Enterprise fault clustering early warning method and system based on production equipment data analysis

Abstract

The invention discloses an enterprise fault clustering early warning method and system based on production equipment data analysis, and relates to the technical field of equipment fault monitoring. Monitoring equipment units and data monitoring nodes thereof are determined, and a monitoring node cluster expansion tree based on a process relationship is constructed; then, performing independence division and abnormal mark classification on the historical monitoring data to obtain normal and abnormal historical monitoring sub-data; mapping the sub-data to an expansion tree, and generating normal and abnormal monitoring data cluster expansion trees; triggering features are extracted through data interval monitoring, and a feature representation library is formed; and finally, analyzing real-time data by using the database, matching feature performance and early warning a fault. According to the method, the limitation of traditional threshold comparison is overcome by integrating the equipment process relationship and the multi-dimensional data, systematic and advanced fault prediction is realized, and the reliability of industrial production is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment failure monitoring, and particularly to an enterprise fault clustering early warning method and system based on production equipment data analysis. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, the complexity and connectivity of production equipment systems have been significantly improved, involving multiple devices and diverse monitoring nodes (such as temperature, pressure, and vibration sensors), generating a large amount of real-time and historical data. These data provide an important basis for fault prediction and maintenance. However, existing fault judgment technologies mainly rely on threshold comparison methods, that is, by setting upper and lower threshold values of monitoring parameters to determine whether the operating state of the equipment is abnormal. Although this method is simple to implement, it has significant limitations in practical applications. First, threshold comparison usually can only trigger an alarm when a fault occurs or an abnormal parameter exceeds the threshold, making it difficult to give an early warning of potential faults, resulting in enterprises being unable to take preventive measures in a timely manner, increasing the risk of equipment damage and production interruption. Second, single-threshold judgment ignores the process correlation between devices and the multi-dimensional dynamic characteristics of monitoring data, making it difficult to capture early signs of complex systematic faults. In addition, the fixed threshold setting has poor adaptability to different devices and operating conditions, easily leading to false alarms or missed alarms, reducing the accuracy and reliability of early warning. Therefore, there is an urgent need for a method that can transcend the limitations of threshold comparison and achieve proactive early warning before a fault based on multi-dimensional data analysis, so as to improve the advance and accuracy of production equipment fault prediction and ensure the continuity and safety of industrial production. Summary of the Invention

[0003] The purpose of the present invention is to provide a fault early warning method and system that can more effectively warn production equipment.

[0004] The present invention discloses an enterprise fault clustering early warning method based on production equipment data analysis, including:

[0005] Step S100, determining equipment units to be monitored, and determining a plurality of data monitoring nodes corresponding to each equipment unit, classifying the plurality of data monitoring nodes into the same monitoring node cluster, and sequentially connecting the monitoring node clusters based on the process connection relationship between the equipment to obtain a monitoring node cluster expansion tree;

[0006] Step S200, independently partitioning historical monitoring data to obtain a plurality of historical monitoring sub-data, and classifying the historical monitoring sub-data based on the abnormal marks recorded in each historical monitoring sub-data to obtain normal historical monitoring sub-data and abnormal historical monitoring sub-data;

[0007] Step S300: Map the normal historical monitoring sub-data and abnormal historical monitoring sub-data to the monitoring node cluster expansion tree to obtain several normal monitoring data cluster expansion trees and abnormal monitoring data cluster expansion trees;

[0008] Step S400: Conduct data interval monitoring on each monitoring sub-data in the monitoring data cluster expansion tree, record the interval trigger characteristics of each monitoring sub-data, form the characteristic performance of the data cluster expansion tree, and construct the characteristic performance library with several characteristic performances of the data cluster expansion tree;

[0009] Step S500: Analyze the real-time monitoring data using the characteristic performance library, determine the characteristic performance of the data cluster expansion tree adapted to the real-time monitoring data, and based on the characteristic performance of the data cluster expansion tree, give an early warning of faults.

[0010] In the disclosed embodiments of the present invention, the method for independent partitioning of historical monitoring data includes:

[0011] Step S201: Based on the change of the data parameters corresponding to each device unit in the historical monitoring data over time, construct data parameter curves and align all the data parameter curves corresponding to the device units;

[0012] Step S202: Uniformly split the data parameter curves to form several parameter curve segments. Use the same dynamic time window to scan and analyze the aligned several data parameter curves, calculate the average parameter value for each parameter curve segment within the window, and obtain the average parameter value sequence within the window;

[0013] Step S203: Dynamically adjust the window period of the dynamic time window, and after each adjustment, shift the window according to the window period, compare the average parameter value sequences corresponding to different windows, and determine whether the average parameter value sequences conform to the periodic cycle characteristic. If they conform, perform independent partitioning on each data parameter curve according to the window period at this time.

[0014] In the disclosed embodiments of the present invention, the method for determining whether the average parameter value sequences satisfy the periodic cycle characteristic includes:

[0015] Step S2031: Perform averaging calculation on the average parameter values of the same order in the selected several average parameter value sequences to obtain the reference average parameter value sequence;

[0016] Step S2032: Analyze the matching parameters between each average parameter value sequence and the reference average parameter value sequence, judge the number of the first matching parameters whose matching parameters are greater than or equal to the preset value, calculate the matching situation ratio of the number of the first matching parameters to the total number of all matching parameters. If the matching situation ratio is greater than or equal to the preset value, determine that the average parameter value sequences conform to the periodic cycle characteristic.

[0017] In the embodiments disclosed by the present invention, the method for monitoring the data range of each piece of monitoring data in the monitoring data cluster expansion tree includes:

[0018] Step S301: Based on the variation of the data parameters corresponding to each device unit in the historical monitoring data over time, construct a data parameter curve;

[0019] Step S302: Randomly intercept the data parameter curve several times to obtain several data parameter curve segments. For each data parameter curve segment, construct several horizontal analysis lines. Each horizontal analysis line corresponds to a data parameter. Statistically calculate the length of the curve segment mapped within the preset range above and below the horizontal analysis line, calculate the sum of the lengths of the curve segments corresponding to all horizontal analysis lines, denote it as the total curve segment length, and determine the proportion of the length of the curve segment corresponding to each horizontal analysis line to the total curve segment length;

[0020] Step S303: Based on the preset proportion interval to which the proportion of the curve length corresponding to the horizontal analysis line belongs, determine the data range monitoring length of the data parameter corresponding to the horizontal analysis line, and based on the data range monitoring length, divide the preset range above and below the data parameter into several data monitoring ranges.

[0021] In the embodiments disclosed by the present invention, the method for monitoring the data range of each monitoring sub - data in the monitoring data expansion tree includes:

[0022] Step S401: Determine the intermediate value of each data monitoring range, and recognize the intermediate value as the interval trigger expression value of the data monitoring range;

[0023] Step S402: Determine the data monitoring range to which each monitoring sub - data in the monitoring data cluster expansion tree belongs at different time nodes, and based on the determined data monitoring range, determine the interval trigger expression value corresponding to the corresponding time node, and map the interval trigger expression values corresponding to different time nodes to the monitoring data nodes corresponding to the monitoring data cluster expansion tree to obtain the characteristic performance of the data cluster expansion tree.

[0024] In the embodiments disclosed by the present invention, the method for constructing a characteristic performance library from several characteristic performances of the data cluster expansion tree includes:

[0025] Step S403: Determine the sequence of the interval trigger expressions of the monitoring data nodes in each characteristic performance of the data cluster expansion tree, and determine the trigger interval between the monitoring data nodes triggered by the interval trigger expression to form a monitoring data node trigger sequence;

[0026] Step S404: Using the monitoring data nodes that appear successively in the monitoring data node trigger sequence as the first retrieval classification condition, and the trigger intervals that appear successively as the second retrieval classification condition, classify the feature manifestations of the data cluster expansion tree one by one to obtain a feature manifestation library.

[0027] In the embodiments disclosed in the present invention, the method for analyzing real-time monitoring data using the feature manifestation library includes:

[0028] Step S501: Analyze the real-time monitoring data, map the corresponding real-time monitoring sub-data to the monitoring node cluster expansion tree, and perform data interval monitoring on the real-time monitoring sub-data of each monitoring data node. Using the order of the real-time monitoring sub-data triggered by the interval and the trigger intervals between them as retrieval conditions, find several initial adapted feature manifestations of the data cluster expansion tree in the feature manifestation library.

[0029] Step S502: Determine the interval trigger expression value of each real-time monitoring sub-data, and substitute several interval trigger expression values into different feature manifestations of the data cluster expansion tree for comparison. If the difference in expression values between the corresponding interval trigger expression values is less than or equal to the preset value, then the corresponding interval trigger expression value is determined as the adapted interval trigger expression value. If the proportion of the adapted interval trigger expression values is greater than or equal to the preset value, then the feature manifestation of the data cluster expansion tree at this time is determined as the final adapted feature manifestation of the data cluster expansion tree.

[0030] In the embodiments disclosed in the present invention, an enterprise fault clustering and early warning system based on production equipment data analysis is also disclosed, including:

[0031] The first module is used to determine the equipment units to be monitored, determine several data monitoring nodes corresponding to each equipment unit, classify the several data monitoring nodes into the same monitoring node cluster, and sequentially connect the monitoring node clusters based on the technological connection relationship between the equipment to obtain a monitoring node cluster expansion tree.

[0032] The second module is used to independently partition the historical monitoring data to obtain several historical monitoring sub-data, and classify the historical monitoring sub-data based on the abnormal marks recorded in each historical monitoring sub-data to obtain normal historical monitoring sub-data and abnormal historical monitoring sub-data.

[0033] The third module is used to map the normal historical monitoring sub-data and the abnormal historical monitoring sub-data to the monitoring node cluster expansion tree to obtain several normal monitoring data cluster expansion trees and abnormal monitoring data cluster expansion trees.

[0034] The fourth module is used to monitor the data interval of each monitored sub - data in the monitored data cluster expansion tree, record the interval trigger characteristics of each monitored sub - data, form the characteristic performance of the data cluster expansion tree, and construct a characteristic performance library by using several characteristic performances of the data cluster expansion tree;

[0035] The fifth module is used to analyze the real - time monitored data by using the characteristic performance library, determine the characteristic performance of the data cluster expansion tree adapted to the real - time monitored data, and give an early warning of faults based on the characteristic performance of the data cluster expansion tree.

[0036] The present invention discloses an enterprise fault clustering and early warning method and system based on production equipment data analysis, which relates to the technical field of equipment fault monitoring. It determines the monitoring equipment unit and its data monitoring nodes, and constructs a monitoring node cluster expansion tree based on the process relationship; then, it conducts independent partitioning and abnormal marking classification on the historical monitoring data to obtain normal and abnormal historical monitored sub - data; maps these sub - data to the expansion tree to generate normal and abnormal monitored data cluster expansion trees; extracts trigger characteristics through data interval monitoring to form a characteristic performance library; finally, uses this library to analyze real - time data, match the characteristic performance and give an early warning of faults. By integrating the equipment process relationship and multi - dimensional data, the present invention overcomes the limitations of traditional threshold comparison, realizes systematic and early fault prediction, and improves the reliability of industrial production.

[0037] The technical solution of the present invention will be further described in detail below through the drawings and embodiments. Description of the Drawings

[0038] Figure 1 It is a method step diagram of the enterprise fault clustering and early warning method based on production equipment analysis disclosed in the embodiment of the present invention. Detailed Embodiment

[0039] The technical solution of the present invention will be further described below through the drawings and embodiments.

[0040] The technical solution of the present invention will be clearly and completely described below in conjunction with the drawings and specific embodiments. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention and cannot be construed as limiting the protection scope of the present invention. Those skilled in the art can make some non - essential improvements and adjustments according to the content of the present invention below. In the present invention, unless otherwise clearly defined and limited, the technical terms used in the present invention should be the common meanings understood by those skilled in the art of the present invention.

[0041] Embodiment:

[0042] The present invention discloses an enterprise fault clustering and early warning method based on production equipment data analysis. Refer to Figure 1 , including:

[0043] Step S100, determine the device units to be monitored, and determine a number of data monitoring nodes corresponding to each device unit. Group the number of data monitoring nodes into the same monitoring node cluster, and based on the technological connection relationship between devices, sequentially connect the monitoring node clusters to obtain a monitoring node cluster expansion tree.

[0044] Among them, a monitoring node cluster refers to grouping multiple data monitoring nodes (such as sensors, measurement points) of a device unit into a set to form a logically monitoring unit. In production equipment, each device unit may have multiple monitoring points (such as temperature, pressure, vibration sensors). These monitoring points are grouped into "monitoring node clusters" for unified management and analysis of data. For example, a motor may have three monitoring points: current, rotation speed, and temperature, which together form a monitoring node cluster, facilitating the structured organization of multi-dimensional data of the device and providing a basis for the subsequent construction of the expansion tree.

[0045] The monitoring node cluster expansion tree is a tree-like data structure formed by sequentially connecting multiple monitoring node clusters based on the technological connection relationship between devices. Production equipment is usually connected according to the technological process (such as devices A to B to C in an assembly line). This term describes a model that organizes monitoring node clusters in the technological order, resembling a tree. The root node may be the starting device, and the branches represent subsequent devices and their monitoring points. For example, an assembly line in a factory can be represented as: raw material device (cluster 1) → processing device (cluster 2) → finished product device (cluster 3), forming a tree-like topology. By reflecting the dependencies and data flows between devices through the tree structure, it is convenient to analyze the propagation path of faults.

[0046] In a complex industrial production environment, the equipment system usually consists of multiple interconnected equipment units, and each unit may involve the monitoring of multiple physical parameters, such as temperature, pressure, vibration, etc. The principle of step S100 is to establish a structured data analysis framework for these devices. By systematically organizing the technological relationships between the monitoring points and the devices, it provides a basis for subsequent fault analysis. First, identify the equipment units that need to be monitored, such as motors, pumps, or conveyor belts on the production line. These units are the core components of the production process. Then, identify the corresponding data monitoring nodes for each equipment unit, usually sensors or measurement points, such as the current sensor, speed sensor, and temperature sensor of the motor. These monitoring nodes are the basic units for data acquisition and are responsible for recording the operating status of the equipment in real time. For the convenience of management and analysis, multiple monitoring nodes of the same equipment unit are classified into a "monitoring node cluster" to form a logical monitoring unit. For example, the current, speed, and temperature data of a motor are organized into a cluster, similar to integrating multi-dimensional data into a whole for unified processing. Subsequently, based on the technological connection relationships between the devices, such as the output of device A in the assembly line being the input of device B, the monitoring node clusters are connected in the technological order to form a "monitoring node cluster expansion tree". This tree-like structure takes the starting device of the technological process as the root node, and the branches represent the downstream devices and their monitoring clusters. For example, an assembly line in a factory may be represented as raw material equipment (cluster 1) → processing equipment (cluster 2) → finished product equipment (cluster 3), similar to a topological tree. The principle of this structured modeling draws on the tree-shaped data structure in graph theory. By abstractly representing the production system through nodes (monitoring clusters) and edges (technological relationships), it reflects the dependencies and data flows between the devices. The core role of the monitoring node cluster expansion tree is that it not only retains the monitoring information of a single device but also captures the system-level correlations, enabling fault analysis to expand from the isolated device level to the entire technological process. For example, when a device fails, the tree structure can help track whether the fault will spread to downstream devices or whether it is caused by upstream anomalies. This systematic perspective solves the limitations of single-device analysis in traditional methods and provides a spatial framework for data mapping and feature extraction in subsequent steps. In addition, the construction of the expansion tree also facilitates the visualization and management of data.

[0047] Step S200: Independently partition the historical monitoring data to obtain several historical monitoring sub-data. Based on the anomaly marks recorded in each historical monitoring sub-data, classify the historical monitoring sub-data to obtain normal historical monitoring sub-data and abnormal historical monitoring sub-data.

[0048] Historical monitoring sub - data refers to several data subsets obtained after independently partitioning historical monitoring data. Historical monitoring data is usually long - time - series data (such as sensor readings of a device running for one year). By a certain method (such as based on time windows or periodic features), these data are segmented into independent subsets, and each subset is called "historical monitoring sub - data". For example, when divided by the device operation cycle, the data of each day may be a subset. According to the anomaly markers recorded in the historical monitoring sub - data, it is classified into normal or abnormal subsets.

[0049] Historical monitoring data is a valuable record of industrial equipment operation, usually existing in the form of long - time series, such as temperature, pressure and other data recorded by sensors within a year. These data contain periodicity, noise and abnormal fluctuations, making it difficult to directly use them for failure mode analysis. The principle of step S200 lies in that through systematic data processing, a large amount of historical data is segmented into analyzable subsets, and the normal and abnormal states are distinguished, providing a basis for subsequent feature extraction and pattern construction. First of all, the historical data needs to be "independently partitioned", that is, the continuous time series is segmented into multiple independent "historical monitoring sub - data". This process depends on the identification of data patterns. For example, the device operation may have a fixed cycle (such as the on - off rule of the device every day). The partitioning method may be based on dynamic time windows or periodic detection algorithms (the periodic cycle features are mentioned in the subsequent claims), ensuring that each subset is statistically representative. For example, the data is segmented into daily or shift - by - shift data subsets according to the device operation cycle, and each subset reflects a complete and independent operation state. The purpose of partitioning is to decompose the complex large - data set into smaller and more easily processed chunks while retaining the key behavior patterns. After the partitioning is completed, based on the anomaly markers recorded in each segment of sub - data (such as fault logs, over - limit alarms), it is classified to generate two types of data: normal historical monitoring sub - data and abnormal historical monitoring sub - data. Normal sub - data corresponds to the scenario of stable device operation, for example, the sensor readings are always within the safe range; abnormal sub - data contains faults or abnormal events, such as a sudden increase in temperature or vibration exceeding the standard. The classification process is similar to data annotation in supervised learning and depends on predefined anomaly criteria, which may be generated through manual records or automatic threshold detection. The significance of this classification is that it clearly distinguishes historical data into "healthy" and "sick" states, providing a reference for subsequent feature comparison. Technically, the independent partitioning may involve time - series analysis techniques, such as autocorrelation analysis to detect cycles or Fourier transform to identify frequency components, while the classification may combine simple rule - based judgments or complex state - machine models. The role of step S200 is not only data sorting but also laying a foundation for systematic analysis. By classifying the data into normal and abnormal categories, the system can clearly learn the behavioral differences of the device in different states, providing a clear starting point for the extraction of failure modes.

[0050] Step S300: Map the normal historical monitoring sub-data and the abnormal historical monitoring sub-data to the monitoring node cluster expansion tree to obtain a number of normal monitoring data cluster expansion trees and abnormal monitoring data cluster expansion trees.

[0051] The core principle of step S300 is to combine the historical monitoring sub-data sorted out in step S200 with the device topology structure constructed in step S100 to form a systematic data representation, so as to capture the overall patterns of device behavior in normal and abnormal states. Specifically, in this step, the normal historical monitoring sub-data and the abnormal historical monitoring sub-data are respectively mapped to the monitoring node cluster expansion tree, generating two types of results: normal monitoring data cluster expansion trees and abnormal monitoring data cluster expansion trees. The mapping process is to allocate time series data to the corresponding nodes of the tree structure, and each node (monitoring point) records its value or behavior in the sub-data. For example, a monitoring node cluster may contain the temperature and vibration data of a motor. After mapping, the normal sub-data may show that the temperature is stable at 50 °C and the vibration is within the normal range; while the abnormal sub-data may show that the temperature rises to 100 °C and the vibration exceeds the standard. The significance of this mapping is that it embeds isolated time series data into a spatial structure reflecting process relationships, forming a spatio-temporal combined analysis framework. Technically, the mapping may involve data standardization processing (such as normalization) to ensure the comparability of data at different monitoring points, and it may also be necessary to align timestamps to ensure the correspondence between sub-data and tree nodes. The generated normal monitoring data cluster expansion tree describes the data distribution and the association pattern between nodes in the healthy state of the device system. For example, the data of all nodes are within the expected range and the process flow is stable. On the contrary, the abnormal monitoring data cluster expansion tree reveals the data deviation in the fault state. For example, the data of a certain node is significantly abnormal, or a chain reaction occurs between multiple nodes. In principle, this process is similar to projecting data into a high-dimensional space, using the tree structure to organize multi-dimensional features, combining time domain (data changes over time) and spatial domain (process connections between devices) information. This method solves the limitation of only focusing on a single device or a single parameter in traditional analysis and can capture the propagation path of system-level faults. For example, if the abnormality of an upstream device triggers the fault of a downstream device, the abnormal expansion tree can clearly show this causal relationship. The role of step S\alpha is to provide a structured data representation for subsequent feature extraction, making the fault analysis no longer limited to isolated data points, but based on the dynamic behavior of the entire process flow. In addition, the mapping process also facilitates visualization and diagnosis, and may display the distribution of abnormal nodes through a tree diagram to help engineers quickly locate problems. The two generated expansion trees establish a comparison benchmark for normal and abnormal states and provide direct input for the feature extraction in step S400. In summary, step S300 constructs a systematic data model by integrating historical data with device topology, not only retaining the dynamic characteristics of the data, but also reflecting the process association between devices, laying a key foundation for deeply exploring fault patterns.

[0052] Step S400: Monitor the data range of each monitored sub - data in the monitored data cluster expansion tree, record the interval trigger characteristics of each monitored sub - data, form the feature representation of the data cluster expansion tree, and construct a feature representation library from several feature representations of the data cluster expansion tree.

[0053] Among them, data range monitoring is a process of dividing intervals and monitoring features for each monitored sub - data in the monitored data cluster expansion tree. This is a method for fine - grained analysis of data. The monitored data (such as temperature, pressure) is divided into multiple intervals (such as low temperature, medium temperature, high temperature) according to the numerical range, and it is detected which interval the data falls into. For example, temperature data may be divided into intervals such as [0 - 50°C], [50 - 100°C], etc., and its distribution law is analyzed.

[0054] The interval trigger characteristic is to record the characteristic that the monitored sub - data triggers a specific interval during data range monitoring. When the monitored data falls into a certain predefined interval, this event is recorded as "trigger". For example, when the temperature enters the interval [100 - 150°C], it may trigger a "high temperature" characteristic. The trigger characteristics may include the trigger time, frequency, duration, etc.

[0055] Among them, the feature representation of the data cluster expansion tree is a set of features formed by recording the interval trigger characteristics of the monitored sub - data, which describes the overall behavior of the monitored data cluster expansion tree. This is a comprehensive expression of the data characteristics of all monitored nodes in the expansion tree. For example, an abnormal data cluster expansion tree may be manifested as "node A has frequent high - temperature triggers, and node B has vibration exceeding the limit". The feature representation is a structured summary of these trigger characteristics, which may include the trigger order, time interval, etc.

[0056] The feature representation library is a database or model library formed by collecting and organizing several feature representations of the data cluster expansion tree. This is a "knowledge base" that stores the features of various fault and normal modes. For example, the feature representation library may contain "motor overheating fault characteristics", "pressure anomaly characteristics", etc., and each feature corresponds to a specific performance of the data cluster expansion tree.

[0057] The principle of step S400 is to abstract the behavior patterns in the monitoring data cluster expansion tree into a comparable feature set through refined data analysis and feature extraction, and construct a knowledge base of fault and normal patterns to provide a reference for real-time warning. This step contains multiple sub-processes. First, "data interval monitoring" is performed on each monitoring sub-data in the monitoring data cluster expansion tree (the normal or abnormal tree from step S300). Specifically, the monitoring data (such as temperature, pressure) is divided into multiple intervals according to the numerical range. For example, the temperature is divided into intervals such as 0 - 50°C, 50 - 100°C, 100 - 150°C, etc., and it is detected which interval the data falls into. The principle of this interval division is similar to data binning, which discretizes continuous data into discrete states for facilitating the analysis of its distribution law. Then, the "interval trigger features" of each sub-data are recorded, that is, a feature event is triggered when the data enters a certain interval. For example, when the temperature enters 100 - 150°C, it may trigger the "high temperature" feature, and the recorded features include the trigger time, frequency, duration, etc. These features reflect the dynamic behavior of the data. For example, the frequency of high temperature triggering may indicate a potential overheating risk. Subsequently, the trigger features of all monitoring nodes are integrated to form the "feature performance of the data cluster expansion tree", which is a structured feature set describing the behavior of the entire tree. For example, the feature performance of an abnormal tree may be "node A has frequent high temperature triggers, node B has vibration overlimit, and the trigger interval is 2 seconds", which synthesizes the trigger order, time relationship, and the association between nodes. The generation of this feature performance is similar to feature engineering in pattern recognition, aiming to extract the most discriminative information. Finally, the feature performances of multiple normal and abnormal data cluster expansion trees are collected and organized to form the "feature performance library". This library is a knowledge base that stores templates of various operating modes, such as "motor overheating fault features", "pressure abnormality features", etc., and each feature corresponds to a typical performance of an expansion tree.

[0058] In step S500, the feature performance library is used to analyze the real-time monitoring data, determine the feature performance of the data cluster expansion tree that matches the real-time monitoring data, and based on the feature performance of the data cluster expansion tree, a warning is issued for the fault.

[0059] The principle of step S500 lies in using the feature representation library constructed in step S400 to quickly analyze real-time monitoring data, identify potential faults, and issue early warnings, thereby achieving proactive fault prevention. The process starts with collecting real-time monitoring data of production equipment. These data are similar to historical data and contain multi-dimensional parameters (such as temperature, vibration). First, map the real-time data to the monitoring node cluster expansion tree in step S100 to generate a real-time data cluster expansion tree, which reflects the operating state of the current device system. The mapping process is similar to step S300 to ensure that the data is aligned with the device topology. Subsequently, perform data interval monitoring on the real-time data and extract its "interval trigger features", such as detecting which nodes trigger high temperature or abnormal vibration. These features are extracted in the same way as in step S400, based on the same interval division criteria, to ensure that the real-time features can be compared with the templates in the feature representation library. Next, the system matches the trigger features of the real-time data with the patterns in the feature representation library to find the closest "data cluster expansion tree feature representation". The matching process may be based on the calculation of feature similarity, such as comparing the trigger order, time interval, and numerical deviation, similar to distance metrics (such as Euclidean distance) or classification algorithms (such as nearest neighbor) in pattern recognition. If the features of the real-time data highly match an abnormal pattern in the library, such as being similar to the "motor overheating" feature representation, it is determined that there is a risk of this fault. Finally, based on the matching result, the system issues a warning signal, indicating the possible fault type, location, and severity, such as "Warning: Motor A may be overheating". Technically, this process requires the support of efficient algorithms to meet the real-time requirements. Indexing technology may be used to optimize library queries, or lightweight models may be used to reduce calculation latency.

[0060] In the embodiments disclosed in the present invention, the method for independently partitioning historical monitoring data includes:

[0061] Step S201: Based on the changes in the data parameters corresponding to each device unit in the historical monitoring data over time, construct data parameter curves and align the data parameter curves corresponding to all device units.

[0062] Among them, the data parameter curve is a continuous or discrete curve constructed based on the monitoring data of the device unit changing over time, which is a visual expression of the monitoring data. For example, a temperature sensor records data once per second, and connecting these points forms a temperature curve that changes over time. The data parameter refers to the physical quantity being measured (such as temperature, pressure).

[0063] Step S202: Uniformly split the data parameter curves to form several parameter curve segments, use the same dynamic time window to scan and analyze the aligned several data parameter curves, calculate the average parameter value for each parameter curve segment within the window, and obtain an average parameter value sequence within the window;

[0064] Among them, the average parameter value sequence is an ordered numerical sequence formed by calculating the average value of the parameter curve segments within a dynamic time window. By sliding a time window (such as 1 hour), the average value of the data parameter curve segments within the window is taken to generate a sequence. For example, the average value of the temperature per minute within the window may form a sequence [25.3, 26.1, 25.8, …].

[0065] Step S203: Dynamically adjust the window period of the dynamic time window, and after each adjustment, shift the window according to the window period, compare the average parameter value sequences corresponding to different windows, and determine whether there is a periodic cycle feature between the average parameter value sequences. If so, perform an independent division on each data parameter curve according to the current window period.

[0066] Among them, the periodic cycle feature means that the average parameter value sequences show periodic repetition characteristics. It refers to that the data presents a similar change pattern within a specific time period (such as every day, every hour). For example, the temperature of the device rises at 8 am and drops at 8 pm every day, and this rule is recognized as a periodic cycle feature.

[0067] In the embodiments disclosed in the present invention, the method for determining whether the average parameter value sequences satisfy the periodic cycle feature includes:

[0068] Step S2031: Perform an averaging calculation on the average parameter values of the same order in a selected number of average parameter value sequences to obtain a reference average parameter value sequence;

[0069] Step S2032: Analyze the matching parameter between each average parameter value sequence and the reference average parameter value sequence, judge the number of first matching parameters whose matching parameter is greater than or equal to a preset value, and calculate the matching situation ratio of the number of first matching parameters to the total number of all matching parameters. If the matching situation ratio is greater than or equal to the preset value, it is determined that there is a periodic cycle feature between the average parameter value sequences.

[0070] Among them, the expression for calculating the matching parameter between the average parameter value sequence and the reference average parameter value sequence is:

[0071]

[0072] Where, W is the anastomosis parameter, β(i) is the difference parameter value anastomosis judgment function. If the difference parameter value between the average parameter value corresponding to the i-th order and the reference average parameter value is less than or equal to the preset value, β(i) outputs 1; otherwise, it outputs 0. n is the total number of average parameter values in the average parameter value sequence or the total number of reference average parameter values in the reference average parameter value sequence. α(x) is the sequence continuous change anastomosis judgment function. If the change direction of the average parameter value between the x-th orders is the same as the change direction of the reference average parameter value, α(x) outputs 1; otherwise, it outputs 0. K is the continuous change anastomosis influence adjustment coefficient, and b is the continuous change anastomosis influence adjustment constant.

[0073] This formula accurately identifies the periodic pattern of equipment data through the dual constraints of static numerical deviation and dynamic trend continuity, and solves the defect that the traditional threshold method ignores the time series correlation. The parameters K and b provide flexibility and can adjust the strictness of trend matching according to specific scenarios, and are applicable to the early warning of multi-device linkage failures in complex industrial environments.

[0074] In the embodiments disclosed in the present invention, the method for monitoring the data interval of each monitoring data in the monitoring data cluster expansion tree includes:

[0075] Step S301: Based on the change of the data parameters corresponding to each equipment unit in the historical monitoring data over time, construct a data parameter curve.

[0076] Step S302: Randomly intercept the data parameter curve several times to obtain several data parameter curve segments. For each data parameter curve segment, construct several horizontal analysis lines. Each horizontal analysis line corresponds to a data parameter. Statistically, the length of the curve segment mapped within the preset interval above and below the horizontal analysis line is calculated, and the sum of the curve segment lengths corresponding to all horizontal analysis lines is recorded as the total curve segment length. Then, determine the proportion of the curve segment length corresponding to each horizontal analysis line to the total curve segment length.

[0077] Step S303: Based on the preset proportion interval to which the proportion of the curve length corresponding to the horizontal analysis line belongs, determine the data interval monitoring length of the data parameter corresponding to the horizontal analysis line, and based on the data interval monitoring length, divide the preset range above and below the data parameter into several data monitoring intervals.

[0078] In the embodiments disclosed in the present invention, the method for monitoring the data interval of each monitoring sub-data in the monitoring data expansion tree includes:

[0079] Step S401: Determine the intermediate value of each data monitoring interval and regard the intermediate value as the interval trigger expression value of the data monitoring interval.

[0080] Step S402: Determine the data monitoring intervals to which each monitoring sub - data in the monitoring data cluster expansion tree belongs at different time nodes. Based on the determined data monitoring intervals, determine the interval trigger expression values corresponding to the time nodes, and map the interval trigger expression values corresponding to different time nodes to the monitoring data nodes corresponding to the monitoring data cluster expansion tree to obtain the characteristic performance of the data cluster expansion tree.

[0081] In the embodiments disclosed in the present invention, the method of constructing a characteristic performance library from several characteristic performances of the data cluster expansion tree includes:

[0082] Step S403: Determine the sequence of the interval trigger expressions of the monitoring data nodes in each characteristic performance of the data cluster expansion tree, and determine the trigger intervals between the monitoring data nodes triggered by the interval trigger expressions to form a monitoring data node trigger sequence.

[0083] Step S404: Use the monitoring data nodes that appear in sequence in the monitoring data node trigger sequence as the first retrieval classification condition, and the trigger intervals that appear in sequence as the second retrieval classification condition to classify the characteristic performances of the data cluster expansion tree one by one to obtain the characteristic performance library.

[0084] In the embodiments disclosed in the present invention, the method of analyzing real - time monitoring data using the characteristic performance library includes:

[0085] Step S501: Analyze the real - time monitoring data, map the corresponding real - time monitoring sub - data to the monitoring node cluster expansion tree, and perform data interval monitoring on the real - time monitoring sub - data of each monitoring data node. Use the order of the real - time monitoring sub - data triggered by the interval trigger expression and the trigger intervals between them as retrieval conditions to find several initially adapted characteristic performances of the data cluster expansion tree in the characteristic performance library.

[0086] Step S502: Determine the interval trigger expression values of each real - time monitoring sub - data, and substitute the several interval trigger expression values into different characteristic performances of the data cluster expansion tree for comparison. If the difference in the expression values between the corresponding interval trigger expression values is less than or equal to the preset value, then determine the corresponding interval trigger expression value as the adapted interval trigger expression value. If the proportion of the adapted interval trigger expression values is greater than or equal to the preset value, then determine the characteristic performance of the data cluster expansion tree at this time as the finally adapted characteristic performance of the data cluster expansion tree.

[0087] In the embodiments disclosed in the present invention, an enterprise fault clustering early - warning system based on production equipment data analysis is also disclosed, including:

[0088] The first module is used to determine the device units to be monitored, determine a number of data monitoring nodes corresponding to each device unit, classify the number of data monitoring nodes into the same monitoring node cluster, and sequentially connect the monitoring node clusters based on the process connection relationship between devices to obtain a monitoring node cluster expansion tree;

[0089] The second module is used to perform independent partitioning on historical monitoring data to obtain a number of historical monitoring sub-data, and classify the historical monitoring sub-data based on the anomaly marks recorded in each historical monitoring sub-data to obtain normal historical monitoring sub-data and abnormal historical monitoring sub-data;

[0090] The third module is used to map the normal historical monitoring sub-data and the abnormal historical monitoring sub-data to the monitoring node cluster expansion tree to obtain a number of normal monitoring data cluster expansion trees and abnormal monitoring data cluster expansion trees;

[0091] The fourth module is used to perform data interval monitoring on each monitoring sub-data in the monitoring data cluster expansion tree, record the interval trigger characteristics of each monitoring sub-data, form a data cluster expansion tree feature representation, and construct a number of data cluster expansion tree feature representations into a feature representation library;

[0092] The fifth module is used to analyze real-time monitoring data using the feature representation library, determine the data cluster expansion tree feature representation adapted to the real-time monitoring data, and issue a fault warning based on the data cluster expansion tree feature representation.

[0093] Through the description of the above embodiments, those skilled in the art can clearly understand that the present invention can be implemented through hardware, or can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product, and the software product can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various implementation scenarios of the present invention.

[0094] The present invention discloses an enterprise fault clustering and early warning method and system based on production equipment data analysis, which relates to the technical field of equipment fault monitoring. It determines the monitored equipment units and their data monitoring nodes, and constructs a monitoring node cluster expansion tree based on the process relationship; then, it conducts independent partitioning and anomaly marking classification on the historical monitoring data to obtain normal and abnormal historical monitoring sub-data; maps these sub-data to the expansion tree to generate normal and abnormal monitoring data cluster expansion trees; extracts triggering features through data interval monitoring to form a feature performance library; finally, uses this library to analyze real-time data, match the feature performance and give early warnings of faults. By integrating the equipment process relationship and multi-dimensional data, the present invention overcomes the limitations of traditional threshold comparison, realizes systematic and anticipatory fault prediction, and improves the reliability of industrial production.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An enterprise fault clustering early warning method based on production equipment data analysis, characterized in that, Including: Step S100: Determine the device units to be monitored, and determine a number of data monitoring nodes corresponding to each device unit. Group the number of data monitoring nodes into the same monitoring node cluster, and based on the technological connection relationship between devices, sequentially connect the monitoring node clusters to obtain a monitoring node cluster expansion tree. Step S200: Independently partition the historical monitoring data to obtain a number of historical monitoring sub-data. Based on the anomaly marks recorded in each historical monitoring sub-data, classify the historical monitoring sub-data to obtain normal historical monitoring sub-data and abnormal historical monitoring sub-data. Step S300: Map the normal historical monitoring sub-data and abnormal historical monitoring sub-data onto the monitoring node cluster expansion tree to obtain a number of normal monitoring data cluster expansion trees and abnormal monitoring data cluster expansion trees. Step S400: Monitor the data interval of each monitoring sub-data in the monitoring data cluster expansion tree, and record the interval trigger characteristics of each monitoring sub-data to form the characteristic performance of the data cluster expansion tree. Construct the characteristic performance of the data cluster expansion tree into a characteristic performance library. Step S500: Analyze the real-time monitoring data using the characteristic performance library, determine the characteristic performance of the data cluster expansion tree adapted to the real-time monitoring data, and based on the characteristic performance of the data cluster expansion tree, give an early warning of faults.

2. The enterprise fault clustering and early warning method based on production equipment data analysis according to claim 1, characterized in that The method for independently partitioning the historical monitoring data includes: Step S201: Based on the change of the data parameters corresponding to each device unit in the historical monitoring data over time, construct data parameter curves and align the data parameter curves corresponding to all device units. Step S202: Uniformly split the data parameter curves to form a number of parameter curve segments. Use the same dynamic time window to scan and analyze the aligned data parameter curves, calculate the average parameter value of each parameter curve segment within the window to obtain the average parameter value sequence within the window. Step S203: Dynamically adjust the window period of the dynamic time window, and after each adjustment, shift the window according to the window period, compare the average parameter value sequences corresponding to different windows, and determine whether the average parameter value sequences conform to the periodic cycle characteristic. If they conform, perform independent partitioning on each data parameter curve according to the window period at this time.

3. The enterprise fault clustering and early warning method based on production equipment data analysis according to claim 2, wherein The method for determining whether the average parameter value sequences satisfy the periodic cycle characteristic includes: Step S2031: Calculate the average value of the average parameter values of the same order in the selected average parameter value sequences to obtain a reference average parameter value sequence. Step S2032: Analyze the matching parameters between each average parameter value sequence and the reference average parameter value sequence, and judge the number of the first matching parameters whose matching parameters are greater than or equal to the preset value. Calculate the matching situation ratio of the number of the first matching parameters to the total number of all matching parameters. If the matching situation ratio is greater than or equal to the preset value, determine that the average parameter value sequences conform to the periodic cycle characteristic.

4. The enterprise fault clustering and early warning method based on production equipment data analysis according to claim 1, characterized in that The method for monitoring the data interval of each monitoring data in the monitoring data cluster expansion tree includes: Step S301: Based on the change of the data parameters corresponding to each device unit in the historical monitoring data over time, construct data parameter curves. Step S302: Randomly intercept the data parameter curve several times to obtain several data parameter curve segments. For each data parameter curve segment, construct several horizontal analysis lines. Each horizontal analysis line corresponds to a data parameter. Statistically calculate the length of the curve segment mapped within the preset interval above and below the horizontal analysis line, calculate the sum of the curve segment lengths corresponding to all horizontal analysis lines, denote it as the total curve segment length, and determine the proportion of the curve segment length corresponding to each horizontal analysis line to the total curve segment length; Step S303: Based on the preset proportion interval to which the proportion of the curve length corresponding to the horizontal analysis line belongs, determine the data interval monitoring length of the data parameter corresponding to the horizontal analysis line, and based on the data interval monitoring length, divide the preset range above and below the data parameter into intervals to obtain several data monitoring intervals.

5. The enterprise fault clustering early warning method based on production equipment data analysis according to claim 4, characterized in that The method for performing data interval monitoring on each monitoring sub-data in the monitoring data expansion tree includes: Step S401: Determine the intermediate value of each data monitoring interval, and regard the intermediate value as the interval trigger expression value of the data monitoring interval; Step S402: Determine the data monitoring interval to which each monitoring sub-data in the monitoring data cluster expansion tree belongs at different time nodes, and based on the determined data monitoring interval, determine the interval trigger expression value corresponding to the corresponding time node, and map the interval trigger expression values corresponding to different time nodes to the monitoring data nodes corresponding to the monitoring data cluster expansion tree to obtain the characteristic performance of the data cluster expansion tree.

6. The enterprise fault clustering and early warning method based on production equipment data analysis according to claim 1, characterized in that, The method for constructing several characteristic performances of the data cluster expansion tree into a characteristic performance library includes: Step S403: Determine the order of the interval trigger expressions of the monitoring data nodes in each characteristic performance of the data cluster expansion tree, and determine the trigger interval between the monitoring data nodes triggered by the interval trigger expression to form a monitoring data node trigger sequence; Step S404: Use the monitoring data nodes that appear in sequence in the monitoring data node trigger sequence as the first retrieval classification condition, and use the trigger intervals that appear in sequence as the second retrieval classification condition to classify the characteristic performances of the data cluster expansion tree one by one to obtain the characteristic performance library.

7. The enterprise fault clustering early warning method based on production equipment data analysis according to claim 6, characterized in that, The method for analyzing real-time monitoring data using the characteristic performance library includes: Step S501: Analyze the real-time monitoring data, map the corresponding real-time monitoring sub-data to the monitoring node cluster expansion tree, and perform data interval monitoring on the real-time monitoring sub-data of each monitoring data node. Use the order of the real-time monitoring sub-data triggered by the interval trigger expression and the trigger interval between them as retrieval conditions to find several initial matching characteristic performances of the data cluster expansion tree in the characteristic performance library; Step S502: Determine the interval trigger expression value of each real-time monitoring sub-data, and substitute several interval trigger expression values into different characteristic performances of the data cluster expansion tree for comparison. If the difference in the expression values between the corresponding interval trigger expression values is less than or equal to the preset value, then the corresponding interval trigger expression value is recognized as the matching interval trigger expression value. If the proportion of the matching interval trigger expression values is greater than or equal to the preset value, then the characteristic performance of the data cluster expansion tree at this time is recognized as the final matching characteristic performance of the data cluster expansion tree.

8. An enterprise fault clustering early warning system based on production equipment data analysis, characterized in that, including: The first module is used to determine the device units to be monitored, determine several data monitoring nodes corresponding to each device unit, classify the several data monitoring nodes into the same monitoring node cluster, and sequentially connect the monitoring node clusters based on the technological connection relationship between devices to obtain a monitoring node cluster expansion tree; The second module is used to perform independence partitioning on historical monitoring data to obtain several historical monitoring sub-data, and classify the historical monitoring sub-data based on the anomaly marks recorded in each historical monitoring sub-data to obtain normal historical monitoring sub-data and abnormal historical monitoring sub-data; The third module is used to map the normal historical monitoring sub-data and the abnormal historical monitoring sub-data to the monitoring node cluster expansion tree to obtain several normal monitoring data cluster expansion trees and abnormal monitoring data cluster expansion trees; The fourth module is used to perform data interval monitoring on each monitoring sub-data in the monitoring data cluster expansion tree, record the interval trigger characteristics of each monitoring sub-data, form a data cluster expansion tree feature representation, and construct several data cluster expansion tree feature representations into a feature representation library; The fifth module is used to analyze the real-time monitoring data using the feature representation library, determine the data cluster expansion tree feature representation adapted to the real-time monitoring data, and issue a fault warning based on the data cluster expansion tree feature representation.

Images