoring time window to generate time-dimensional sequence data. Missing value processing, noise removal, and standardization are performed on the time-dimensional sequence data to generate time-dimensional standard sequence data. The time-dimensional standard sequence data is then associated with operating condition information to generate a time-dimensional operational status sequence. The operating condition information includes shift information, order cycle time, equipment start / stop status, load level, and operating mode.

[0044] Specifically, such as Figure 3 As shown, due to significant differences in the sampling frequencies of sensor-acquired data, control system operating status data, and business management data, time alignment and synchronization are fundamental to data fusion. For example, for high-frequency sampling data such as vibration, statistical characteristics such as root mean square value, peak factor, or frequency band energy are extracted within a unified monitoring time window (e.g., 1 second) for downsampling. For low-frequency event data such as PLC status, linear interpolation or zero-order hold is used to fill in the gaps within the unified monitoring time window, generating time-dimension sequence data. In the missing value processing and noise removal process, when data is discontinuously missing, mean imputation, linear interpolation, or probability imputation based on historical distribution are used; when data is continuously missing, the missing state is marked and used as an input for model observation uncertainty; simultaneously, sliding window midpoint filtering or outlier detection methods are used to remove obvious acquisition errors. In the standardization process, different physical quantities such as temperature, vibration, and current are normalized or standardized using standard deviation to eliminate dimensional differences and generate time-dimension standard sequence data. Finally, the standard sequence data is correlated with shift information, order cycle time, equipment start / stop status, load level, and operating mode to generate a time-dimensional operating status sequence. This embodiment, through the aforementioned homogenization preprocessing mechanism, unifies multi-source heterogeneous data into a single time-series framework, solving the problems of scattered data sources, inconsistent structures, and difficulties in joint analysis in industrial settings, thus providing a more comprehensive data foundation for subsequent anomaly identification.

[0045] The status nodes include normal state, slightly abnormal state, moderately abnormal state, severely abnormal state, and fault state; the measurable parameters represented by the observation nodes include vibration amplitude, temperature rise rate, current fluctuation coefficient, pressure deviation, start-stop frequency, cycle deviation, speed fluctuation, and number of abnormal stops; the exogenous variables represented by the external operating condition nodes include load rate, ambient temperature, task intensity, shift difference, and control mode.

[0046] Specifically, such as Figure 4As shown, the status nodes of the equipment health layer, as latent variables, are specifically divided into five levels: normal state, slightly abnormal state, moderately abnormal state, severely abnormal state, and fault state. This is used to finely characterize the gradual degradation process of the equipment from health to failure. The observation nodes of the feature observation layer represent measurable parameters including vibration amplitude, temperature rise rate, current fluctuation coefficient, pressure deviation, start-stop frequency, cycle time deviation, speed fluctuation, and number of abnormal pauses. These parameters are external manifestations of the combined effect of the equipment's internal health status and external operating conditions, and can be directly obtained through sensors or business systems. The external operating condition nodes of the operating condition constraint layer represent exogenous variables including load rate, ambient temperature, task intensity, shift differences, and control mode. These variables are independent of the equipment's own state but significantly affect the equipment's operating performance. It should be understood that the specific physical quantities listed above are merely exemplary and not limiting. In other embodiments, status nodes may be divided into fewer or more levels according to equipment type, and observation nodes and external operating condition nodes may also include other parameters that can reflect the equipment status or operating environment. This embodiment establishes a correlation between data features and physical mechanisms by uniformly mapping different physical quantities to the nodes of the anomaly monitoring model. This enables the model to distinguish between changes caused by physical degradation and changes caused by operating condition fluctuations, avoiding false alarms caused by changes in operating conditions and significantly improving the robustness and interpretability of the model under complex operating conditions.

[0047] After completing the network topology construction and node physical quantity mapping, parameter learning is required for the anomaly monitoring model to enable it to perform dynamic reasoning for specific equipment and production line scenarios. Model parameters include at least the initial state probability, state transition probability, conditional probability from a state node to an observation node, and the probability of conditional influence of operating nodes on state or observation nodes. Parameter learning can be conducted in several ways: initializing the conditional probability table based on expert experience to provide reasonable prior knowledge for the model; statistical estimation based on samples such as historical normal data, historical abnormal data, maintenance records, and downtime logs, using large-sample statistical regularities to correct the initial parameters; for data containing latent variables or missing observations, iterative estimation using the expectation-maximization method to gradually converge to the optimal parameters; for continuous observation variables, discretization before modeling, or continuous probability forms such as Gaussian conditional distribution can be used for modeling. Through the above parameter learning process, the anomaly monitoring model can accurately depict the probabilistic evolution of equipment from normal to fault states under different operating conditions, providing a reliable parameter foundation for subsequent online inference.

[0048] This application provides a detailed description of an exemplary embodiment of the online state reasoning process, anomaly determination conditions, and hierarchical early warning mechanism.

[0049] The operation state sequence is input into the anomaly monitoring model, and the anomaly data is output, including: based on the operation state sequence, the anomaly monitoring model calculates the posterior probability of each state node at the current time, and the predicted probability of entering an abnormal state or fault state within a preset time step in the future.

[0050] Specifically, after receiving a time-dimensional sequence of operating states, the anomaly monitoring model uses a forward inference algorithm to combine the values ​​of the observed nodes at the current moment with the values ​​of the external operating condition nodes to calculate the posterior probability of each state node (such as normal state, minor abnormal state, moderate abnormal state, severe abnormal state, and fault state) in the equipment health layer at the current moment. The posterior probability reflects the true confidence level of the equipment in each potential health level based on the current observation evidence. Subsequently, using the state degradation probability matrix defined by the state transition edges across time slices, a prediction algorithm is used to extrapolate the predicted probability of the equipment entering an abnormal or fault state within a preset time step (e.g., the next 5 time windows). The predicted probability not only considers the current degree of degradation but also incorporates historical evolution trends, thus enabling the quantification of future fault risks in advance. In addition to forward inference, a smoothing inference algorithm can be used to more accurately estimate the state at intermediate moments using past and future observation data, or the most probable state sequence inference method can be used to obtain the optimal path for the evolution of the equipment state over time, thereby more comprehensively depicting the dynamic change process of the equipment state. In addition, during the inference process, the model can also output the key observed variables that lead to an increased probability of anomalies and their causal links, and match the current anomaly characteristics with the historical anomaly patterns stored in the anomaly pattern database. If similar historical anomaly patterns exist, the model can combine historical handling experience to provide a more accurate inferred anomaly type and recommended maintenance measures for the current anomaly.

[0051] When the predicted probability is consistently higher than the preset threshold within a preset time period, or when the growth rate of the predicted probability exceeds the preset growth threshold in multiple consecutive time windows, the data for the corresponding time period will be identified as abnormal data.

[0052] Specifically, this embodiment employs a dual-judgment logic to extract abnormal data, adapting to different types of fault evolution patterns. For gradual degradation faults, the abnormal symptoms develop relatively slowly. When the predicted probability remains above a preset threshold (e.g., 0.7) for a preset time period (e.g., three consecutive time windows), it indicates a stable abnormal trend in the device, and the data for the corresponding time period is identified as abnormal data. For sudden deterioration faults, the condition may deteriorate rapidly in a short period. When the growth rate of the predicted probability exceeds a preset growth threshold (e.g., 0.15 per window) over multiple consecutive time windows, even if the absolute probability value has not yet reached the preset threshold, the data for the corresponding time period is also identified as abnormal data. Through this dual-judgment logic, the model can comprehensively cover both slow degradation and rapid deterioration abnormal modes, effectively avoiding missed detections.

[0053] The abnormal data includes the start time of the abnormality, the end time, the associated device identifier, the main abnormality characteristics, the risk level, the presumed abnormality type, the location of the fault, and the recommended maintenance measures.

[0054] Specifically, once abnormal data is identified, the system automatically generates structured abnormal data records. The start and end times of the abnormality define the time interval of its evolution; the associated equipment identifier clarifies the specific physical equipment at the source of the risk; the main abnormal characteristics record changes in key observation nodes that lead to increased probability (such as excessive vibration amplitude, excessively rapid temperature rise, etc.); the risk level is divided into different levels based on the absolute value and growth rate of the predicted probability; the inferred anomaly type and fault location are inferred based on a typical causal chain structure; and the recommended maintenance measures match the corresponding handling strategies according to the inferred anomaly type.

[0055] A multi-level early warning mechanism is set up based on abnormal data; when the predicted probability exceeds the corresponding threshold, the corresponding level of early warning output is triggered. The early warning output includes at least one of the following methods: monitoring interface alarm, message push, work order generation, or integration with the enterprise management system.

[0056] Specifically, the multi-level early warning mechanism is divided into multiple levels based on the numerical range of predicted probabilities and risk levels. For example, a Level 1 early warning is triggered when the predicted probability exceeds the first threshold, alerting maintenance personnel to pay attention; a Level 2 early warning is triggered when the predicted probability exceeds the second threshold, recommending inspection; and a Level 3 early warning is triggered when the predicted probability exceeds the third threshold, recommending load reduction or shutdown for maintenance. The early warning output method can be flexibly combined according to the level. A Level 1 early warning can be displayed only on the monitoring interface; a Level 2 early warning can be sent to relevant personnel via message push (such as SMS or application notifications); and a Level 3 early warning automatically generates a maintenance work order and interfaces with the enterprise management system (such as an enterprise resource planning system or equipment management system) to achieve a closed-loop flow of maintenance tasks. Furthermore, combining cross-device impact reasoning, an emergency early warning is triggered when a risk of a complete line shutdown is determined. That is, when the condition of upstream equipment deteriorates and propagates to downstream equipment through related nodes, causing the predicted probabilities of multiple devices to simultaneously exceed the linkage alarm threshold, the system triggers a linkage alarm, prompting maintenance personnel to intervene before the entire line is shut down. The emergency early warning simultaneously pushes alarm information to multiple workstations and automatically generates collaborative handling work orders. Through a multi-level early warning mechanism, the system can intervene in the early stages of equipment failure. Maintenance personnel can take different levels of maintenance measures according to the warning level, thereby avoiding sudden downtime and optimizing the allocation of maintenance resources.

[0057] This application provides a detailed description of an exemplary embodiment of the structured storage of abnormal data and the closed-loop self-updating mechanism of the anomaly monitoring model.

[0058] Abnormal data and their corresponding state evolution paths are structured and stored as an abnormal pattern database; the abnormal pattern database is used to update the model parameters of the abnormal monitoring model.

[0059] Specifically, existing technologies often end the process after outputting an anomaly warning signal, resulting in the loss of anomaly evolution data. The model can only maintain the static parameters from the initial training, making it difficult to adapt to dynamic changes such as equipment aging and changes in operating conditions. This embodiment not only extracts anomaly data but also stores the anomaly data and its corresponding state evolution paths in a structured manner, forming an anomaly pattern database. The state evolution path records the complete probabilistic trajectory of the equipment transitioning from a normal state through a minor anomaly state, a moderate anomaly state, to a severe anomaly state, or a fault state. By using the anomaly pattern database to update the model parameters of the anomaly monitoring model, the model can self-correct based on newly occurring real anomaly samples, forming a closed-loop link of data acquisition, anomaly inference, pattern accumulation, and parameter update. This fundamentally solves the problem of the decline in prediction accuracy of traditional static models over time, enabling the model to have continuous optimization capabilities.

[0060] The abnormal pattern database stores the following: original abnormal data fragments, preprocessed standardized features, state transition paths, corresponding operating conditions, fault labels or maintenance result labels, abnormal occurrence precursor features, typical causal chain structures, risk levels and handling recommendations.

[0061] Specifically, such as Figure 6 As shown, the anomaly pattern database adopts a multi-dimensional feature-structured organization. The original anomaly data fragments retain the original waveforms or logs of sensors, control systems, and business systems during the anomaly period, for subsequent in-depth analysis. Preprocessed standardized features store time-series feature vectors that have undergone time alignment, missing value handling, and standardization, facilitating direct use in model calculations. State transition paths record the probability distribution sequence of state nodes inferred by the anomaly monitoring model over time. Corresponding operating conditions record the values ​​of external operating condition nodes such as load rate, ambient temperature, and task intensity at the time of the anomaly, used to distinguish between operating condition fluctuations and actual faults. Fault labels or maintenance result labels are entered by maintenance personnel after maintenance is completed, serving as monitoring signals to guide parameter updates. Anomaly precursor features extract subtle changes in observed nodes before a sharp increase in predicted probability. Typical causal chain structures record the key observed nodes leading to the anomaly and their dependencies. Risk levels and handling suggestions provide reference templates for rapid responses to similar anomalies. Through this multi-dimensional structured storage, the anomaly pattern database is not only a data warehouse but also a domain knowledge graph containing physical mechanisms, evolutionary logic, and handling experience.

[0062] The model parameters of the anomaly monitoring model are updated using the anomaly pattern database, including: updating the conditional probability of the anomaly monitoring model using incremental learning based on newly added anomaly data or maintenance confirmation results in the anomaly pattern database, and adjusting the nodes and dependencies in the anomaly monitoring model.

[0063] Specifically, when a certain amount of new abnormal data accumulates in the abnormal pattern database or a maintenance confirmation result is received, the system triggers a model self-update mechanism. Unlike the traditional offline training that re-collects all data, this embodiment uses incremental learning, only using newly added abnormal samples to locally update the model parameters. For example, when a maintenance result label confirms that an abnormality is bearing wear, the system will enhance the conditional probability between the "vibration amplitude" observation node and the "severe abnormal state" state node under that abnormal pattern, while fine-tuning the transition probability from "minor abnormal state" to "moderate abnormal state" in the state transition edge. In addition, if new abnormal data contains observation feature combinations that the existing model cannot explain, the system can also add observation nodes or adjust the dependencies between nodes in the abnormal monitoring model. At the same time, the system can also dynamically adjust the warning threshold based on the statistical results in the abnormal pattern database, so that the warning sensitivity adapts to equipment aging or changes in operating conditions; when the equipment undergoes technical upgrades or the process changes, the external operating condition nodes and their dependencies in the operating condition constraint layer are adjusted accordingly to maintain the consistency between the model and the actual operating environment. For example, if a certain type of motor is found to exhibit current fluctuations in a specific frequency band before a stator fault occurs, and this frequency band feature is not included in the existing model, a corresponding observation node can be added to the feature observation layer, and a causal connection between it and the fault state can be established. This closed-loop self-updating mechanism enables the anomaly monitoring model to continuously evolve with the accumulation of operational data, not only improving the accuracy of identifying known fault types but also enhancing the ability to perceive unknown new faults, thus achieving continuous model optimization and knowledge accumulation.

[0064] In one exemplary embodiment of this application, by introducing a preliminary screening mechanism before model inference, the online calculation range of the anomaly monitoring model can be effectively narrowed, and the system's computational load can be reduced.

[0065] Before inputting the operating status sequence into the anomaly monitoring model, the industrial equipment anomaly monitoring method also includes: extracting suspected anomaly data segments based on the operating status sequence and inputting the suspected anomaly data segments into the anomaly monitoring model.

[0066] Specifically, industrial equipment operates in a normal state for most of its operating time. If all time window operating state sequences were indiscriminately input into the anomaly monitoring model for probabilistic inference, it would consume a large amount of computational resources, making it difficult to meet the needs of real-time online monitoring in industrial settings. Therefore, this embodiment adds an initial screening step before model inference. Through lightweight statistical calculations, it quickly filters out obviously normal data segments, sending only suspected abnormal data segments into the anomaly monitoring model, thereby significantly reducing the computational load of online model inference.

[0067] Extract suspected abnormal data segments, specifically: calculate the statistical characteristics of the running state sequence based on a sliding window. The statistical characteristics include mean, variance, skewness, kurtosis, peak-to-peak value, frequency band energy, and rate of change.

[0068] Specifically, a sliding window slides across the operating state sequence with a fixed step size to extract statistical characteristics of the data within the window. The mean and variance reflect the average level and fluctuation range of the equipment's operating parameters; skewness and kurtosis characterize the symmetry and sharpness of the data distribution. When a local defect occurs in the equipment, the impact signal causes a long tail phenomenon in the data distribution, significantly increasing skewness and kurtosis; peak-to-peak value captures extreme fluctuations within the window; frequency band energy is calculated by performing a Fast Fourier Transform on the time-domain signal to determine the energy distribution of a specific frequency band. Different fault types often correspond to energy surges in specific frequency bands; the rate of change reflects the drastic change of parameters over time. It should be understood that the above statistical characteristics are merely exemplary and not limiting. In other embodiments, other statistical characteristics, such as waveform factor or pulse factor, can be added or replaced according to the equipment type and monitoring requirements.

[0069] Based on statistical characteristics, the degree of deviation of the operating state sequence relative to the baseline operating condition is calculated, the change in characteristic correlation is calculated, and the deviation between the actual values ​​of the observation nodes and the expected values ​​under normal operating conditions is calculated.

[0070] Specifically, the baseline operating condition refers to the historical statistical characteristic distribution of the equipment under healthy conditions and corresponding operating conditions. The degree of deviation is measured by calculating the Mahalanobis or Euclidean distance between the current window statistical characteristics and the corresponding characteristic distribution under the baseline operating condition; the larger the distance, the further the current operating state deviates from the normal state. Changes in feature correlation are used to capture abnormal coupling relationships between multiple variables. For example, under normal operating conditions, vibration amplitude and load rate are positively correlated. If the vibration amplitude rises sharply at a certain moment while the load rate remains unchanged, the feature correlation changes. The disruption of this cooperative relationship is often a precursor to complex faults. The deviation between the actual value of the observation node and the expected value under normal operating conditions is obtained by calculating the residual between the actual observed value and the expected value output based on the health baseline model. The magnitude of the residual directly reflects the degree of equipment performance degradation. This embodiment ensures that the initial screening mechanism can cover different types of abnormal patterns through multi-dimensional deviation calculation, avoiding missed detections of single indicators.

[0071] Based on the degree of offset, changes in feature correlation, and deviation, suspected abnormal data segments are extracted according to preset screening rules. The screening rules include: when at least one of the degree of offset, changes in feature correlation, or deviation is greater than its corresponding preset threshold, the corresponding time period of the running state sequence is marked as a suspected abnormal data segment.

[0072] Specifically, the preset threshold is set based on the statistical distribution of historical normal data. For example, the threshold can be set as the boundary of the 99.9% confidence interval of the baseline operating condition distribution. The initial screening rule uses "OR" logic, meaning that if any one of the offset degree, feature correlation change, or deviation exceeds its corresponding preset threshold, the system marks that time period as a suspected anomalous data segment. This lenient marking strategy aims to minimize the false negative rate, preferring to introduce a certain number of false positives for accurate judgment by the subsequent anomaly monitoring model rather than overlooking any weak anomalous signs. The marked suspected anomalous data segments, along with their pre-set length of context data, are truncated and used as input to the anomaly monitoring model. In this embodiment, through the above-mentioned initial screening mechanism, more than 80% of normal data segments are filtered out without sacrificing anomaly sensitivity, significantly reducing the computational load of the online inference of the anomaly monitoring model. This allows the system to meet the needs of high-frequency real-time monitoring in industrial sites while ensuring monitoring accuracy.

[0073] This application provides an exemplary embodiment, using a tobacco logistics center warehousing and sorting line scenario as an example to more clearly illustrate the technical solution. It should be understood that this scenario is merely illustrative and not restrictive; the technical solution of this application is also applicable to other continuous and interconnected industrial scenarios, such as express delivery sorting lines, automobile assembly lines, or food packaging lines.

[0074] In this scenario, the industrial equipment includes interconnected devices such as conveyors, sorting machines, and elevators. The system first acquires multi-source heterogeneous operational data from the industrial equipment. This multi-source heterogeneous operational data includes at least sensor data, control system operational status data, and business management data. Specifically, vibration sensors, temperature sensors, and current sensors are deployed for key equipment such as conveyor motors, sorting machine actuators, and elevator drive units to collect sensor data. The system then connects to a Programmable Logic Controller (PLC) and a Supervisory Control and Data Acquisition (SCADA) system to obtain control system operational status data, such as equipment start / stop status and operating speed. Simultaneously, it interfaces with a Warehouse Management System (WMS) to obtain business management data, such as order takt time and shift scheduling information.

[0075] Subsequently, the multi-source heterogeneous operational data is preprocessed to generate an operational status sequence with a time dimension. Specifically, the preprocessing process uses a 1-second unified monitoring time window for time alignment and synchronization. For high-frequency sampled vibration data, statistical features such as root mean square value, peak factor, and specific frequency band energy are extracted and downsampled within each 1-second window; for medium-frequency data such as temperature and current, the mean and rate of change within the window are extracted; for low-frequency event data such as PLC status, zero-order hold is used to complete the data within the 1-second window. Simultaneously, missing value processing, noise removal, and standardization are performed on the data, and the processed data is correlated with operating condition information (such as shift information, order cycle time, equipment start / stop status, load level, and operating mode) to ultimately generate an operational status sequence with a time dimension.

[0076] Next, the running state sequence is input into the anomaly monitoring model, which outputs anomaly data. The anomaly monitoring model comprises a monitoring network of several devices. Each device's monitoring network includes a device health layer, a condition constraint layer, and a feature observation layer within a time slice, as well as state transition edges across time slices. The device health layer has state nodes to characterize the device's current potential health level and degradation degree. The condition constraint layer has external condition nodes to characterize exogenous variables related to the device's operating environment and conditions. The feature observation layer has observation nodes to characterize measurable parameters collected by sensors and the business system. State nodes point to observation nodes, external condition nodes point to observation nodes, and state nodes in adjacent time slices are connected via state transition edges. The monitoring networks of adjacent devices are connected through device association nodes, with the state node of an upstream device pointing to the state node of a downstream device via the device association node. Specifically, in this sorting line scenario, the status nodes of the equipment health layer are divided into normal state, slightly abnormal state, moderately abnormal state, severely abnormal state, and fault state; the observation nodes of the feature observation layer include vibration amplitude, temperature rise rate, current fluctuation coefficient, and cycle deviation; the external operating condition nodes of the operating condition constraint layer include load rate and shift difference.

[0077] During model inference, based on the operational state sequence, an anomaly monitoring model is used to calculate the posterior probability of each state node at the current moment, as well as the predicted probability of entering an abnormal or faulty state within a preset time step. Specifically, when the system is running online, for each new 1-second window of data received, the predicted probability of the device entering an abnormal or faulty state within the next 5 windows is calculated. When the predicted probability is consistently higher than a preset threshold within a preset time period, or when the growth rate of the predicted probability exceeds a preset growth threshold over multiple consecutive time windows, the data for the corresponding time period is identified as abnormal data. For example, when the predicted probability of the sorting machine's abnormal state is greater than 0.7 for three consecutive windows, and the fault probability growth rate exceeds 0.15 per window, the system identifies the data for the corresponding time period as abnormal data and outputs abnormal data including the anomaly start time, end time, associated device identifier, main anomaly characteristics, risk level, presumed anomaly type, fault location, and recommended maintenance measures.

[0078] A multi-level early warning mechanism is set up based on abnormal data. When the predicted probability exceeds the corresponding threshold, the corresponding level of early warning output is triggered. The early warning output includes at least one of the following methods: monitoring interface alarm, message push, work order generation, or integration with the enterprise management system. For example, when the predicted probability exceeds 0.7, a level 2 early warning is triggered, and a message push is sent to the maintenance personnel to suggest inspection; when the predicted probability exceeds 0.9, a level 3 early warning is triggered, and a maintenance work order is automatically generated and integrated with the enterprise management system.

[0079] like Figure 5As shown, in the tobacco logistics center's warehousing and sorting line, there is a strict process coupling between upstream conveyor equipment, intermediate sorting equipment, and downstream execution equipment. When the drive belt of the upstream conveyor equipment experiences slight misalignment or roller jamming, its vibration amplitude and current fluctuation coefficient begin to rise slowly, causing the state node of the upstream conveyor equipment to transition from a normal state to a slightly abnormal state. At this time, due to the minor degree of abnormality, univariate threshold monitoring often fails to trigger an alarm. However, the speed fluctuations and cycle deviations of the upstream conveyor equipment are transmitted to the downstream sorting machine through the physical conveyor belt, causing the downstream sorting machine to frequently accelerate and decelerate to adapt to the cycle changes of the upstream. In the anomaly monitoring model, the state node of the upstream conveyor equipment points to the state node of the downstream sorting machine through the equipment association node. This directed edge across equipment quantifies the slight anomaly risk of the upstream as a modulating effect on the state transition probability of the downstream equipment. Therefore, even if the observed data of the downstream sorting machine itself has not deteriorated significantly, the model will increase the predicted probability of the downstream sorting machine entering an abnormal state based on the deterioration of the upstream equipment's state node through the equipment association node. When the overall risk score reaches the alarm threshold, the system triggers an alarm in advance, prompting maintenance personnel to intervene before upstream conveyor equipment causes a substantial malfunction in the downstream sorting machine. This embodiment, through the above mechanism, effectively avoids line-wide downtime caused by overlooking minor anomalies in a single piece of equipment, significantly improving the overall operational stability and maintenance foresight of the integrated production line.

[0080] Based on the same inventive concept as the above-described method embodiments, this application provides an industrial equipment anomaly monitoring system in one embodiment. The system includes a data acquisition module, a preprocessing module, and an anomaly monitoring module. The modules interact with each other via a communication bus or internal network to jointly complete the fusion of multi-source heterogeneous data, dynamic state reasoning, and anomaly early warning functions.

[0081] The data acquisition module is used to acquire multi-source heterogeneous operating data of industrial equipment. The multi-source heterogeneous operating data includes at least sensor data, control system operating status data, and business management data.

[0082] Specifically, the data acquisition module, as the system's input, can connect to field devices via industrial Ethernet, programmable logic controller (PLC) communication bus, or message queue telemetry transmission protocol. For example, the data acquisition module may include a data acquisition card or edge computing gateway to receive high-frequency physical quantity data from vibration and temperature sensors in real time. Simultaneously, it reads operational status data such as equipment start / stop status from the control system via an open platform communication unified architecture protocol, and obtains business management data such as order cycle time and shift schedules from business management software such as warehouse management systems via application programming interfaces (APIs). The data acquisition module then timestamps the received multi-source heterogeneous operational data before sending it to the preprocessing module.

[0083] The preprocessing module is used to preprocess multi-source heterogeneous operating data to generate an operating state sequence with a time dimension.

[0084] Specifically, the preprocessing module can be implemented by a data processing thread in a field-programmable gate array (FPGA) or an edge server. Upon receiving heterogeneous operational data from multiple sources, this module performs operations such as time alignment and synchronization, missing value handling, noise removal, standardization, and operational condition label association. For example, for high-frequency sampled vibration data, the preprocessing module extracts statistical features such as the root mean square value according to a unified monitoring time window of 1 second; for low-frequency control system state data, it uses the zero-order hold method for completion. The preprocessing module maps data from different sources and frequencies to the same time-series framework, generating a time-dimensional operational state sequence, which is then pushed to the anomaly monitoring module for in-depth analysis.

[0085] The anomaly monitoring module is used to input the running status sequence into the anomaly monitoring model and output anomaly data.

[0086] Specifically, the anomaly monitoring module is the core computing unit of the system, which can be deployed on a cloud server or a local high-performance computing station. It executes the inference algorithm of the probabilistic graphical model through a central processing unit or a graphics processing unit. The anomaly monitoring model includes a monitoring network of several devices. Each device's monitoring network includes a state node, an external condition node, and an observation node within a time slice, as well as state transition edges across time slices. The state node characterizes the device's current potential health level and degradation degree. The external condition node characterizes exogenous variables related to the device's operating environment and conditions. The observation node characterizes measurable parameters collected by sensors and the business system. The state node points to the observation node, and the external condition node points to the observation node. State nodes in adjacent time slices are connected through the state transition edges. The monitoring networks of adjacent devices are connected through device association nodes, with the state node of an upstream device pointing to the state node of a downstream device via the device association node.

[0087] After receiving the operating state sequence, the anomaly monitoring module maps it to the observation nodes of the feature observation layer and the external operating condition nodes of the operating condition constraint layer. Subsequently, based on the inference engine of a dynamic Bayesian network, and combining the dependencies defined by state transition edges and device-related nodes, the module calculates the posterior probability of each state node at the current moment and the predicted probability within a preset future time step. When the predicted probability meets the anomaly judgment condition, the anomaly monitoring module outputs anomaly data containing information such as the anomaly time, device identifier, and risk level. This embodiment, by materializing the above method steps into a system module, enables complex dynamic causal modeling and cross-device risk propagation calculations to run stably in the industrial field's hardware and software architecture, achieving automated workflow from data acquisition to anomaly early warning.

[0088] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0089] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores multi-source heterogeneous operating data, operating status sequences, abnormal data, and abnormal pattern databases generated during industrial equipment anomaly monitoring. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the industrial equipment anomaly monitoring method.

[0090] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0091] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0092] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0093] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0094] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0095] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0096] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0097] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0098] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for monitoring anomalies in industrial equipment, characterized in that, The industrial equipment anomaly monitoring method includes: Acquire multi-source heterogeneous operating data of industrial equipment, wherein the multi-source heterogeneous operating data includes at least sensor-collected data, control system operating status data, and business management data; The multi-source heterogeneous operating data is preprocessed to generate an operating state sequence with a time dimension; The operational status sequence is input into the anomaly monitoring model, which outputs anomaly data. The anomaly monitoring model includes a monitoring network of several devices. Each device's monitoring network includes a state node, an external condition node, and an observation node within a time slice, as well as state transition edges across time slices. The state node characterizes the device's current potential health level and degradation degree. The external condition node characterizes exogenous variables of the device's operating environment and conditions. The observation node characterizes measurable parameters collected by sensors and business systems. The state node points to the observation node, and the external condition node points to the observation node. State nodes in adjacent time slices are connected through the state transition edges. The monitoring networks of adjacent devices are connected through device association nodes, and the state node of an upstream device points to the state node of a downstream device through the device association node.

2. The industrial equipment anomaly monitoring method according to claim 1, characterized in that, The preprocessing of the multi-source heterogeneous operating data specifically includes: The multi-source heterogeneous operating data are synchronized and time-aligned according to a unified monitoring time window to generate time-dimension sequence data. The time-dimension sequence data is processed by missing value handling, noise removal, and standardization to generate standard time-dimension sequence data. The standard sequence data with the time dimension is associated with the operating condition information to generate a time-dimensional operating status sequence; the operating condition information includes shift information, order cycle time, equipment start / stop status, load level, and operating mode.

3. The industrial equipment anomaly monitoring method according to claim 1, characterized in that, The status nodes include normal state, slightly abnormal state, moderately abnormal state, severely abnormal state, and fault state; the measurable parameters represented by the observation nodes include vibration amplitude, temperature rise rate, current fluctuation coefficient, pressure deviation, start-stop frequency, cycle deviation, speed fluctuation, and number of abnormal stops; the exogenous variables represented by the external operating condition nodes include load rate, ambient temperature, task intensity, shift difference, and control mode.

4. The industrial equipment anomaly monitoring method according to claim 1, characterized in that, The step of inputting the operating state sequence into the anomaly monitoring model and outputting anomaly data includes: Based on the operating state sequence, the anomaly monitoring model is used to calculate the posterior probability of each state node at the current time, as well as the predicted probability of entering an abnormal state or fault state within a preset time step in the future. When the predicted probability is consistently higher than a preset threshold within a preset time period, or when the growth rate of the predicted probability exceeds a preset growth threshold within multiple consecutive time windows, the data for the corresponding time period is identified as the abnormal data. The abnormal data includes the start time of the abnormality, the end time, the associated device identifier, the main abnormal characteristics, the risk level, the presumed abnormality type, the location of the fault, and the recommended maintenance measures.

5. The industrial equipment anomaly monitoring method according to claim 4, characterized in that, A multi-level early warning mechanism is set up based on the abnormal data; When the predicted probability exceeds the corresponding threshold, a warning output of the corresponding level is triggered. The warning output includes at least one of the following methods: monitoring interface alarm, message push, work order generation, or integration with the enterprise management system.

6. The industrial equipment anomaly monitoring method according to claim 1, characterized in that, After inputting the running state sequence into the anomaly monitoring model and outputting the anomaly data, the method further includes: The abnormal data and its corresponding state evolution path are structured and stored as an abnormal pattern database; The model parameters of the anomaly monitoring model are updated using the anomaly pattern database.

7. The industrial equipment anomaly monitoring method according to claim 6, characterized in that, The abnormal pattern database stores the following: original abnormal data fragments, preprocessed standardized features, state transition paths, corresponding operating conditions, fault labels or maintenance result labels, abnormal occurrence precursor features, typical causal chain structures, risk levels and handling suggestions. Updating the model parameters of the anomaly monitoring model using the anomaly pattern database includes: Based on the newly added abnormal data or maintenance confirmation results in the abnormal pattern database, the conditional probability of the abnormal monitoring model is updated using an incremental learning method, and the nodes and dependencies in the abnormal monitoring model are adjusted.

8. The industrial equipment anomaly monitoring method according to claim 1, characterized in that, Before inputting the operating state sequence into the anomaly monitoring model, the industrial equipment anomaly monitoring method further includes: Based on the aforementioned operating state sequence, suspected abnormal data segments are extracted and input into the anomaly monitoring model; the extraction of suspected abnormal data segments specifically involves: The statistical characteristics of the running state sequence are calculated based on a sliding window. The statistical characteristics include mean, variance, skewness, kurtosis, peak-to-peak value, frequency band energy, and rate of change. Based on the statistical characteristics, calculate the degree of deviation of the operating state sequence relative to the baseline operating condition, calculate the change in feature correlation, and calculate the deviation between the actual value of the observation node and the expected value under normal operating conditions; Based on the degree of offset, the change in feature correlation, and the deviation, suspected abnormal data segments are extracted according to a preset initial screening rule. The initial screening rule includes: when at least one of the degree of offset, the change in feature correlation, or the deviation is greater than its corresponding preset threshold, the corresponding time period of the running state sequence is marked as a suspected abnormal data segment.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the industrial equipment anomaly monitoring method according to any one of claims 1-8.

10. An industrial equipment anomaly monitoring system, characterized in that, The system includes: The data acquisition module is used to acquire multi-source heterogeneous operating data of industrial equipment. The multi-source heterogeneous operating data includes at least sensor data, control system operating status data, and business management data. The preprocessing module is used to preprocess the multi-source heterogeneous operating data to generate an operating state sequence with a time dimension. The anomaly monitoring module is used to input the running status sequence into the anomaly monitoring model and output anomaly data; The anomaly monitoring model includes a monitoring network of several devices. Each device's monitoring network includes a state node, an external condition node, and an observation node within a time slice, as well as state transition edges across time slices. The state node characterizes the device's current potential health level and degradation degree. The external condition node characterizes exogenous variables of the device's operating environment and conditions. The observation node characterizes measurable parameters collected by sensors and business systems. The state node points to the observation node, and the external condition node points to the observation node. State nodes in adjacent time slices are connected through the state transition edges. The monitoring networks of adjacent devices are connected through device association nodes, and the state node of an upstream device points to the state node of a downstream device through the device association node.

Images