Industrial equipment degradation monitoring method and system based on digital twinning

Abstract

The application discloses an industrial equipment degradation monitoring method and system based on digital twinning, and belongs to the technical field of industrial equipment state monitoring and intelligent operation and maintenance. The method comprises the following steps: establishing a unique equipment identifier; establishing a multi-dimensional label set according to equipment attributes; accessing real-time monitoring data, historical operation data, inspection records, repair records and the like; binding the data and the label set to form a specific data set; generating a digital twin object containing static attributes, a current state, a historical change trajectory and a risk portrait; automatically matching differentiated monitoring rules according to label combinations; adopting a double-channel model to perform state recognition and degradation level determination; outputting alarm or early warning information; and updating the twin object and the label set according to review and disposal results. The application can improve the integrity of the equipment digital object, realize differentiated adaptation of monitoring rules, and help improve the ability of degradation trend recognition and early warning.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment condition monitoring and intelligent operation and maintenance technology, specifically to a method and system for monitoring the deterioration of industrial equipment based on digital twins. Background Technology

[0002] Real-time monitoring of the operating status of industrial equipment and accurate prediction of its deterioration trends are crucial for ensuring production safety and reducing the risk of unplanned downtime. Digital twin technology, by constructing a mapping of physical equipment in virtual space, provides a feasible technical approach for tracking the status of equipment throughout its entire lifecycle.

[0003] Current industrial equipment monitoring systems can collect and display parameters such as temperature, vibration, current, voltage, pressure, flow rate, and speed online. Some systems also have over-limit alarm capabilities, historical curve query, and report output capabilities. However, several shortcomings still exist in practical industrial applications.

[0004] On the one hand, existing systems can typically only record basic ledger information such as equipment name, installation location, and site number, making it difficult to form a complete, calculable, associative, and updatable digital object around a single piece of equipment. Information such as the equipment's process section affiliation, operating conditions, environmental status, key component composition, maintenance history, and fault characteristics are often scattered across multiple business systems or manually recorded, lacking a systematic integration and association mechanism, resulting in a disconnect between monitoring data and the business context.

[0005] On the other hand, in terms of degradation monitoring and alarm strategies, most systems still rely primarily on uniform thresholds or a few fixed rules for judgment. When there are diverse equipment types and complex operating conditions, this crude judgment method is difficult to adapt to the differentiated monitoring needs of different equipment, and is prone to false alarms, missed alarms, or delayed warnings. Especially in the stage where equipment parameters have not yet reached the preset fixed thresholds, but have already shown a continuous deterioration trend, existing solutions generally lack effective identification and early warning capabilities.

[0006] In addition, existing digital twin construction methods are mostly customized for single device types, lacking a general construction framework across device types. Furthermore, the updates of the twin objects often rely on manual intervention, making it difficult to reflect changes in device status in a timely manner and failing to form a closed-loop iterative mechanism for monitoring, identification, early warning, handling, and updating. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for monitoring the deterioration of industrial equipment based on digital twins, which can improve the integrity of digital objects of equipment, realize the differentiated adaptation of monitoring rules, and improve the ability to identify deterioration trends and provide early warnings.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for monitoring the deterioration of industrial equipment based on digital twins, comprising: Step 1: Obtain the target device's unique device number, device type, system, location, key components, and basic information, and establish a unique device identifier; Step 2: Based on the target equipment's equipment type, location, operating mode, environmental conditions, key components, maintenance history, risk level, and typical failure modes, establish a multi-dimensional tag set for the target equipment; Step 3: Collect or access real-time monitoring data of the target device, and simultaneously access historical operating data, inspection records, maintenance records, alarm records and work order data related to the target device; Step 4: Bind the access data to the multi-dimensional tag set according to the device's unique identifier to form a specific data set for the target device; Step 5: Generate a digital twin object of the target device based on a specific dataset. The digital twin object includes the device's static attributes, current status, historical change trajectory, and risk profile. Step 6: Based on the tag combination of the target equipment, automatically match the corresponding monitoring rules from the preset rule library so that different judgment standards are used for different equipment, different components and different working conditions; Step 7: Based on the comparative analysis of real-time monitoring data and historical operating data, and combined with the installed rules, perform status identification, trend determination, risk assessment, and degradation level determination of the target equipment; Step 8: When the target device reaches the hard threshold, combined threshold, or meets the preset abnormal conditions, output alarm information; when the target device does not reach the hard threshold but shows a continuous deterioration trend, output warning information. Step 9: Update the digital twin object and multidimensional tag set based on the information from early warning, alarm, on-site verification, maintenance and handling and result confirmation to achieve continuous iteration and long-term tracking.

[0009] Furthermore, in step 2, a multi-dimensional tag set is established for the target equipment based on its equipment type, location, operating mode, environmental conditions, key components, maintenance history, risk level, and typical failure modes. Specifically: Define a label dimension space consisting of multiple label dimensions, each label dimension corresponding to a label value range. The label dimensions include equipment type label, process location label, operating mode label, environmental condition label, key component type label, maintenance history label, risk level label, and typical failure mode label. For continuous attributes in each label dimension, a preset segmentation threshold is used to map them to discrete label values, and the discrete label values ​​of each label dimension are combined into a label vector. The weights of each label dimension are adaptively assigned according to the device type to form a label weight vector, where the weight of each label dimension is obtained by normalizing the importance score of that dimension in the corresponding device category; Based on the statistical information of different tag dimensions appearing together as fault correlation factors in the historical equipment failure case library, the correlation strength between tag dimensions is calculated, and a tag correlation matrix is ​​established.

[0010] Furthermore, in step 5, a digital twin object of the target device is generated based on a specific dataset, specifically as follows: Basic data information and multidimensional tag sets are extracted from a specific dataset. Numerical attributes are arranged into numerical vectors. Tag vectors are one-hot encoded to obtain tag encoded vectors, which are then concatenated to form a device static attribute vector. Extract the mean feature vector, standard deviation feature vector, and maximum deviation feature vector from the most recent preset number of sampling points of the standardized real-time monitoring data matrix, and concatenate them to form the current state vector; After extracting all operational data within the backtracking period from the historical operational dataset and standardizing it, the backtracking period is divided into multiple equal-length time periods. The state feature vector of each time period is extracted and arranged in chronological order to form a historical change trajectory. By combining multi-dimensional tag sets, historical alarm records, and historical maintenance records, a risk index is calculated, which is a weighted composite of the fault susceptibility sub-index, the historical fault frequency sub-index, and the operational deterioration rate sub-index, thus forming a risk profile.

[0011] Furthermore, in step 6, based on the tag combination of the target device, the corresponding monitoring rules are automatically matched from the preset rule base, specifically as follows: A two-layer matching algorithm based on label vector weighted cosine similarity and rule condition coverage is adopted, wherein: The first layer calculates the weighted cosine similarity between the label vector of the device to be matched and the rule condition vector of each rule in the preset rule base. The weighted cosine similarity is calculated by weighting each dimension with the label weight vector. The second layer calculates the rule condition coverage, which represents the proportion of the actual tags of the device that satisfy the rule condition constraints in the non-wildcard dimension; The weighted cosine similarity and rule condition coverage are weighted and fused to obtain a comprehensive matching score. The matching threshold is dynamically adjusted according to the current risk index of the equipment. Rules with a comprehensive matching score not lower than the matching threshold are arranged in descending order of score, and a preset number of rules are selected to form an assembly rule set.

[0012] Furthermore, when calculating the coverage of the rule conditions, if all dimensions in the rule condition vector are wildcards, the value is directly taken as the preset maximum value to avoid calculation errors caused by a denominator of zero.

[0013] Furthermore, in step 7, a dual-channel degradation assessment model is used to identify the state of the target equipment, determine the trend, assess the risk, and determine the degradation level. This model includes a deviation assessment channel and a trend evolution assessment channel, wherein: The deviation assessment channel constructs a normal reference center and a normal reference covariance matrix using the historical state feature vectors during normal equipment operation. It calculates the Mahalanobis distance of the current state vector relative to the normal reference as a deviation index and maps the deviation index to a deviation level. The deviation threshold is dynamically adjusted according to the degradation tolerance coefficient in the assembly rules. The trend evolution assessment channel performs trend testing and slope estimation on the mean sequence of each measuring point over multiple time periods, calculates the trend intensity index of each measuring point, and obtains the trend intensity index by multiplying the standardized value of the test statistic and the normalized slope estimate. The overall trend evolution index of the equipment is obtained by combining the trend intensity indices of all measuring points, and the trend evolution index is mapped to the trend level. The deviation level and trend level are fused together using adaptive weights to obtain a comprehensive degradation index. The adaptive weights are dynamically adjusted based on the equipment risk index and the rule priority weight of the highest priority rule among the assembled rules. The final degradation level is determined based on the comprehensive degradation index.

[0014] Furthermore, in step 8, when the target device reaches a hard threshold, a combined threshold, or meets a preset abnormal condition, an alarm message is output, specifically: Perform dynamic hard threshold judgment, combined threshold judgment, and early warning of continuous deterioration trend, among which: In the dynamic hard threshold judgment, the upper and lower limits of the hard threshold of each measuring point are dynamically adjusted according to the alarm threshold and load compensation factor in the assembled rules. The load compensation factor is determined by the deviation between the current load rate and the rated load rate and the load sensitivity coefficient of the measuring point. In the combined threshold judgment, an alarm is triggered when the weighted sum of the deviation ratios of multiple associated measuring points reaches a preset combined trigger threshold value. In the continuous deterioration trend warning, the sliding slope sequence is calculated within the trend warning window. When the direction consistency condition, the amplitude increasing condition, and the deterioration level correlation condition are met simultaneously and the hard threshold alarm is not triggered, the warning information is output. The direction consistency condition requires that the proportion of the same sign in the sliding slope sequence is not less than a preset proportion. The amplitude increasing condition requires that the recent average value of the absolute value of the sliding slope is greater than the previous average value. When the target device reaches a hard threshold, a combined threshold, or meets a preset abnormal condition, an alarm message is output.

[0015] Furthermore, in step 9, the digital twin object and the multidimensional tag set are updated, specifically as follows: Update the digital twin object: append the new time period state feature vector collected after maintenance to the historical change trajectory and mark the trajectory segment points, recalculate the current state vector, re-count the number of unplanned maintenance and the total number of maintenance based on the maintenance record to update the historical fault frequency sub-index and recalculate the risk index; The multidimensional label set is updated as follows: the repair level label value is reassessed based on the repair depth, the number of repairs within the preset time period is recalculated to update the repair frequency label, the risk level label is remapped based on the updated risk index, the weight of the label dimension related to the component type involved in this repair is increased and normalized, and the co-occurrence information of this failure mode and related label dimensions is included in the statistical update of the label correlation matrix.

[0016] This invention also provides a digital twin-based industrial equipment degradation monitoring system, applied to the aforementioned digital twin-based industrial equipment degradation monitoring method, comprising: The device identification module is used to obtain the unique device number, device type, system to which it belongs, location, key components and basic information of the target device, and to establish a unique device identifier; The tag building module is used to create a multi-dimensional tag set for the target equipment based on its type, location, operating mode, environmental conditions, key components, maintenance history, risk level, and typical failure modes. The data access module is used to collect or access real-time monitoring data of the target equipment, and synchronously access historical operating data, inspection records, maintenance records, alarm records and work order data related to the target equipment. The data association module is used to bind the access data with a multi-dimensional tag set based on the device's unique identifier, forming a specific data set for the target device; A digital twin construction module is used to generate a digital twin object of the target device based on the specific data set. The digital twin object includes the device's static attributes, current status, historical change trajectory, and risk profile. The tag mapping analysis module is used to automatically match the corresponding monitoring rules from the preset rule base based on the tag combination of the target device, so that different judgment criteria are used for different devices, different components and different working conditions; The degradation monitoring module is used to compare and analyze real-time monitoring data with historical data, and combine the installed rules to identify the status of the target equipment, determine the trend, assess the risk, and determine the degradation level. The output module is used to output alarm information when the device reaches the hard threshold, combined threshold, or meets the preset abnormal conditions, and to output warning information when the device does not reach the hard threshold but shows a continuous deterioration trend. The twin update module is used to update the digital twin object and multidimensional tag set based on information such as early warning, alarm, on-site verification, maintenance and handling and result confirmation, so as to achieve continuous iteration and long-term tracking.

[0017] In summary, the present invention has at least one of the following beneficial technical effects: First, by constructing a multi-dimensional tag set containing multiple tag dimensions, the system can systematically represent multi-source attributes such as equipment type, process location, operation mode, environmental conditions, key components, maintenance history, risk level, and typical failure modes. This transforms each piece of equipment from a simple ledger object into a calculable, associative, and updatable digital twin object, enhancing the information integrity and computability of the equipment object.

[0018] Second, by uniformly binding real-time monitoring data, historical operation data, inspection records, maintenance records, alarm records, and work order data with a multi-dimensional tag set, a specific data set for the target equipment is formed, realizing the organic integration of monitoring data and business context, and providing a unified data foundation for subsequent status identification and degradation assessment.

[0019] Third, through the tag-driven differentiated rule assembly mechanism, the corresponding monitoring rules can be automatically matched from the preset rule library based on the tag combination of different devices, so that different devices, different components and different operating conditions adopt differentiated judgment standards, which helps to improve the accuracy of early warning and alarm, and reduce the probability of false alarms and missed alarms.

[0020] Fourth, by adopting a dual-channel degradation assessment model, the equipment status is comprehensively judged from two dimensions: deviation assessment and trend evolution assessment. The degradation level is synthesized through an adaptive fusion mechanism, which can identify the trend of equipment evolving from normal to abnormal earlier. Even if the monitoring parameters have not yet reached the hard threshold, the continuous deterioration trend can be effectively identified and warned in advance.

[0021] Fifth, through the twin update mechanism, information such as early warning, alarm, on-site verification, maintenance and handling and result confirmation are fed back to the digital twin object and multi-dimensional tag set, realizing the dynamic adjustment and continuous iteration of tag weight, risk level, correlation matrix and historical change trajectory, so that equipment profile and operation experience can be continuously accumulated and optimized, providing evolving data support for subsequent intelligent operation and maintenance decisions. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 Radar chart of multidimensional label weight distribution for centrifugal pumps; Figure 3 A schematic diagram of the sliding slope and hard threshold for vibration measurement point trend early warning. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention 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 merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0024] like Figure 1 As shown, the present invention proposes a method for monitoring the deterioration of industrial equipment based on digital twins, comprising: Step 1: Obtain the target device's unique device number, device type, system, location, key components, and basic information, and establish a unique device identifier; Step 2: Based on the target equipment's equipment type, location, operating mode, environmental conditions, key components, maintenance history, risk level, and typical failure modes, establish a multi-dimensional tag set for the target equipment; Step 3: Collect or access real-time monitoring data of the target device, and simultaneously access historical operating data, inspection records, maintenance records, alarm records and work order data related to the target device; Step 4: Bind the access data to the multi-dimensional tag set according to the device's unique identifier to form a specific data set for the target device; Step 5: Generate a digital twin object of the target device based on a specific dataset. The digital twin object includes the device's static attributes, current status, historical change trajectory, and risk profile. Step 6: Based on the tag combination of the target equipment, automatically match the corresponding monitoring rules from the preset rule library so that different judgment standards are used for different equipment, different components and different working conditions; Step 7: Based on the comparative analysis of real-time monitoring data and historical operating data, and combined with the installed rules, perform status identification, trend determination, risk assessment, and degradation level determination of the target equipment; Step 8: When the target device reaches the hard threshold, combined threshold, or meets the preset abnormal conditions, output alarm information; when the target device does not reach the hard threshold but shows a continuous deterioration trend, output warning information. Step 9: Update the digital twin object and multidimensional tag set based on the information from early warning, alarm, on-site verification, maintenance and handling and result confirmation to achieve continuous iteration and long-term tracking.

[0025] Next, the above method will be elaborated in detail with specific implementation examples and parameter settings, as follows: In step 1, the unique device number, device type, system to which it belongs, location, key components, and basic data of the target device are obtained, and a unique device identifier is established, specifically as follows: In practice, the unique identifier for each device adopts a hierarchical coding structure, consisting of four levels of coding segments joined together by connectors. These four levels are: system affiliation coding segment, device type coding segment, installation location coding segment, and device serial number coding segment. The coding structure is as follows: [System Affiliation Code Segment] - [Device Type Code Segment] - [Installation Location Code Segment] - [Device Serial Number Code Segment]; The system affiliation code segment uses three uppercase letters, determined by the standard abbreviation of the process system to which the equipment belongs. For example, PWS corresponds to the power system, LBS to the lubrication system, and CLS to the cooling system. The equipment type code segment uses four characters: the first two are the equipment category code, and the last two are the equipment subcategory code. For example, CP01 corresponds to a centrifugal pump, and RC02 to a reciprocating compressor. The installation location code segment uses six characters: the first three are the area code, and the last three are the location code. For example, A01003 corresponds to location 03 in area A. The equipment serial number code segment uses four digits, numbered sequentially starting from 0001.

[0026] Simultaneously with establishing the unique identifier for the device, key component information is extracted and component sub-identifiers are created. Assume the target device contains... The first key component, Sub-identifiers of key components Defined as a unique identifier for the device Additional component type suffix and component serial number ,Right now: ; in This represents a string concatenation operation. This is an abbreviation for component type. It is a three-digit serial number.

[0027] Basic data information, including equipment design parameters, rated operating parameters, manufacturer information, and commissioning date, is uniformly converted into a set of structured key-value pairs. Stored in JSON format and uniquely identified by the device. Related. The definition of is: ; in For the first The name key of the basic data attribute, For the corresponding attribute value, This represents the total number of basic data attribute items.

[0028] After this step, the output device's unique identifier is obtained. Component sub-identifier set and basic data key-value pair set These three elements serve as the basis for device identity anchoring in subsequent steps.

[0029] In step 2, a multi-dimensional tag set is established for the target equipment based on its equipment type, location, operating mode, environmental conditions, key components, maintenance history, risk level, and typical failure modes. Specifically: First, define the label dimensional space. Let the multidimensional label set be composed of... It consists of several label dimensions, each label dimension A value range corresponding to a label In this embodiment The tag dimensions include: device type tag Process location label Operating mode label Ambient temperature rating label Ambient humidity level label Environmental corrosion rating label Key component type labels Design life label Recent Repair Level Label Repair frequency label Risk level label Typical Failure Mode Labels Media type label Load level label and operating years label .

[0030] Each tag dimension The value of Discretize the values ​​based on the actual attributes of the equipment. For continuous attributes, use a piecewise mapping function. Map it to discrete label values: ; in For tag dimensions The corresponding original attribute value, To preset the segmentation threshold, These are the discrete label values ​​corresponding to each segment. For tag dimensions The number of segments.

[0031] Will The values ​​of each label dimension are combined to form a label vector. : ; in Indicates transpose

[0032] Furthermore, weights are assigned to each label dimension. This forms the label weight vector. Weight A weight allocation strategy based on device type is used for calculation. Let the device type label be... Corresponding device category index , This represents the total number of equipment categories. For each equipment category... Based on fault statistics and expert experience for this major category of equipment, the label dimensions are determined in advance. Importance rating Then the weight for: ; in Satisfying the normalization condition .

[0033] At the same time, establish a tag association matrix. , of which elements Representing label dimensions With tag dimension The strength of the correlation between them. The calculation is based on the tag dimension in the historical equipment failure case library. Values ​​and Label Dimensions The conditional probability of all values ​​occurring together: ; in This indicates the tag dimension in the historical failure case library. and tag dimensions The number of cases that were also recorded as factors related to the failure. Representing label dimensions The total number of cases recorded as factors related to failure. For tag dimensions and The joint entropy value in the failure case is used to penalize the impact of random co-occurrence.

[0034] This step outputs the label vector of the multidimensional label set. Label weight vector Label association matrix These three elements constitute a complete mathematical representation of the device's multidimensional tag set, which is passed to step 4 for data association and binding, and simultaneously passed to step 6 for rule matching.

[0035] In step 3, real-time monitoring data of the target device is collected or accessed, and historical operating data, inspection records, maintenance records, alarm records, and work order data related to the target device are accessed simultaneously. Specifically: Real-time monitoring data is acquired through sensors deployed at various measuring points on the target device. Assume the target device is configured with a total of... The sensor measurement point, the first Each sensor at time The raw values ​​collected are The sensor sampling period is Real-time data is continuously input in time-series format. Define a time window. Real-time monitoring data matrix within : ; in For the current moment, The length of the time window. The sampling period. , This represents the number of sampling points within the time window.

[0036] Standardize the raw data. For the first... Each measuring point, standardized value The calculation formula is: ; in and The first The historical mean and standard deviation of data from each measuring point under normal equipment operating conditions are used as a standardized reference benchmark. If Then take .

[0037] Outlier detection is performed on the standardized data. An improved Hampel detection method is used, with a sliding window width of [value missing]. For time Observations Calculate the median within the window. and median absolute deviation : ; ; when At that time, the judgment These are outliers, where The anomaly detection coefficient is taken as... Outliers are replaced by linear interpolation of the value at the previous valid time point of the measurement point.

[0038] Historical operational data, inspection records, maintenance records, alarm records, and work order data are accessed from the enterprise data platform via a data interface and aligned according to a unified timestamp format. Each type of historical data forms a corresponding structured data table, uniquely identified by the equipment. As a foreign key association. Inspection records include inspection time, inspection location, inspection results, and anomaly description; maintenance records include maintenance time, maintenance location, maintenance type, maintenance content, and replacement parts list; alarm records include alarm time, alarm type, alarm level, and handling result; work order data includes work order number, work order type, creation time, completion time, and work order status.

[0039] This step outputs a standardized real-time monitoring data matrix. Historical running dataset (Original values), standardized reference baseline parameter set Inspection record set Inspection and maintenance record set Alarm record set Work order dataset All of them are then passed to step 4.

[0040] In step 4, the access data is uniformly bound to the multi-dimensional tag set based on the device's unique identifier, forming a specific data set for the target device, specifically as follows: Unique device identifier Use the primary key to establish a data association index structure. Then, combine the outputs of step 1, step 2, and step 3 according to... Perform correlation aggregation. This is a specific data set for the target device. Defined as a 7-tuple: ; in: This is a unique identifier for the device, derived from step 1; This is a set of component sub-identifiers, derived from step 1; The basic data set of key-value pairs originates from step 1; The triplet of the multidimensional tag set originates from step 2; The standardized real-time monitoring data matrix is ​​derived from step 3; The historical dataset (raw values) is derived from step 3; As an auxiliary dataset, it consists of inspection record sets. Inspection and maintenance record set Alarm record set Work order dataset The combined components all originate from step 3; The standardized reference parameter set is derived from step 3.

[0041] To ensure the accuracy and consistency of data association, Perform integrity verification. Verification rules include: It must not be empty and its format must conform to the hierarchical encoding structure defined in step 1; Each All with As a prefix; The data entry must contain at least one valid data column from a measurement point. If the verification fails, return to step 3 and re-execute the data entry.

[0042] This step outputs a validated set of specific data. It serves as the sole data input source for generating the digital twin object in step 5.

[0043] In step 5, a digital twin object of the target device is generated based on a specific dataset. This digital twin object includes the device's static attributes, current state, historical change trajectory, and risk profile, specifically: Generation of device static attributes. From Extract and Construct device static attribute vectors .Will Numerical attributes are arranged into numerical vectors in a preset order. ,Will Label vectors in The tag encoding vector is obtained after one-hot encoding. The pieces are joined together to form: ; The current state is generated. This involves standardizing the real-time monitoring data matrix. The recent Statistical feature extraction is performed on each sampling point. The value is such that Integers covering the duration of observations over the most recent hour. Extracted features include: Mean eigenvector : ; in for The List.

[0044] Standard deviation eigenvector : ; in This indicates element-wise multiplication.

[0045] Maximum deviation eigenvector Defined as the most recent The maximum absolute value of the standardized values ​​of the sampling points: ; Current state vector It is formed by concatenating the above three feature vectors: ; Generation of historical change trajectories. From historical running datasets. Extract all operational data of the target device since its commissioning or within a preset backtracking period, and use a standardized reference baseline parameter set. Perform the same standardization process as in step 3 to obtain standardized historical data. Divide the backtracking period into... There are several equal-length time intervals, each with a duration of [duration value missing]. For the first Extract the state feature vector within each time period. Extraction method and current state vector Consistent. Historical trajectory of change. Defined as a sequence of state feature vectors arranged in chronological order: ; in For the first The end time of each time period.

[0046] Trend fitting is performed on the historical change trajectory. A locally weighted scatter smoothing method is used for the th... Average values ​​of each measuring point at each time period The sequence formed Fit a smooth curve. Smoothing value This is obtained by solving the following weighted least squares problem: ; in Trigonometric kernel function , This is a bandwidth parameter, with a value of [value missing]. .

[0047] Risk profile generation. Based on a multi-dimensional tag set, historical alarm records, and historical maintenance records, the equipment risk index is calculated. The risk index is a weighted composite of three sub-indices: Fault Susceptibility Index Based on the tag association matrix and label weight vector Calculations reflect the failure tendency determined by the inherent properties of the equipment: ; in For tag dimensions The weight, Label association matrix Mid-dimensional and The strength of the association.

[0048] Historical Failure Frequency Sub-index Inspection and maintenance record set Chinese statistics in the past Number of unplanned repairs within a time period Total number of repairs The ratio: ; in To prevent smoothing constants with a denominator of zero.

[0049] Operating degradation rate sub-index Based on historical change trajectory Calculation of the rate of change of the eigenvectors of the middle state. For the ... The mean sequence of each measurement point The slope of the linear fit is ,but: ; in For the first The reference value for the maximum allowable degradation rate of the sensor type corresponding to each measuring point is pre-set in the basic data set. middle.

[0050] Risk Index The synthesis formula is: ; in For the preset weighting coefficients, satisfy Default value , , .

[0051] After this step, a complete digital twin object will be output. ,in This is a vector of static attributes of the device. Let this be the current state vector. For the trajectory of historical changes, This is a risk index. The data is passed to step 6 for guiding rule matching, to step 7 for status identification and degradation monitoring, and to step 8 for early warning and alarm decision-making.

[0052] In step 6, based on the tag combination of the target equipment, the corresponding monitoring rules are automatically matched from the preset rule base, so that different judgment criteria are used for different equipment, different components, and different operating conditions. Specifically: Building the preset rule base. It is a structured set of rules, each rule Defined as a quadruple: ; in This is the rule condition vector, which defines the label conditions to which the rule applies; This is the rule parameter vector, which defines the set of monitoring parameters used after the rule is activated; Define the judgment logic after the rule is triggered, as a description of the rule action; This refers to the priority weight of the rule.

[0053] Rule condition vector With label vector Being in the same label space, its dimensions are the same as same. Each component The value can be one of the following three cases: taking a specific label value means that the dimension must match the value exactly; taking a wildcard means that the dimension does not participate in the condition constraint; taking a set of label values ​​means that the dimension matches any value in the set to satisfy the condition.

[0054] Rule parameter vector This includes a set of threshold parameters used for monitoring and judgment after the rule is activated. For the... Rule 1 Includes, but is not limited to: alarm upper limit threshold for each measuring point Alarm lower limit threshold Warning upper limit threshold Warning threshold The threshold for the number of consecutive outliers required for trend determination and degradation tolerance coefficient .

[0055] Tag-driven rule matching algorithm. This step employs an improved two-layer matching algorithm based on tag vector weighted cosine similarity and rule condition coverage to automatically match monitoring rules from a preset rule base. This algorithm introduces tag weight vector weighting and condition coverage penalty terms into the traditional cosine similarity algorithm, and adopts an adaptive matching threshold mechanism.

[0056] The first layer of the algorithm calculates the weighted cosine similarity of the label vectors. For the label vectors of the devices to be matched... and the first in the rule base Rule condition vector of a rule First of all The tag values ​​in the data are numerically encoded to obtain the result. Numerical vectors of the same dimension The encoding method is as follows: For exact match components, the index of the corresponding label value in the value range is divided by the normalized value of the value range size; for wildcard components, the average normalized value of the indexes of all possible label values ​​in that dimension is taken; for set match components, the average normalized value of the indexes of each label value in the set is taken. Weighted cosine similarity. Defined as: ; in For tag dimensions The weights (derived from the label weight vector in step 2) ), for The Middle Numerical encoding of dimension label values for The Middle Numerical encoding of dimensional condition values To prevent constants with a denominator of zero.

[0057] The second layer of the algorithm calculates the rule condition coverage. Condition coverage. To what extent do the actual labels of the devices satisfy the constraints of the rule condition vector? ; in This is an indicator function that takes the value when the condition is true. Otherwise take ; This indicates that the dimension is not a wildcard; This indicates that the device tag value meets the rule conditions and constraints.

[0058] Overall Matching Score Weighted fusion of the two-layer results: ; in The balance coefficient is set to a default value. .

[0059] Adaptive matching threshold According to the equipment risk index (Risk index derived from the digital twin object in step 5) Dynamically adjusted: ; in The basic matching threshold is set to a default value. ; To adjust the coefficients, the default value is set to... When the equipment risk index is high, the matching threshold is automatically lowered, allowing more rules to enter the candidate set and improving the sensitivity of monitoring.

[0060] All satisfied The rules are sorted in descending order of overall matching score, and the top score is taken. These rules constitute the assembly rule set of the equipment. , The default value is If the number of candidate rules is insufficient Then all of them will be included. For the assembly rule set... For each rule in the dataset, its rule parameter vector is extracted simultaneously. and rule action description And according to the rule priority weight Arrange them in descending order to form a rule execution sequence with clear priorities.

[0061] This step outputs the assembly rule set for the target device. and its corresponding set of rule parameter vectors The rule priority sequence is passed to step 7 for status identification and degradation monitoring.

[0062] In step 7, based on a comparative analysis of real-time monitoring data and historical operating data, and in conjunction with the established rules, the target equipment is subjected to status identification, trend determination, risk assessment, and degradation level determination. Specifically: This step employs an improved dual-channel degradation assessment model, including a deviation assessment channel and a trend evolution assessment channel. The outputs of the two channels are synthesized into the final degradation level determination result through an adaptive fusion mechanism.

[0063] Channel 1: Deviation Assessment Channel.

[0064] This channel assesses the degree of deviation of the current state from a historical normal operating baseline. This is based on the historical change trajectory in the digital twin object from step 5. Based on this, select equipment that is in its initial stage of operation and has no maintenance records. The state feature vectors for each time period are used to construct the normal operation baseline envelope. Let the set of normal state feature vectors be... Calculate the normal reference center and normal baseline covariance matrix : ; ; in The regularization coefficient is . The identity matrix is ​​used, and the regularization term is used to ensure the invertibility of the covariance matrix.

[0065] Current state vector (From step 5) Mahalanobis distance relative to the normal baseline Deviation index: ; Deviation index Mapped to deviation level The mapping function employs a piecewise nonlinear mapping: ; in These are the deviation thresholds. These thresholds are not fixed values, but rather determined based on the assembly rule set. Corresponding rule parameter vector Degradation tolerance coefficient Dynamic settings are implemented. For the current device, the parameter vector of the first matching rule is extracted from the assembly rule set, and the degradation tolerance coefficient is taken from it. To adjust the threshold: ; in For the first The basic threshold for level division is set to the default value. , , , ; This is the scaling factor, the default value is... .

[0066] Channel Two: Trend Evolution Assessment Channel.

[0067] This channel assesses the ongoing evolution of the equipment's condition, based on the historical change trajectory described in step 5. and the current state vector Trend extraction is performed on the feature sequence of each measurement point. Taking the first... Each measuring point is at Mean sequence over a period of time For the analysis object, a combined method of improved Mann-Kendall trend test and Sen slope estimation is used to determine the trend.

[0068] Calculate the Mann-Kendall statistic : ; in For symbolic functions: ; Standardized test statistics for: ; Among them, variance The calculation takes into account the modification of the data structure: ; in The number of groups with the same value in the sequence. For the first The number of identical values ​​in a group.

[0069] Sen slope estimate for: ; Trend strength index The normalized product of the standardized test statistic and the Sen slope: ; in For standard normal distribution Quantiles correspond ; For the first Reference degradation rate at each measuring point.

[0070] In summary The trend intensity at each measuring point yields the overall trend evolution index of the equipment. : ; Mapping trend evolution indicators to trend levels : ; in , , , This is the threshold for trend level division.

[0071] Dual-channel fusion and degradation level determination: Deviation level and trend level The comprehensive degradation index is obtained through adaptive fusion. : ; Among them, the fusion weight and It's not a fixed value, but rather based on the equipment risk index. (From step 5) and assembly rule set The rule priority weight of the highest priority rule Perform adaptive adjustments: ; ; And apply constraints When the calculation result exceeds the boundary, the boundary value is taken.

[0072] Deterioration level According to the comprehensive deterioration index determination: ; After this step, the degradation level is output. Overall deterioration index Deviation level Trend Level and trend intensity indicators at each measuring point All of these are transmitted to step 8 for early warning and alarm output decisions.

[0073] In step 8, when the target device reaches the hard threshold, combined threshold, or meets the preset abnormal conditions, an alarm message is output; when the target device does not reach the hard threshold but shows a continuous deterioration trend, a warning message is output, specifically: This step, based on traditional threshold judgment, introduces a dynamic threshold adjustment mechanism and a trend deterioration confirmation mechanism to improve alarm accuracy and early warning foresight.

[0074] Dynamic hard threshold determination: A hard threshold refers to a standardized monitoring value at a single measuring point exceeding a preset absolute upper or lower limit. In traditional solutions, the hard threshold is a fixed value; in this step, the hard threshold is based on the assembly rule set. The rule parameter vector in Dynamically adjust according to the current operating conditions of the equipment.

[0075] For the For each measurement point, the upper limit of its dynamic hard threshold and lower limit for: ; ; in and The highest priority rule in the assembly rule set The first rule parameter vector in the first rule parameter vector Alarm upper and lower limit thresholds for each measuring point Load compensation factor: ; in The current load rate, Rated load rate, For the first The load sensitivity coefficient of each measuring point, and its range. .

[0076] When any measuring point satisfy or When this occurs, a hard threshold alarm is triggered.

[0077] Combined threshold judgment: A combined threshold refers to an alarm triggered only when the standardized monitoring values ​​of multiple measuring points simultaneously meet a set of associated conditions. These combined conditions are derived from the assembly rule set. Description of rules and actions Extract from. Suppose a combination rule involves Each measuring point has one measuring point. Current deviation ratio for: ; Combined trigger function Defined as the weighted sum of the deviation ratios of each measuring point: ; in For the first The contribution weight of each measurement point in the combined rule is derived from the rule parameter vector. The preset value in the middle satisfies .

[0078] when When the combined threshold alarm is triggered, the following conditions are met: To combine trigger threshold values, the default value is set to... .

[0079] Warning of continued deterioration trend: When the device has not reached the hard threshold but shows a continuous deterioration trend, this step uses an improved sliding window trend confirmation mechanism to output early warning information.

[0080] Define the trend alert window to include The number of consecutive sampling points. For the number of... For each measurement point, calculate the sliding slope sequence within the trend warning window. Each of them For continuous within the window Least square fit slope at each point: ; in , , Values .

[0081] Confirmation of a continued deteriorating trend requires the simultaneous fulfillment of three conditions: Condition 1, Consistency of direction: The proportion of the same sign (both positive or both negative) in the sliding slope sequence is not less than : ; in The Sen slope estimate obtained in step 7. Default .

[0082] Condition 2, Increasing Amplitude: The absolute value sequence of the sliding slope shows an increasing trend, that is, the most recent... The mean of the absolute values ​​of the slopes is greater than the previous one. The mean of the numbers: ; in , Default .

[0083] Condition 3, Degradation level correlation: The degradation level output in step 7 Achieve Level II attention or above.

[0084] When all three conditions above are met simultaneously and no hard threshold alarm is triggered at any of the current measuring points, a warning message is output. The warning message includes: the identifier of the warning device. Warning monitoring point markers, current degradation level Trend strength index (From step 7) and the estimated time to reach the hard threshold .

[0085] Expected time to reach the hard threshold Extrapolation calculations are performed based on current trends. For monitoring points that trigger warnings... Let its current standardized value be The closest sliding slope The mean of the values ​​is The target hard threshold is (Taking positive deviation as an example), then: ; like Then take Preset maximum estimated time ,default Hour.

[0086] After this step, the output alarm or warning information is transmitted to step 9 for twin update.

[0087] In step 9, based on information from early warnings, alarms, on-site verification, repair and handling, and result confirmation, the digital twin object and multi-dimensional tag set are updated to achieve continuous iteration and long-term tracking. Specifically: Updates to digital twin objects: After step 8 outputs a warning or alarm message, the warning or alarm event record is appended to the event log of the digital twin object. Event Log Defined as a sequence of event records ordered by time, each event record ,in For event timestamps, , Identify the measurement point as the source of the event. For details of the incident.

[0088] Once the on-site verification is completed, the verification results will be available. ( This indicates a false alarm. (Indicating the information is true) Feedback is sent to the system. If And once the repair is completed, perform the following update operations: First, update the historical trajectory of change. The new time-period state feature vector collected after the equipment is put back into operation after maintenance will be appended to... At the same time, the maintenance time is marked as the trajectory segment point, indicating that the equipment status undergoes a step recovery at this point.

[0089] Second, update the current state vector. During the stable operation period after maintenance, data will be collected again. Each sampling point is recalculated according to the method described in step 5. .

[0090] Third, update the historical failure frequency sub-index in the risk profile. The repair records corresponding to this incident will be included in the statistics, and the calculation will be recalculated. and ,renew .

[0091] Fourth, update the risk index. Based on the updated and recalculated Recalculate according to the synthesis formula in step 5. .

[0092] Update of the multidimensional tag set: Once the repair is completed, update the multidimensional tag set based on the repair content and replaced parts. Specific updates include: Repair level label Update: The repair level label value has been reassessed based on the depth and scope of this repair.

[0093] Repair frequency label Update: The most recent The number of repairs within a given time period is recalculated, and the tag segmentation mapping results are updated. Default sky.

[0094] Risk level label Update: Based on the updated risk index Perform a remapping. The mapping rules are as follows: ; Label weight vector Update: If the types of parts involved in this repair are highly correlated with a specific tag dimension, then the weight of that tag dimension will be appropriately increased. Let's assume this repair involves tag dimensions... Its weight increment for: ; The updated weights have been normalized: ; in when ,otherwise .

[0095] Tag Relevance Matrix Update: Incorporate co-occurrence information of the current failure mode and related tag dimensions into the statistics, and update the corresponding positions. The value of . Let the set of tag dimensions involved in this fault be . For any and The update method is as follows: ; in The joint entropy is recalculated after adding new cases.

[0096] Updated multidimensional tag set and the updated digital twin object The feedback is then sent back to step 6, triggering a new round of rule assembly, thus forming a complete closed-loop iterative mechanism for monitoring, identification, early warning, handling, and updating.

[0097] The present invention also provides an industrial equipment degradation monitoring system based on digital twins, which includes nine modules, corresponding to steps 1 to 9 of the above method.

[0098] The device identification module is used to obtain the unique device number, device type, system to which it belongs, installation location, key components, and basic information of the target device, and to establish a unique device identifier. This module executes all the processing logic of step 1 above, including hierarchical coding structure generation, component sub-identifier establishment, and basic data key-value pair structuring, and outputs... , and .

[0099] The tag construction module is used to create a multi-dimensional tag set for the target equipment based on its type, process location, operating mode, environmental conditions, key components, maintenance history, risk level, and typical failure modes. This module executes all the processing logic in step 2 above, including tag dimension definition, segmented mapping, weight calculation, and correlation matrix construction, and outputs... .

[0100] The data access module is used to collect or access real-time monitoring data from the target device, and simultaneously access historical operating data, inspection records, maintenance records, alarm records, and work order data related to the target device. This module executes all the processing logic in step 3 above, including sensor data acquisition, standardization processing, outlier detection, and structured data access, and outputs... , , , , , and .

[0101] The data association module is used to bind access data with a multi-dimensional tag set based on the device's unique identifier, forming a specific data set for the target device. This module executes all the processing logic in step 4 above, including seven-tuple aggregation and integrity verification, and outputs... .

[0102] The digital twin construction module is used to generate a digital twin object of the target device based on the specific dataset. The digital twin object includes the device's static attributes, current state, historical change trajectory, and risk profile. This module executes all the processing logic in step 5 above, including static attribute vector construction, current state feature extraction, historical trajectory fitting, and risk index synthesis, and outputs... .

[0103] The tag mapping analysis module automatically matches corresponding monitoring rules from a preset rule base based on the tag combinations of the target device, enabling different judgment criteria to be applied to different devices, components, and operating conditions. This module executes all the processing logic in step 6 above, including preset rule base management, weighted cosine similarity calculation, condition coverage calculation, and adaptive matching threshold adjustment, and outputs... and corresponding parameters.

[0104] The degradation monitoring module compares and analyzes real-time monitoring data with historical data, and combines this with pre-installed rules to identify the status of target equipment, determine trends, assess risks, and classify degradation levels. This module executes all the processing logic in step 7 above, including deviation assessment channel calculation, trend evolution assessment channel calculation, dual-channel adaptive fusion, and degradation level determination, and outputs the results. , , , and .

[0105] The output module is used to output alarm information when the device reaches a hard threshold, a combined threshold, or meets preset abnormal conditions, and to output warning information when the device does not reach a hard threshold but shows a continuous deterioration trend. This module executes all the processing logic in step 8 above, including dynamic hard threshold judgment, combined threshold judgment, and confirmation of continuous deterioration trend warning, and outputs alarm or warning event.

[0106] The twin update module is used to update the digital twin object and multi-dimensional tag set based on early warning, alarm, on-site verification, maintenance and handling, and result confirmation information to achieve continuous iteration and long-term tracking. This module executes all the processing logic in step 9 above, including event log management, historical trajectory update, risk index recalculation, tag weight update, and correlation matrix update, and feeds the update results back to the tag mapping analysis module.

[0107] The above nine modules are sequentially connected via data flow: the output of the device identification establishment module is transmitted to the tag construction module and the data access module; the outputs of the tag construction module and the data access module are jointly transmitted to the data association module; the output of the data association module is transmitted to the digital twin construction module; the output of the digital twin construction module is transmitted to the tag mapping analysis module and the degradation monitoring module; the output of the tag mapping analysis module is transmitted to the degradation monitoring module; the output of the degradation monitoring module is transmitted to the output module; the output of the output module is transmitted to the twin update module; and the output of the twin update module is fed back to the tag mapping analysis module and the digital twin construction module, forming a closed-loop iteration.

[0108] This invention provides several embodiments to further illustrate the above-described system and method: Example 1: Construction of Multidimensional Tag Set and Digital Twin Object for Centrifugal Pump Equipment This embodiment uses a key centrifugal pump in the circulating water system of a petrochemical plant as the target device to demonstrate the process of constructing a multidimensional tag set and generating a digital twin object.

[0109] First, obtain the basic information of the equipment through step 1: the equipment number is PWS-CP01-A01003-0005, the equipment type is a centrifugal pump, the system belongs to the power system, the installation location is position 03 in area A, and the key components include the impeller, bearings, and mechanical seal. The basic data shows a rated flow rate of 200 cubic meters per hour, a rated head of 50 meters, a rated power of 45 kilowatts, a rated speed of 2950 revolutions per minute, and a commissioning date of March 2018.

[0110] Following step 2, a multidimensional label set with 15 label dimensions is established for the centrifugal pump. The values ​​of each label dimension and the weights calculated in step 2 are shown in Table 1.

[0111] Table 1. Values ​​and weights of the multidimensional label set for centrifugal pumps

[0112] Centrifugal pumps operate in high-temperature and high-humidity environments and are critical process equipment that operates continuously. Therefore, the process location label and operating mode label have high weight. Due to recent maintenance, the maintenance level and failure mode labels also play an important role in the risk profile. Figure 2 The weight distribution of the main dimensions in the multidimensional label set of the centrifugal pump is displayed intuitively in the form of a radar chart.

[0113] After obtaining the specific dataset of the centrifugal pump, a digital twin object is generated according to step five. The static attribute vector is obtained by concatenating the equipment parameters with the tag vector after thermal encoding; the current state vector is obtained by statistically analyzing the standardized data of vibration, temperature, and current from the most recent 60 sampling points (sampling period of 1 minute), and its mean feature vector is... The standard deviation eigenvector is The maximum deviation eigenvector is The risk profile calculation yields a failure susceptibility index. Historical failure frequency sub-index Operating degradation rate sub-index Synthetic risk index It is at a low to medium risk level. This twin object will be used for subsequent rule assembly and degradation monitoring.

[0114] This embodiment specifically demonstrates the quantitative construction process and weight allocation basis of the multidimensional label set, and verifies the feasibility of characterizing the device differentiation attributes based on label vectors and weight vectors in this invention.

[0115] Example 2: Application of the dual-channel degradation assessment model in centrifugal pump condition identification For the centrifugal pump in Example 1, a dual-channel degradation assessment is performed on its current state after 48 hours of continuous operation. The system has already equipped the device with five high-matching rules through step six, with the degradation tolerance coefficient being the highest priority rule parameter. .

[0116] The deviation assessment channel selects the state vectors of the first 30 normal periods during the initial operation of the equipment to construct a normal baseline. The center of the normal baseline is calculated. and covariance matrix (After regularization). Current state vector Mahalanobis distance relative to the normal reference Based on the dynamically adjusted boundary threshold Determine the level of deviation. (Notice).

[0117] The trend evolution assessment channel performed Mann-Kendall test and Sen slope estimation on the mean sequence of the three measuring points of vibration, temperature and current of the centrifugal pump for nearly 100 time periods, and obtained the trend intensity index and statistics of each measuring point as shown in Table 2.

[0118] Table 2 Trend Analysis Results for Each Measurement Point

[0119] Overall equipment trend evolution indicators Determine the trend level by comparing with the boundary threshold. (Note). At this time, the equipment risk index... The priority weight of the highest priority rule Calculate the fusion weights for the deviation channels: ; Even after applying constraints Within the range, Overall deterioration index: ; The final degradation level is determined to be Level II (Caution). Although the vibration and temperature measurement points have shown some deviation and an upward trend, they have not yet reached a serious level, and the system will output a status warning at the Caution level.

[0120] This embodiment verifies that the dual-channel degradation assessment model can integrate information on both deviation and trend under multi-source parameters, and output a reasonable degradation level by fusion through adaptive weights, thus avoiding the false alarms or missed alarms that may be caused by single-channel assessment.

[0121] Example 3: Verification of Early Warning of Continuous Deterioration Trend and Dynamic Threshold Alarm Taking the centrifugal pump from Example 1 as an example, during continuous operation, the standardized values ​​of its vibration measurement points showed a slow upward trend, but had not yet reached the upper limit of the dynamic hard threshold. The system entered the trend warning monitoring window at the 500th sampling point, with a window length of... The slip slope sequence was calculated. Table 3 lists the vibration normalization values ​​and calculated slip slopes for the most recent 15 sampling points (taking...). ).

[0122] Table 3. Vibration measuring point sliding slope sequence (partial)

[0123] Statistical analysis showed that 100% of the 30 sliding slopes had the same sign (all positive), thus satisfying the condition of directional consistency. The average absolute value of the last 5 slopes is 0.018, which is greater than the average absolute value of the previous 5 slopes (0.012), satisfying the amplitude increase condition. The current degradation level is determined to be Level II (note) by step seven, satisfying the correlation condition. Therefore, the system outputs a warning message at sampling point 500: Warning device identifier PWS-CP01-A01003-0005, warning measurement point is a vibration measurement point, and the estimated time to reach the upper limit of the hard threshold. The calculations are as follows: the current vibration normalization value is 0.72, the average of the last 5 slopes is 0.0176, and the upper limit of the dynamic hard threshold is... (After load compensation) The system issued an early warning of "increasing degradation trend" about 27 minutes in advance, which bought time for on-site response.

[0124] Figure 3 The sliding slope change of the vibration measuring point within the warning window and the hard threshold line are given, intuitively demonstrating the process of trend deterioration. This embodiment demonstrates that when the device has not yet triggered the hard threshold alarm, the sliding window trend confirmation mechanism can accurately identify the continuous deterioration trend and issue an early warning, as well as the dynamic threshold adjustment and quantitative calculation process of the estimated time to reach the threshold, verifying the beneficial effect of this invention in terms of early warning foresight.

[0125] Through the above three embodiments, the practicality, accuracy and technical advantages of the method and system provided by this invention in industrial equipment condition monitoring and deterioration early warning are systematically verified from three aspects: multi-dimensional label set construction and digital twin generation, dual-channel degradation assessment fusion, trend early warning and dynamic threshold.

[0126] Embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0127] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0128] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0129] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0130] Contents not described in detail in this specification are prior art known to those skilled in the art. It is hereby indicated that the above description is intended to help those skilled in the art understand this invention, but does not limit the scope of protection of this invention. Any equivalent substitutions, modifications, improvements, or simplifications of the above descriptions that do not depart from the essential content of this invention fall within the scope of protection of this invention.

Claims

1. A method for monitoring the deterioration of industrial equipment based on digital twins, characterized in that, include: Step 1: Obtain the target device's unique device number, device type, system, location, key components, and basic information, and establish a unique device identifier; Step 2: Based on the target equipment's equipment type, location, operating mode, environmental conditions, key components, maintenance history, risk level, and typical failure modes, establish a multi-dimensional tag set for the target equipment; Step 3: Collect or access real-time monitoring data of the target device, and simultaneously access historical operating data, inspection records, maintenance records, alarm records and work order data related to the target device; Step 4: Bind the access data to the multi-dimensional tag set according to the device's unique identifier to form a specific data set for the target device; Step 5: Generate a digital twin object of the target device based on a specific dataset. The digital twin object includes the device's static attributes, current status, historical change trajectory, and risk profile. Step 6: Based on the tag combination of the target equipment, automatically match the corresponding monitoring rules from the preset rule library so that different judgment standards are used for different equipment, different components and different working conditions; Step 7: Based on the comparative analysis of real-time monitoring data and historical operating data, and combined with the installed rules, perform status identification, trend determination, risk assessment, and degradation level determination of the target equipment; Step 8: When the target device reaches the hard threshold, combined threshold, or meets the preset abnormal conditions, output alarm information; When the target device has not reached the hard threshold but shows a continuous deterioration trend, an early warning message is output. Step 9: Update the digital twin object and multidimensional tag set based on the information from early warning, alarm, on-site verification, maintenance and handling and result confirmation to achieve continuous iteration and long-term tracking.

2. The method for monitoring the deterioration of industrial equipment based on digital twins according to claim 1, characterized in that, In step 2, a multi-dimensional tag set is established for the target equipment based on its equipment type, location, operating mode, environmental conditions, key components, maintenance history, risk level, and typical failure modes. Specifically: Define a label dimension space consisting of multiple label dimensions, each label dimension corresponding to a label value range. The label dimensions include equipment type label, process location label, operating mode label, environmental condition label, key component type label, maintenance history label, risk level label, and typical failure mode label. For continuous attributes in each label dimension, a preset segmentation threshold is used to map them to discrete label values, and the discrete label values ​​of each label dimension are combined into a label vector. The weights of each label dimension are adaptively assigned according to the device type to form a label weight vector, where the weight of each label dimension is obtained by normalizing the importance score of that dimension in the corresponding device category; Based on the statistical information of different tag dimensions appearing together as fault correlation factors in the historical equipment failure case library, the correlation strength between tag dimensions is calculated, and a tag correlation matrix is ​​established.

3. The method for monitoring the deterioration of industrial equipment based on digital twins according to claim 2, characterized in that, In step 5, a digital twin object of the target device is generated based on a specific dataset, specifically as follows: Basic data information and multidimensional tag sets are extracted from a specific dataset. Numerical attributes are arranged into numerical vectors. Tag vectors are one-hot encoded to obtain tag encoded vectors, which are then concatenated to form a device static attribute vector. Extract the mean feature vector, standard deviation feature vector, and maximum deviation feature vector from the most recent preset number of sampling points of the standardized real-time monitoring data matrix, and concatenate them to form the current state vector; After extracting all operational data within the backtracking period from the historical operational dataset and standardizing it, the backtracking period is divided into multiple equal-length time periods. The state feature vector of each time period is extracted and arranged in chronological order to form a historical change trajectory. By combining multi-dimensional tag sets, historical alarm records, and historical maintenance records, a risk index is calculated, which is a weighted composite of the fault susceptibility sub-index, the historical fault frequency sub-index, and the operational deterioration rate sub-index, thus forming a risk profile.

4. The method for monitoring the deterioration of industrial equipment based on digital twins according to claim 3, characterized in that, In step 6, based on the tag combination of the target device, the corresponding monitoring rules are automatically matched from the preset rule base, specifically as follows: A two-layer matching algorithm based on label vector weighted cosine similarity and rule condition coverage is adopted, wherein: The first layer calculates the weighted cosine similarity between the label vector of the device to be matched and the rule condition vector of each rule in the preset rule base. The weighted cosine similarity is calculated by weighting each dimension with the label weight vector. The second layer calculates the rule condition coverage, which represents the proportion of the actual tags of the device that satisfy the rule condition constraints in the non-wildcard dimension; The weighted cosine similarity and rule condition coverage are weighted and fused to obtain a comprehensive matching score. The matching threshold is dynamically adjusted according to the current risk index of the equipment. Rules with a comprehensive matching score not lower than the matching threshold are arranged in descending order of score, and a preset number of rules are selected to form an assembly rule set.

5. The method for monitoring the deterioration of industrial equipment based on digital twins according to claim 4, characterized in that, When calculating the coverage of the rule conditions, if all dimensions in the rule condition vector are wildcards, the value is directly taken as the preset maximum value to avoid calculation errors caused by a denominator of zero.

6. The method for monitoring the deterioration of industrial equipment based on digital twins according to claim 4, characterized in that, In step 7, a dual-channel degradation assessment model is used to identify the status of the target equipment, determine the trend, assess the risk, and determine the degradation level. This model includes a deviation assessment channel and a trend evolution assessment channel. The deviation assessment channel constructs a normal reference center and a normal reference covariance matrix using the historical state feature vectors during normal equipment operation. It calculates the Mahalanobis distance of the current state vector relative to the normal reference as a deviation index and maps the deviation index to a deviation level. The deviation threshold is dynamically adjusted according to the degradation tolerance coefficient in the assembly rules. The trend evolution assessment channel performs trend testing and slope estimation on the mean sequence of each measuring point over multiple time periods, calculates the trend intensity index of each measuring point, and obtains the trend intensity index by multiplying the standardized value of the test statistic and the normalized slope estimate. The overall trend evolution index of the equipment is obtained by combining the trend intensity indices of all measuring points, and the trend evolution index is mapped to the trend level. The deviation level and trend level are fused together using adaptive weights to obtain a comprehensive degradation index. The adaptive weights are dynamically adjusted based on the equipment risk index and the rule priority weight of the highest priority rule among the assembled rules. The final degradation level is determined based on the comprehensive degradation index.

7. The method for monitoring the deterioration of industrial equipment based on digital twins according to claim 6, characterized in that, In step 8, when the target device reaches a hard threshold, a combined threshold, or meets a preset abnormal condition, an alarm message is output, specifically: Perform dynamic hard threshold judgment, combined threshold judgment, and early warning of continuous deterioration trend, among which: In the dynamic hard threshold judgment, the upper and lower limits of the hard threshold of each measuring point are dynamically adjusted according to the alarm threshold and load compensation factor in the assembled rules. The load compensation factor is determined by the deviation between the current load rate and the rated load rate and the load sensitivity coefficient of the measuring point. In the combined threshold judgment, an alarm is triggered when the weighted sum of the deviation ratios of multiple associated measuring points reaches a preset combined trigger threshold value. In the continuous deterioration trend warning, the sliding slope sequence is calculated within the trend warning window. When the direction consistency condition, the amplitude increasing condition, and the deterioration level correlation condition are met simultaneously and the hard threshold alarm is not triggered, the warning information is output. The direction consistency condition requires that the proportion of the same sign in the sliding slope sequence is not less than a preset proportion. The amplitude increasing condition requires that the recent average value of the absolute value of the sliding slope is greater than the previous average value. When the target device reaches a hard threshold, a combined threshold, or meets a preset abnormal condition, an alarm message is output.

8. The method for monitoring the deterioration of industrial equipment based on digital twins according to claim 7, characterized in that, In step 9, the digital twin object and the multidimensional tag set are updated, specifically as follows: Update the digital twin object: append the new time period state feature vector collected after maintenance to the historical change trajectory and mark the trajectory segment points, recalculate the current state vector, re-count the number of unplanned maintenance and the total number of maintenance based on the maintenance record to update the historical fault frequency sub-index and recalculate the risk index; The multidimensional label set is updated as follows: the repair level label value is reassessed based on the repair depth, the number of repairs within the preset time period is recalculated to update the repair frequency label, the risk level label is remapped based on the updated risk index, the weight of the label dimension related to the component type involved in this repair is increased and normalized, and the co-occurrence information of this failure mode and related label dimensions is included in the statistical update of the label correlation matrix.

9. A digital twin-based industrial equipment degradation monitoring system, applied to the digital twin-based industrial equipment degradation monitoring method described in any one of claims 1-8, characterized in that, include: The device identification module is used to obtain the unique device number, device type, system to which it belongs, location, key components and basic information of the target device, and to establish a unique device identifier; The tag building module is used to create a multi-dimensional tag set for the target equipment based on its type, location, operating mode, environmental conditions, key components, maintenance history, risk level, and typical failure modes. The data access module is used to collect or access real-time monitoring data of the target equipment, and synchronously access historical operating data, inspection records, maintenance records, alarm records and work order data related to the target equipment. The data association module is used to bind the access data with a multi-dimensional tag set based on the device's unique identifier, forming a specific data set for the target device; A digital twin construction module is used to generate a digital twin object of the target device based on the specific data set. The digital twin object includes the device's static attributes, current status, historical change trajectory, and risk profile. The tag mapping analysis module is used to automatically match the corresponding monitoring rules from the preset rule base based on the tag combination of the target device, so that different judgment criteria are used for different devices, different components and different working conditions; The degradation monitoring module is used to compare and analyze real-time monitoring data with historical data, and combine the installed rules to identify the status of the target equipment, determine the trend, assess the risk, and determine the degradation level. The output module is used to output alarm information when the device reaches the hard threshold, combined threshold, or meets the preset abnormal conditions, and to output warning information when the device does not reach the hard threshold but shows a continuous deterioration trend. The twin update module is used to update the digital twin object and multidimensional tag set based on information such as early warning, alarm, on-site verification, maintenance and handling and result confirmation, so as to achieve continuous iteration and long-term tracking.

Images