Monitoring data fusion analysis method and device, equipment and storage medium

Abstract

The invention relates to a monitoring data fusion analysis method, device and equipment and a storage medium, and the method comprises the steps: obtaining multi-source heterogeneous data of production equipment and a sensing device, and adding a metadata label to the multi-source heterogeneous data to obtain monitoring data; performing protocol analysis on the monitoring data based on a preset protocol rule to generate a protocol data sequence; performing multi-source aggregation and space-time alignment on the protocol data sequence based on a preset time window to generate a fusion feature set, and calculating a dynamic association degree between the fusion feature set and the multi-source heterogeneous data; according to the fault early warning management and control method and system, information islands in a factory monitoring system can be broken through, discrete multi-source data are converted into systematic operation and maintenance knowledge, and the accuracy and timeliness of fault prediction are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of monitoring data, and in particular to a monitoring data fusion analysis method, device, equipment and storage medium. BACKGROUND

[0002] With the in-depth promotion of intelligent manufacturing and industrial internet technology, modern factories are accelerating the deployment of various production equipment, sensing devices and process monitoring systems, forming a large-scale multi-source heterogeneous data environment. However, the core dilemma faced by the current factory monitoring system is that due to significant differences in communication protocols, data formats and sampling time sequences of different equipment and systems, multi-source data is difficult to effectively integrate and uniformly analyze, forming a serious information island. This fundamental data barrier makes it impossible for the factory to conduct deep fusion analysis and accurate prediction of the production state, restricting the implementation of predictive maintenance of equipment and the intelligent management and control level of the production process. SUMMARY

[0003] The main purpose of the present application is to provide a monitoring data fusion analysis method, device, equipment and storage medium, which can break the information island in the factory monitoring system and convert discrete multi-source data into systematic operation and maintenance knowledge, thereby improving the accuracy and timeliness of fault prediction.

[0004] To achieve the above purpose, the present application provides a monitoring data fusion analysis method, comprising: obtaining multi-source heterogeneous data of production equipment and sensing devices, and adding metadata tags to the multi-source heterogeneous data to obtain monitoring data; performing protocol analysis on the monitoring data based on a preset protocol rule to generate a protocol data sequence; performing multi-source aggregation and space-time alignment on the protocol data sequence based on a preset time window to generate a fusion feature set, and calculating a dynamic correlation degree between the fusion feature set and the multi-source heterogeneous data; performing fault warning analysis according to the dynamic correlation degree and the fusion feature set to obtain fault warning management and control information.

[0005] Further, the obtaining of the multi-source heterogeneous data of the production equipment and the sensing devices, and the adding of the metadata tags to the multi-source heterogeneous data to obtain the monitoring data, comprises: obtaining multi-source heterogeneous data from data interfaces corresponding to the production equipment and the sensing devices; extracting a metadata tag template from a pre-set tag module library according to the data interface, and analyzing the multi-source heterogeneous data to obtain collection time information and equipment identity information; filling the collection time information and the equipment identity information into the metadata tag template to form the metadata tag; binding the metadata tag with the multi-source heterogeneous data to generate the monitoring data.

[0006] Further, the protocol analysis is performed on the monitoring data based on a preset protocol rule to generate a protocol data sequence, including: initially separating the monitoring data according to a separation rule of the preset protocol rule to obtain data payload and a protocol identifier; extracting a data field structure definition corresponding to the protocol identifier from the preset protocol rule; performing field positioning and segmentation on the data payload according to the data field structure definition to obtain to-be-verified field data; performing validity verification and format conversion on the to-be-verified field data based on a format specification rule of the preset protocol rule to obtain converted field data; arranging and assembling all the converted field data according to a protocol sequence rule of the preset protocol rule to generate the protocol data sequence.

[0007] Further, the multi-source aggregation and space-time alignment are performed on the protocol data sequence based on a preset time window to generate a fusion feature set, including: time grouping the protocol data sequence according to the preset time window to obtain initial data groups; sorting and interval interpolating each initial data group according to a timestamp interval of the preset time window to obtain aligned grouping data; respectively analyzing statistical features of numerical fields, correlation features between different fields, and frequency features of event fields of the aligned grouping data; integrating the statistical features, the correlation features, and the frequency features of all the grouping data according to a preset feature combination rule to generate the fusion feature set.

[0008] Further, the dynamic correlation degree between the fusion feature set and the multi-source heterogeneous data is calculated, including: extracting fusion feature data from the fusion feature set, and extracting corresponding original feature data from the multi-source heterogeneous data according to the fusion feature data; performing bias analysis on the fusion feature data and the original feature data to form a difference data sequence; comparing each bias value in the difference data sequence with a preset correlation degree threshold; when the bias value is less than the correlation degree threshold, outputting that the fusion feature data and the original feature data are in a strong correlation state, otherwise outputting that they are in a weak correlation state; According to the preset time window, the strong correlation state and the weak correlation state are dynamically aggregated, and a dynamic correlation degree is output.

[0009] Further, the fault early warning analysis is performed according to the dynamic correlation degree and the fusion feature set, and fault early warning management information is obtained, including: Each feature item in the fusion feature set is matched with a preset fault feature library, and if the matching is successful, the feature item is marked as a to-be-judged feature; The dynamic correlation degree corresponding to each to-be-judged feature is extracted, and the dynamic correlation degree is compared with a preset early warning judgment threshold; If the dynamic correlation degree is lower than the preset early warning judgment threshold, the to-be-judged feature is excluded from the early warning judgment process; If the dynamic correlation degree is not lower than the preset early warning judgment threshold, the to-be-judged feature is marked with an early warning identifier, and the to-be-judged feature is added to an early warning feature set; When the number of early warning identifiers in the early warning feature set exceeds a preset identifier threshold, an early warning signal is generated; The fault early warning management information is generated according to the early warning signal and the early warning feature set.

[0010] Further, the matching of each feature item in the fusion feature set with the preset fault feature library, if the matching is successful, the feature item is marked as a to-be-judged feature, includes: Each feature item in the fusion feature set is extracted; Each feature item is compared with a fault feature item in the preset fault feature library one by one; If the feature item meets the identification requirement and the feature value range requirement of any fault feature item, it is determined that the feature item matches successfully; Each feature item that matches successfully is marked as the to-be-judged feature.

[0011] The application also provides a monitoring data fusion analysis device applied to the monitoring data fusion analysis method. The acquisition module is used to acquire multi-source heterogeneous data of production equipment and sensing devices, and to add metadata tags to the multi-source heterogeneous data to obtain monitoring data; The analysis module is used to perform protocol analysis on the monitoring data based on a preset protocol rule to generate a protocol data sequence; The association module is configured to perform multi-source aggregation and space-time alignment on the protocol data sequence based on a preset time window, generate a fusion feature set, and calculate a dynamic correlation degree between the fusion feature set and the multi-source heterogeneous data; The processing module is configured to perform fault early warning analysis according to the dynamic correlation degree and the fusion feature set, and obtain fault early warning management and control information.

[0012] The application further provides a monitoring data fusion analysis system, comprising: A memory configured to store a program; A processor configured to execute the program and implement each step of the monitoring data fusion analysis method according to any one of the preceding embodiments.

[0013] The application further provides a storage medium storing computer instructions configured to enable a computer to execute the method according to any one of the preceding embodiments.

[0014] The monitoring data fusion analysis method, device, equipment and storage medium provided by the application have the following beneficial effects: By establishing a standardized protocol analysis and data alignment mechanism, multi-source heterogeneous data from different protocols and time sequence references are effectively integrated, and information silos in the factory monitoring system are broken. By constructing a fusion feature set and performing dynamic correlation analysis, the internal relationship between multiple feature data can be deeply mined, deep perception of the health status of equipment and early warning of abnormal trends can be realized. Discrete multi-source data is converted into systematic operation and maintenance knowledge, which significantly improves the accuracy and timeliness of fault prediction, and provides reliable technical support for realizing predictive maintenance and intelligent management and control. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a monitoring data fusion analysis method flowchart provided by the application; Figure 2 is a monitoring data fusion analysis device structure diagram provided by the application; Figure 3 is a monitoring data fusion analysis system structure diagram provided by the application.

[0016] The implementation of the application, functional features and advantages will be further described with reference to the accompanying drawings. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical scheme and advantages of the application more clear, the application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and do not limit the application.

[0018] The application will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0019] Referring to Figure 1 The application provides a monitoring data fusion analysis method, comprising: Step S1: acquiring multi-source heterogeneous data of production equipment and sensing devices, and attaching metadata tags to the multi-source heterogeneous data to obtain monitoring data; Specifically, the production equipment and sensing devices specifically include programmable logic controllers, data acquisition and monitoring control systems, various industrial sensors, video monitoring equipment and the like. The original data generated by these sources presents significant heterogeneity in physical interfaces, communication protocols, data formats and sampling periods. Metadata tags are attached to each original data stream. The generation of the metadata tags depends on a preconfigured tag rule library. According to the type of data interface, the key context information is parsed from the message or the interface attribute immediately after the original data message, mainly including the accurate time stamp of data acquisition and the unique identifier of the device generating the data. These information is filled into the corresponding fields of the tag template to form a structured metadata tag. The tag is bound with this original data message to form a tagged monitoring data.

[0020] Step S2: protocol parsing of the monitoring data based on preset protocol rules to generate a protocol data sequence; Specifically, according to the protocol identifier carried by the monitoring data, the corresponding parsing rules are retrieved and loaded from a pre-defined protocol rule library. The rule library contains detailed definitions for various industrial protocols such as OPC UA, Modbus, Profibus, etc., including message structure, starting position and length of fields, data type and encoding method. According to the rules, the original data payload is unpacked and field segmented to extract the original numerical or status information. Strict format conversion and validity check are performed on these original field values, such as converting hexadecimal string to floating point number, mapping status code to pre-defined enumeration value, and checking whether the value is within a reasonable physical range. The field values that pass the check are combined into a clear structured record according to the predetermined data structure. The record inherits the metadata tag information in the original monitoring data. The generated structured records are arranged in chronological order, and finally output as a time-sequenced protocol data sequence.

[0021] Step S3: multi-source aggregation and space-time alignment of the protocol data sequence based on a preset time window to generate a fusion feature set, and calculating a dynamic correlation degree between the fusion feature set and the multi-source heterogeneous data; Specifically, a preset sliding time window mechanism is introduced as the basic unit of data aggregation. The protocol data sequence is divided by time window, and for the data in each window, it is classified according to its device identifier, and multiple structured records from the same monitoring entity (such as a machine tool) but possibly from different sensors (such as temperature, vibration, current) are aggregated together. Due to the slight deviation of the sampling time of these records, an interpolation algorithm based on adjacent time data is used to ensure that all data points are aligned to a unified reference time on the time axis. On the basis of the data grouping of space-time alignment, multi-dimensional feature extraction is carried out. This includes calculating the statistical features of numerical data within the window, analyzing the instantaneous correlation features between different data channels, and counting the occurrence frequency of specific events or alarms within the window. Combining these feature values revealed from different dimensions of the device running state, a multi-dimensional feature vector representing the comprehensive state of the entity within the time window is formed. The feature vectors of all entities form a fusion feature set. Further, to evaluate the dynamic reliability of the fusion features, the method calculates the dynamic correlation degree between its feature values and the evolution trend of the original data in the corresponding time window.

[0022] Step S4: performing fault early warning analysis according to the dynamic correlation degree and the fusion feature set to obtain fault early warning management information.

[0023] Specifically, each feature item in the fusion feature set is matched with a preset fault feature knowledge base. The knowledge base contains the feature patterns and their threshold ranges corresponding to various faults before and after their occurrence in history. If the value of the current feature item meets the description of a certain fault pattern, it is preliminarily marked as a to-be-judged feature. The dynamic correlation degree is introduced as an important noise filtering and confidence judgment basis. The dynamic correlation degree value corresponding to each to-be-judged feature is extracted and compared with the preset early warning judgment threshold. If the correlation degree is too low, it indicates that the feature may be caused by data noise or transient interference, rather than the real device state degradation trend, so it is excluded from the current early warning process. For the to-be-judged features with correlation degree meeting the requirements, their early warning value is confirmed, they are labeled with early warning identifiers and added to the early warning feature set. The set is continuously monitored, and when the number of features with valid early warning identifiers exceeds the preset threshold, it indicates that multiple evidence sources collectively indicate abnormalities, and an early warning signal is triggered. The early warning signal, specific early warning features and their associated device identifiers are combined to generate fault early warning management information.

[0024] The application provides a monitoring data fusion analysis method, which effectively integrates multi-source heterogeneous data from different protocols and time sequence references by establishing a standardized protocol analysis and data alignment mechanism, and breaks the information island in the factory monitoring system. By constructing a fusion feature set and performing dynamic correlation analysis, the internal relationship between multiple feature data can be deeply mined, the depth perception of the equipment health state is realized, and the early warning of abnormal trends is realized. Discrete multi-source data is converted into systematic operation and maintenance knowledge, which significantly improves the accuracy and timeliness of fault prediction, and provides reliable technical support for realizing predictive maintenance and intelligent management and control.

[0025] In one embodiment, the multi-source heterogeneous data of the production equipment and the sensing device are acquired, and metadata tags are attached to the multi-source heterogeneous data to obtain monitoring data, comprising: Acquiring multi-source heterogeneous data from the data interface corresponding to the production equipment and the sensing device; Wherein, the multi-source heterogeneous data refers to a set of original data obtained through the above different interfaces, which has differences in structure and semantics.

[0026] According to the data interface, a metadata tag template is extracted from the preset tag module library, and the multi-source heterogeneous data is parsed to obtain collection time information and equipment identity information; Specifically, while acquiring the original data stream, according to the data interface type identifier used for establishing the connection, a matching query is performed in the preset tag module library. The interface type (such as "Siemens S7-1500 OPC UA", "Modbus RTU temperature sensor") is used as an index, and the corresponding standardized metadata tag template is stored. After successful matching, the template is instantiated and loaded into the memory. The buffered original multi-source heterogeneous data is subjected to preliminary parsing to extract key context information. For network packets containing protocol headers, the encoded timestamp field and device address field are read from the fixed offset of the header. For data rows read from SCADA system log files, the time string and device name code embedded in the text are matched through a predefined regular expression pattern. The collection time information needs special processing of clock synchronization. If the data itself does not carry a high-precision timestamp, the system clock at the time when the data is received is used by the collection agent, and the time source is marked. The equipment identity information is obtained by mapping the fixed identification field in the interface connection parameters or the protocol.

[0027] The collection time information and the equipment identity information are filled into the metadata tag template to form the metadata tag; The metadata tag is associated and bound with the multi-source heterogeneous data to generate the monitoring data.

[0028] The method provided by the embodiment solves the problems of disordered multi-source data sources and difficult traceability by attaching standardized metadata tags to original data streams from different interfaces, thereby establishing a unified identity and time context for each piece of data. The heterogeneous original data is converted into a fused feature set with consistent semantics and timing by presetting protocol rules to analyze and standardize the tagged data, aligning to a unified time window for multi-dimensional feature aggregation, thereby realizing deep fusion and unified representation of cross-protocol and cross-device data. Through multi-level fault early warning analysis based on the fused feature set and its dynamic correlation degree, and using the correlation degree index to effectively screen the preliminary early warning features, the accuracy and reliability of the early warning results are improved while realizing early fault early warning, and false positives caused by data noise or transient interference are effectively reduced.

[0029] In one embodiment, the protocol analysis of the monitoring data based on the preset protocol rules generates protocol data sequences, including: According to the separation rule of the preset protocol rule, the monitoring data is initially separated to obtain data payload and a protocol identifier; Specifically, the metadata header of the monitoring data is analyzed to read preliminary information about the data source interface type. According to the preliminary information, the corresponding separation rule is indexed from the preset protocol rule. The rule may indicate that the protocol identifier directly exists in a specific field of the metadata header, or may need to analyze the start byte mode of the original data body to determine. For example, one rule provides that if the data source interface type is "Ethernet Socket", the first seven bytes of the original data body are checked to see if they are ASCII characters MODBUS, which are used as the protocol identifier. Another rule provides that if the data comes from an OPCUA subscription channel, the protocol identifier is fixed as UA_DataValue. After applying the rule, a protocol identifier string is obtained. At the same time, the original data body is completely stripped out and cached as an independent data payload.

[0030] The data field structure definition corresponding to the protocol identifier is extracted from the preset protocol rule; According to the data field structure definition, field positioning and segmentation are performed on the data payload to obtain to-be-verified field data; Specifically, each field descriptor in the data field structure definition is sequentially traversed. For the first entry in the descriptor list, for example, the status word field is indicated to start at offset 0 bytes, length 2 bytes, starting from the 0th element of the data payload byte array, 2 consecutive bytes are read, this portion of bytes is copied and stored as an independent binary data block, while an ad hoc identifier is attached to the block as specified by the field descriptor. Moving to the next field descriptor, when the descriptor indicates that the temperature reading field starts at offset 2 bytes, length 4 bytes, 4 consecutive bytes are read starting from the 2nd element of the data payload byte array, another independent binary data block is generated and an identifier is attached. This process is looped until all fields described in the structure definition are located and segmented from the data payload. For some fields with variable length or conditional existence, the structure definition contains corresponding judgment logic, for example, according to the value of the previously parsed data length field, the length of the next field is dynamically determined. After the segmentation operation is completed, the data payload byte stream is converted into a set of field data to be verified.

[0031] Perform validity verification and format conversion on the field data to be verified based on the format specification rules of the preset protocol rules to obtain converted field data; Specifically, according to the field identifier, the corresponding detailed rules are searched in the format specification rule library. The verification process first performs format compliance checking: confirms whether the byte length of the data block conforms to the expectation, and excludes truncation errors that may occur during transmission. Then, logical validity verification is performed: for example, for a field representing the status of a device switch, its original value may be one byte, and the verification rule requires that its value must be a legal key value in a predefined dictionary (such as 0x00 representing "off" and 0x01 representing "on"), otherwise it is considered invalid. For a floating-point number field representing a temperature value, the verification rule compares the decoded value with a reasonable physical range (such as -50°C to 200°C), and values outside this range are considered suspicious or invalid. If the field value fails a certain verification, it is handled according to a preset strategy, such as recording an error log, filling in a safe default value, or marking an exception flag. For field values that pass the verification, format conversion is immediately performed. Conversion is performed according to the rules: convert 16-bit integers in big-endian order to system-standard little-endian integers; convert fault codes encoded by a specific manufacturer to a unified enumeration type; convert non-standard floating-point representations to IEEE 754 double-precision format; or interpret the original bytes directly as UTF-8 encoded strings. The converted values are assigned explicit data types and units. After all fields are processed in turn, the converted field data is obtained.

[0032] Arrange and assemble all the converted field data according to the protocol sequence rules of the preset protocol rules to generate the protocol data sequence.

[0033] The method provided by the embodiment ensures accurate identification and preliminary decoupling of different communication protocol data by extracting protocol identifiers and data payloads from the monitoring data according to separation rules. The data payload is positioned and segmented based on the data field structure definition loaded based on the protocol identifier, the original byte stream is converted into discrete field units to be verified, and accurate conversion from an unstructured byte sequence to a structured data element is realized. The data value from different devices is unified into an internal standard format while abnormal data is removed, and the reliability and consistency of data quality are significantly improved through validity verification and standardized format conversion of the field data to be verified according to the format specification rules.

[0034] In one embodiment, the multi-source aggregation and space-time alignment of the protocol data sequence based on the preset time window generates a fusion feature set, including: time grouping of the protocol data sequence according to the preset time window to obtain initial data groups; sorting and interval interpolation of each initial data group according to the timestamp interval of the preset time window to obtain aligned grouping data; Specifically, all data records in the initial data group are sorted. The basis for sorting is the accurate timestamp field carried by each record. After sorting, the records in the group are arranged in chronological order. The timestamp difference between each two adjacent records is calculated in sequence by traversing the sorted record sequence. The difference is compared with a preset maximum allowed sampling interval threshold. If the time difference between adjacent records is less than or equal to the threshold, it is considered that the time sequence is basically continuous and no additional processing is required. If the time difference exceeds the threshold, it is determined that there is a large time interval between the two actual observation points, which may be caused by data packet loss or temporary device offline. Interpolation is needed at the missing time point. The time of the interpolation point is determined according to the preset interpolation time granularity (such as one point per second). The specific value used for interpolation is calculated based on the values and time distance of the two adjacent actual observation points by a predetermined interpolation algorithm, such as linear interpolation or spline interpolation. The generated interpolation data point is constructed as a temporary record with a virtual timestamp and a calculated value, and is inserted into the corresponding position of the original record sequence. After the traversal and interpolation process is completed, the original initial data group is converted into aligned grouping data.

[0035] statistical features, correlation features between different fields, and frequency features of event fields of the aligned grouping data are analyzed respectively; The statistical features, correlation features and frequency features of all the grouping data are integrated according to a preset feature combination rule to generate the fusion feature set.

[0036] The embodiment divides continuous flow data into discrete and periodic analysis units by grouping protocol data sequences based on a preset time window, thereby providing a unified time reference and data range for all subsequent calculations, and realizing standardization and batch processing of the analysis process. By sorting and interval interpolating the data in the group based on the timestamp, the problem of time sequence discontinuity caused by differences in device sampling frequency or data packet loss is effectively solved, thereby generating uniform data sequences aligned on the time axis. By extracting statistical features of numerical fields, correlation features between fields, and frequency features of event fields, the running state of the device within a time window is comprehensively described from three complementary dimensions of central tendency, internal relationship, and discrete events, and multi-angle digital representation of complex states is realized.

[0037] In one embodiment, the dynamic correlation degree between the fusion feature set and the multi-source heterogeneous data is calculated, including: Extracting fusion feature data from the fusion feature set, and extracting corresponding original feature data from the multi-source heterogeneous data according to the fusion feature data; Specifically, the fusion feature set is traversed to select specific fusion feature data to be analyzed. Each selected fusion feature data carries specific information, including the device or entity identifier it describes, the corresponding original physical quantity or logical indicator, and the valid time window range. According to these specific information and according to a mapping relationship table, the original multi-source heterogeneous data pool is traced back to the initial collection and caching. The mapping relationship table records the association between each fusion feature and its original data source. For example, a fusion feature data named "average vibration intensity of main shaft motor" has a mapping relationship pointing to the original data stream with data source identifier "vibration sensor 01", and the time range is aligned. According to this mapping, all original data records in the same time window from the same data source are filtered out from the original data pool. The numerical fields in these records (such as instantaneous vibration acceleration value) are extracted to form an original observation value sequence. When it is necessary to match the fusion feature data (such as statistical mean) at the conceptual level, the same type of primary aggregation calculation corresponding to the fusion feature generation rule is performed on the original observation value sequence, thereby obtaining an original feature data value. Finally, for each processed fusion feature data, an original feature data corresponding thereto is obtained.

[0038] Performing deviation analysis on the fusion feature data and the original feature data to form a difference data sequence; Comparing each deviation value in the difference data sequence with a preset correlation degree threshold; outputting that the fusion feature data and the original feature data are in a strong correlation state if the deviation value is less than the correlation threshold, or in a weak correlation state otherwise; dynamically aggregating the strong correlation state and the weak correlation state according to the preset time window, and outputting the dynamic correlation degree.

[0039] Specifically, a time window range for the current aggregation analysis is determined. Within the aggregation window, all relevant state event records are screened out. Within the aggregation window, the proportion of the number of strong correlation state events to the total number of all state events (i.e., the sum of "strong correlation" and "weak correlation") is calculated. This proportion value, for example, 0.92 or 92%, is calculated and set as the current dynamic correlation degree index. The higher the value, the greater the proportion of the data fusion process maintaining high fidelity in the statistical time period, and the stronger the overall reliability. After the calculation is completed, the dynamic correlation degree value is bound with the aggregation window time identifier corresponding thereto to form a dynamic correlation degree.

[0040] The method provided in the embodiment ensures the strict consistency of data pairs in time, source and semantics for correlation degree evaluation by establishing an accurate mapping and backtracking extraction mechanism between the fusion feature data and the original feature data. By comparing the deviation value with the preset threshold and outputting discrete strong / weak correlation states, the continuous difference is mapped to an explicit quality classification, simplifying the judgment logic of complex data. By dynamically aggregating the discrete correlation states based on the time window, a comprehensive dynamic correlation degree index is generated, so that the reliability trend of the data fusion process in a period of time can be evaluated as a whole.

[0041] In one embodiment, the fault early warning analysis based on the dynamic correlation degree and the fusion feature set obtains fault early warning control information, including: matching each feature item in the fusion feature set with a preset fault feature library, and if the matching is successful, marking the feature item as a to-be-judged feature; extracting the dynamic correlation degree corresponding to each to-be-judged feature, and comparing the dynamic correlation degree with a preset early warning judgment threshold; if the dynamic correlation degree is lower than the preset early warning judgment threshold, excluding the to-be-judged feature from the early warning judgment process; if the dynamic correlation degree is not lower than the preset early warning judgment threshold, marking the to-be-judged feature with an early warning identifier, and adding the to-be-judged feature to an early warning feature set; Specifically, a pre-warning identifier is attached to the current feature to be determined. The identifier is a field of Boolean type. In addition to the identifier, the time stamp of the confirmation and the dynamic correlation degree value on which the confirmation is based are also recorded. After the marking is completed, the feature item is officially transferred from the to-be-determined state to the pre-warning evidence state. Then, an adding operation is performed. The feature item with the marked pre-warning identifier is added to the pre-warning feature set. The pre-warning feature set is organized as a list, a queue or a specific data structure in the memory or cache. When added, in addition to the data of the feature item itself (such as feature identifier, feature value, matched failure mode), the record of the feature item in the pre-warning feature set also contains context information such as source device and time window. The pre-warning feature set plays the role of an effective evidence pool in the entire pre-warning determination process. As the analysis proceeds, feature items that meet the high confidence condition are successively added to the set.

[0042] When the number of pre-warning identifiers in the pre-warning feature set exceeds a preset identifier threshold, a pre-warning signal is generated; The fault pre-warning management information is generated according to the pre-warning signal and the pre-warning feature set.

[0043] Specifically, the pre-warning signal is analyzed to obtain key framework information such as the device triggering the pre-warning and the time. Based on the framework information as an index, all specific feature items causing the triggering of the current pre-warning are extracted from the pre-warning feature set. These feature items record in detail which indicators are abnormal, what the abnormal values are, and what failure mode is matched. A preset control strategy template library is called. According to the main failure mode indicated in the pre-warning signal, a corresponding strategy template is matched from the template library. The basic information of the pre-warning signal, the detailed evidence list in the pre-warning feature set, and the recommended treatment measures (such as “check the lubrication system”, “suggest downtime”) obtained from the strategy template, the notification object list (such as “Team A”, “Engineer B”), the pre-warning level (such as “Level 1”, “Level 2”), etc. are combined and filled into a strategy template, and are converted to generate a fault pre-warning management information.

[0044] The method provided by the embodiment can quickly identify feature abnormalities consistent with known failure modes by matching the fusion feature set with the preset failure feature library, thereby realizing preliminary directional screening and evidence discovery of potential failures. By introducing the dynamic correlation degree as a determination threshold to perform secondary filtering on the matched features, low-confidence abnormal signals caused by poor data source quality or fusion distortion can be effectively excluded, thereby significantly improving the credibility and reliability of the pre-warning evidence. By generating a pre-warning signal only when the number of effective pre-warning features exceeds a preset threshold, false alarms caused by single and isolated abnormal fluctuations can be avoided, thereby enhancing the robustness of the pre-warning decision and ensuring that the pre-warning is triggered only with multi-dimensional evidence support.

[0045] In one embodiment, the matching each feature item in the fusion feature set with the preset fault feature library, if the matching is successful, marking the feature item as a to-be-judged feature, comprises: extracting each feature item in the fusion feature set; comparing each feature item with the fault feature items in the preset fault feature library one by one; Specifically, two-stage conditional matching is performed on each feature item based on the preset fault feature library. The first stage is identifier matching. The feature identifier of the current feature item is read. The identifier is used as a query key to search in the preset fault feature library. The target of the search is to find all the fault feature items in the library whose feature identifier field matches the identifier of the current feature item. If no fault feature item meeting the condition is found, it is determined that the current feature item has no direct correlation with all known fault modes, and the comparison process terminates. If one or more fault feature items meeting the condition are found, the second stage of value matching is entered. For each fault feature item that matches successfully in the first stage, the defined feature value anomaly criterion is read. The criterion can be expressed as a numerical threshold, a numerical interval (such as “[50, 70]”), or a discrete value set (such as “the value belongs to {2, 5, 9}”). The actual feature value of the current feature item is logically compared with the criterion. If the feature value meets the condition specified by the criterion (for example, the actual value 120 is greater than the threshold 100), it is determined that the current feature item and the fault feature item also match successfully in the value level. When the feature item not only corresponds to the parameter monitored by the item, but also its value presents the abnormal pattern defined by the item, it is considered as a successful match.

[0046] If the feature item meets the identification requirement and the feature value range requirement of any of the fault feature items, it is determined that the feature item matches successfully; Each feature item that matches successfully is marked as the to-be-judged feature.

[0047] The method provided by the embodiment can accurately identify abnormal feature items that meet known fault modes by matching the fusion features with the preset fault feature library, thereby realizing rapid directional screening and preliminary evidence locking of potential faults. By introducing a dynamic correlation degree as a confidence threshold to perform secondary filtering on the matching features, low-reliability alarms generated due to fluctuations in data source quality or fusion distortion can be effectively eliminated, thereby significantly improving the credibility of the early warning evidence and the decision accuracy. By using a multi-evidence accumulation threshold mechanism to trigger the early warning, it is ensured that an alarm is generated only when multiple independent high-confidence abnormal features appear at the same time, thereby effectively avoiding false positives caused by single, isolated noise or interference, and enhancing the robustness of the early warning system.

[0048] Refer toFigure 2 As shown, the application also provides a monitoring data fusion analysis device, applied to the monitoring data fusion analysis method of any one of the above, comprising: A collection module, the collection module is used for acquiring multi-source heterogeneous data of production equipment and sensing devices, and attaching metadata tags to the multi-source heterogeneous data, to obtain monitoring data; An analysis module, the analysis module is used for protocol analysis of the monitoring data based on a preset protocol rule, to generate a protocol data sequence; A correlation module, the correlation module is used for multi-source aggregation and space-time alignment of the protocol data sequence based on a preset time window, to generate a fusion feature set, and to calculate a dynamic correlation degree between the fusion feature set and the multi-source heterogeneous data; A processing module, the processing module is used for fault early warning analysis according to the dynamic correlation degree and the fusion feature set, to obtain fault early warning management and control information.

[0049] The monitoring data fusion analysis device provided by the application effectively integrates multi-source heterogeneous data from different protocols and time sequence references by establishing a standardized protocol analysis and data alignment mechanism, breaking the information silos in the factory monitoring system. By constructing a fusion feature set and performing dynamic correlation analysis, the internal relationship between multiple feature data can be deeply mined, deep perception of equipment health status and early warning of abnormal trends can be realized. Discrete multi-source data is converted into systematic operation and maintenance knowledge, significantly improving the accuracy and timeliness of fault prediction, and providing reliable technical support for realizing predictive maintenance and intelligent management and control.

[0050] Reference Figure 3 As shown, the application also provides a monitoring data fusion analysis system, comprising: A memory, used for storing programs; A processor, used for executing the programs, to realize each step of the monitoring data fusion analysis method of any one of the above.

[0051] In this embodiment, the processor and the memory can be connected through a bus or other means. The memory can include volatile memory, such as random access memory; the memory can also include non-volatile memory, such as read-only memory, flash memory, hard disk or solid state disk. The processor can be a general-purpose processor, such as a central processing unit, a digital signal processor, an application-specific integrated circuit, or one or more integrated circuits configured to implement embodiments of the application.

[0052] The application also provides a storage medium storing computer instructions, the computer instructions being used for making a computer execute the method of any one of the above.

[0053] It should be noted that, for the convenience and brevity of description, the specific working processes of the system and each module described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0054] The above only describes the preferred embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields based on the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. A method for fusion and analysis of monitoring data, characterized in that, include: Acquire multi-source heterogeneous data from production equipment and sensing devices, and attach metadata tags to the multi-source heterogeneous data to obtain monitoring data; The monitoring data is parsed based on preset protocol rules to generate a protocol data sequence. Based on a preset time window, the protocol data sequence is aggregated from multiple sources and spatiotemporally aligned to generate a fusion feature set, and the dynamic correlation between the fusion feature set and the multi-source heterogeneous data is calculated. Fault warning analysis is performed based on the dynamic correlation degree and the fusion feature set to obtain fault warning and control information.

2. The monitoring data fusion and analysis method according to claim 1, characterized in that, The process of acquiring multi-source heterogeneous data from production equipment and sensing devices, and attaching metadata tags to the multi-source heterogeneous data to obtain monitoring data, includes: Obtain multi-source heterogeneous data from the data interface corresponding to the production equipment and the sensing device; Based on the data interface, metadata tag templates are extracted from the pre-set tag module library, and the multi-source heterogeneous data is parsed to obtain collection time information and device identity information. The data collection time information and the device identity information are filled into the metadata tag template to form the metadata tag; The metadata tags are associated and bound with the multi-source heterogeneous data to generate the monitoring data.

3. The monitoring data fusion and analysis method according to claim 1, characterized in that, The step of parsing the monitoring data based on preset protocol rules to generate a protocol data sequence includes: The monitoring data is initially separated according to the separation rules of the preset protocol rules to obtain the data payload and protocol identifier; Extract the data field structure definition corresponding to the protocol identifier from the preset protocol rules; Based on the defined data field structure, the data payload is subjected to field location and segmentation to obtain the data of the field to be verified; Based on the format specification rules of the preset protocol rules, the validity check and format conversion of the field data to be verified are performed to obtain the converted field data; The protocol data sequence is generated by arranging and assembling all the transformation field data according to the protocol sequence rules of the preset protocol rules.

4. The monitoring data fusion and analysis method according to claim 1, characterized in that, The step of performing multi-source aggregation and spatiotemporal alignment of the protocol data sequence based on a preset time window to generate a fused feature set includes: The protocol data sequence is time-grouped according to the preset time window to obtain initial data groups; Based on the timestamp interval of the preset time window, each initial data group is sorted and interpolated to obtain aligned grouped data; The statistical characteristics of the numerical fields, the correlation characteristics between different fields, and the frequency characteristics of the event fields of the aligned grouped data are analyzed respectively. Based on preset feature combination rules, the statistical features, correlation features, and frequency features of all the grouped data are integrated to generate the fused feature set.

5. The monitoring data fusion and analysis method according to claim 1, characterized in that, The calculation of the dynamic correlation between the fused feature set and the multi-source heterogeneous data includes: Extract fused feature data from the fused feature set, and extract corresponding original feature data from the multi-source heterogeneous data based on the fused feature data; A deviation analysis is performed between the fused feature data and the original feature data to form a difference data sequence; Each deviation value in the difference data sequence is compared with a preset correlation threshold. If the deviation value is less than the correlation threshold, the output shows that the fused feature data and the original feature data are strongly correlated; otherwise, the output shows that they are weakly correlated. The strongly correlated state and the weakly correlated state are dynamically aggregated according to the preset time window, and the dynamic correlation degree is output.

6. The monitoring data fusion and analysis method according to claim 1, characterized in that, The step of performing fault early warning analysis based on the dynamic correlation degree and the fused feature set to obtain fault early warning and control information includes: Each feature item in the fused feature set is matched with a preset fault feature library. If the match is successful, the feature item is marked as a feature to be determined. Extract the dynamic correlation degree corresponding to each of the features to be determined, and compare the dynamic correlation degree with a preset early warning determination threshold; If the dynamic correlation is lower than the preset early warning judgment threshold, then the feature to be judged is excluded from the early warning judgment process; If the dynamic correlation degree is not lower than the preset warning judgment threshold, then the feature to be judged is marked with a warning identifier, and the feature to be judged is added to the warning feature set; When the number of warning identifiers in the warning feature set exceeds a preset identifier threshold, a warning signal is generated; The fault warning and control information is generated based on the warning signal and the set of warning features.

7. The monitoring data fusion and analysis method according to claim 6, characterized in that, The step of matching each feature item in the fused feature set with a preset fault feature library, and marking the feature item as a feature to be determined if a match is successful, includes: Extract each feature item from the fused feature set; Each of the aforementioned feature items is compared one by one with the fault feature entries in the preset fault feature library; If the feature item meets the identification requirements and feature value range requirements of any of the fault feature items, then the feature item is determined to be a successful match; Each of the features that is successfully matched is marked as the feature to be determined.

8. A monitoring data fusion and analysis device, characterized in that, The monitoring data fusion and analysis method applied to any one of claims 1-7 includes: The acquisition module is used to acquire multi-source heterogeneous data from production equipment and sensing devices, and to attach metadata tags to the multi-source heterogeneous data to obtain monitoring data. The analysis module is used to perform protocol parsing on the monitoring data based on preset protocol rules to generate a protocol data sequence; The association module is used to perform multi-source aggregation and spatiotemporal alignment on the protocol data sequence based on a preset time window, generate a fusion feature set, and calculate the dynamic correlation degree between the fusion feature set and the multi-source heterogeneous data. The processing module is used to perform fault early warning analysis based on the dynamic correlation degree and the fusion feature set to obtain fault early warning and control information.

9. A monitoring data fusion and analysis system, characterized in that, include: Memory, used to store programs; A processor is used to execute the program to implement the various steps of the monitoring data fusion and analysis method as described in any one of claims 1-7.

10. A storage medium, characterized in that, The computer contains computer instructions for causing the computer to perform the method according to any one of claims 1 to 7.

Images