Asset fault early warning system based on machine learning

Abstract

The invention discloses an asset fault early warning system based on machine learning, and the system comprises an obtaining module which is used for obtaining a historical fault log, and generating an important evaluation value of a corresponding asset according to the historical fault log; the setting module is used for setting a time interval of adjacent acquisition time nodes of each asset according to the important evaluation value and acquiring historical state parameters; the determination module is used for performing correlation analysis on the historical state parameters and the state evaluation indexes, determining the feature state parameters according to the analysis result, and determining an influence mapping table of the feature state parameters based on the influence relationship of the feature state parameters between different assets; the construction module is used for constructing a state prediction model and an influence prediction model; and the early warning module is used for analyzing the real-time characteristic state parameters based on the influence prediction model and the state prediction model, and judging whether to send an asset fault early warning signal, so that the prediction accuracy of asset fault early warning is improved, and the factory fault probability and the operation and maintenance cost are reduced.

Description

Technical Field

[0001] The present application relates to the field of asset monitoring technology, and in particular to an asset failure early warning system based on machine learning. Background Art

[0002] The asset monitoring system is one of the key technologies designed to ensure the safe operation of assets in the factory. As the scale of factories and the complexity of assets continue to increase, asset failure warning and timely operation and maintenance have become an important task in the asset monitoring system.

[0003] In the existing technology, different assets are monitored and analyzed through monitoring data obtained in real time by the asset monitoring system to determine whether there is any asset failure. Although it can determine to a certain extent that an asset has failed, it is impossible to predict asset failures or guarantee the accuracy of asset failure warnings, resulting in varying degrees of failures in the factory and increased factory operation and maintenance costs. Summary of the Invention

[0004] To solve the above technical problems, the present application provides an asset fault warning system based on machine learning. By setting collection time nodes at different time intervals according to the important evaluation values ​​of the assets, and screening out the characteristic state parameters of the status evaluation indicators of each asset, a state prediction model is constructed according to the characteristic state parameters and the corresponding historical state evaluation values, and an impact prediction model is constructed according to the characteristic state parameters and the corresponding impact mapping table. The real-time characteristic state parameters are analyzed based on the state prediction model and the impact prediction model to obtain the corresponding predicted comprehensive state evaluation values ​​and generate asset fault warning signals of different levels, thereby improving the prediction accuracy of asset fault warning and reducing the probability of factory failure and operation and maintenance costs.

[0005] In some embodiments of the present application, a machine learning-based asset failure early warning system is provided, comprising: An acquisition module is used to obtain the historical fault logs of each asset in the asset monitoring system and generate an important evaluation value of the corresponding asset based on the historical fault logs; The setting module is used to set the time interval between adjacent collection time nodes of each asset according to the important evaluation value, and obtain the historical status parameters of the corresponding asset in the previous monitoring cycle according to the collection time node; A determination module is used to perform correlation analysis between the historical status parameters of each asset and the corresponding status evaluation indicators, determine the characteristic status parameters based on the analysis results, and determine the influence mapping table of the characteristic status parameters based on the influence relationship between the characteristic status parameters of different assets; A construction module is used to construct a state prediction model for the corresponding asset based on the characteristic state parameters of each asset and the corresponding comprehensive state evaluation value, and to construct an impact prediction model for the corresponding asset based on the characteristic state parameters of each asset and the corresponding impact mapping table; The early warning module is used to obtain real-time characteristic status parameters and determine whether the mutation value of the real-time characteristic status parameters is greater than the standard mutation value. If it is greater, a predicted impact mapping table and a predicted comprehensive status evaluation value are generated based on the impact prediction model and the status prediction model. If it is not greater, a predicted comprehensive status evaluation value is generated based on the status prediction model and it is determined whether to issue an asset failure early warning.

[0006] In some embodiments of the present application, generating an important evaluation value of a corresponding asset based on historical fault logs includes: Obtain several historical fault logs for each asset in the asset monitoring system, analyze each historical fault log, and extract the historical fault level, historical impact, and historical operation and maintenance cost of the corresponding historical fault log; Generate an initial evaluation value for the corresponding asset based on the historical fault level and corresponding historical impact of all historical fault logs of each asset; Generate the compensation coefficient for the corresponding asset based on the historical operation and maintenance costs of all historical fault logs of each asset; The corresponding initial evaluation value is modified according to the compensation coefficient of each asset to obtain the important evaluation value of the corresponding asset.

[0007] In some embodiments of the present application, the time interval between adjacent collection time nodes of each asset is set according to the important evaluation value, including: Presetting a first preset important evaluation value interval, a second preset important evaluation value interval, a third preset important evaluation value interval and a fourth preset important evaluation value interval; When the important evaluation value of the asset is within the first preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to a fourth preset time interval; When the important evaluation value of the asset is within the second preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to a third preset time interval; When the important evaluation value of the asset is within the third preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to the second preset time interval; When the important evaluation value of an asset is within the fourth preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to the first preset time interval.

[0008] In some embodiments of the present application, a correlation analysis is performed between historical status parameters and multiple asset evaluation indicators, and characteristic status parameters of the asset evaluation indicators are determined based on the analysis results, including: Set multiple status evaluation indicators for each asset based on its historical parameters; Taking the time length of the previous monitoring cycle as the time reference line, obtain the historical state parameters of each asset in the previous monitoring cycle and the historical state evaluation value of each state evaluation indicator according to the collection time nodes of the corresponding time interval, and map them to the time reference line to obtain the state parameter-state coefficient relationship diagram of each asset; Perform correlation analysis on each historical state parameter in the state parameter-state coefficient relationship diagram of each asset and the historical state evaluation values ​​of all state evaluation indicators to obtain the dependency coefficient between the historical state parameter at each acquisition time node in the state parameter-state coefficient relationship diagram and the historical state evaluation values ​​of all state evaluation indicators; The collection time nodes whose dependency coefficients between the historical state parameters and the historical state evaluation values ​​of each state evaluation indicator are greater than the preset dependency coefficients are selected and set as the association time nodes between the historical state parameters and each state evaluation indicator; Compare the associated time nodes of the historical state parameters and each state evaluation index with all the acquisition time nodes in the time reference line to generate the ratio of the associated nodes of the historical state parameters and the corresponding state evaluation index; If the ratio of associated nodes between the historical state parameter and the corresponding state evaluation index is greater than the preset associated node ratio, it is determined that the correlation between the historical state parameter and the corresponding state evaluation index is greater than the preset correlation threshold, and the historical state parameter is set as the characteristic state parameter of the corresponding state evaluation index.

[0009] In some embodiments of the present application, determining an influence mapping table of characteristic state parameters based on the influence relationship between characteristic state parameters of different assets includes: Obtain the mutation value of the characteristic state parameter of each state evaluation indicator at each acquisition time node in the previous monitoring cycle, and compare it with the corresponding standard mutation value. If the mutation value is greater than the standard mutation value, set the corresponding acquisition time node as the mutation time node of the current characteristic state parameter; Set the preset period immediately following the mutation time node of the characteristic state parameter as the focus period, obtain the characteristic state parameters of other assets within the focus period, and generate a parameter change curve of the characteristic state parameters of other assets within the focus period; Analyze all parameter change curves to obtain change characteristics of the characteristic state parameters corresponding to the corresponding parameter change curves, wherein the change characteristics include the change trend, change degree and change rate of the corresponding characteristic state parameters; Generate a change evaluation value of the corresponding characteristic state parameter based on the degree of change and change rate of the characteristic state parameters of other assets during the period of interest; The characteristic state parameters of other assets whose change evaluation values ​​are greater than the preset change evaluation value threshold are set as the impact state parameters of the mutation value corresponding to the current characteristic state parameters, and the impact characteristics of the mutation value corresponding to the current characteristic state parameters are set according to the change characteristics of the impact state parameters; An influence mapping table of the corresponding characteristic state parameter is constructed according to the influence state parameters and influence characteristics corresponding to multiple mutation values ​​of each characteristic state parameter, wherein each mutation value, influence state parameter, and influence characteristic in the influence mapping table of each characteristic state parameter is mapped one by one.

[0010] In some embodiments of the present application, before constructing a state prediction model for a corresponding asset based on the characteristic state parameters of each asset and the corresponding historical state evaluation value, the following steps are included: Generate the actual state evaluation value of the corresponding state evaluation indicator at each collection time node in the previous monitoring period based on the characteristic state parameters of the same state evaluation indicator of each asset at each collection time node in the previous monitoring period; The calculation formula of the actual state evaluation value is: ; Where Z is the actual state evaluation value, z0 is the state evaluation conversion coefficient, n is the total number of characteristic state parameters of the state evaluation indicator at the corresponding acquisition time node in the previous monitoring period, t0i is the i-th characteristic state parameter, Ti is the standard state parameter corresponding to the i-th characteristic state parameter, and ai is the weight coefficient of the i-th characteristic state parameter; Compare the actual state evaluation value of each acquisition time node in the previous monitoring cycle with the corresponding historical state evaluation value to obtain the state evaluation value difference, and generate the credibility of the historical state evaluation value of each acquisition time node in the previous monitoring cycle based on the state evaluation value difference; If the credibility is greater than the preset credibility threshold, the historical status evaluation value will not be corrected; If the credibility is less than the preset credibility threshold, the state evaluation value mean is generated based on the historical state evaluation value of the corresponding collection time node in the previous monitoring cycle and the actual state evaluation value, and the state evaluation value difference replaces the historical state evaluation value at the corresponding collection time node.

[0011] In some embodiments of the present application, a state prediction model for the corresponding asset is constructed based on the characteristic state parameters of each asset and the corresponding comprehensive state evaluation value, and an impact prediction model for the corresponding asset is constructed based on the characteristic state parameters of each asset and the corresponding impact mapping table, including: All characteristic state parameters of each asset at the same acquisition time node in the previous monitoring cycle are used as the first training input data set, and the comprehensive state evaluation value of the next adjacent preset time period at the same acquisition time node is used as the corresponding first training output data set; Performing neural network training based on the first training input data set and the first training output data set to obtain a state prediction model for the corresponding asset; The mutation value of each asset's characteristic state parameter at the mutation time node of the previous monitoring cycle is used as the second training input dataset, and the influencing state parameters mapped by the mutation value at the same mutation time node and the influencing features corresponding to the influencing state parameters are used as the second training output dataset; A neural network is trained based on the second training input data set and the second training output data set to obtain an impact prediction model for the corresponding asset.

[0012] In some embodiments of the present application, a real-time feature state parameter is obtained and it is determined whether the mutation value of the real-time feature state parameter is greater than the standard mutation value. If so, a predicted impact mapping table and a predicted comprehensive state evaluation value are generated based on the impact prediction model and the state prediction model. If not, a predicted comprehensive state evaluation value is generated based on the state prediction model, including: Obtain the real-time characteristic state parameters of each asset at the same collection time node within the current monitoring cycle, and calculate the mutation value of each real-time characteristic state parameter; If the mutation values ​​of all real-time characteristic state parameters of the same asset are less than the standard mutation value, the real-time characteristic state parameters of the same acquisition time node are input into the state prediction model of the corresponding asset to obtain the predicted comprehensive state evaluation value of the next adjacent future preset time period of the current acquisition time node. Based on the predicted comprehensive state evaluation value of the future preset time period, it is determined whether to issue an asset fault warning; If the mutation values ​​of all real-time characteristic status parameters of the same asset are not all less than the standard mutation value, the mutation values ​​of the real-time characteristic status parameters that are greater than the standard mutation value are input into the impact prediction model of the corresponding asset to obtain the predicted impact status parameters of the next adjacent future preset time period of the current acquisition time node and the predicted impact characteristics of the predicted impact status parameters. According to the predicted impact characteristics of the predicted impact status parameters of the future preset time period and combined with the predicted comprehensive status evaluation value, it is determined whether to issue an asset fault warning.

[0013] In some embodiments of the present application, determining whether to perform an asset failure warning includes: When the mutation values ​​of all real-time characteristic state parameters of the same asset are less than the standard mutation value, the predicted comprehensive state evaluation value of the corresponding asset in the future preset period is obtained, and a first decline rate of the predicted state evaluation value is generated; If the mutation values ​​of all real-time characteristic state parameters of the same asset are not all less than the standard mutation value, the predicted failure probability of the predicted influencing state parameter in the future preset period is generated based on the predicted influencing characteristics of the predicted influencing state parameter in the future preset period; Generating a plurality of correction coefficients based on the predicted failure probabilities of all predicted influencing state parameters in a future preset period, and correcting the corresponding predicted state evaluation values ​​in the future preset period to obtain a corrected predicted comprehensive state evaluation value in the future preset period, and generating a second rate of decrease of the corrected predicted comprehensive state evaluation value; If the predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset comprehensive status evaluation value threshold and the first decline rate is less than the preset first decline rate threshold, or if the revised predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset comprehensive status evaluation value threshold and the second decline rate is less than the preset second decline rate threshold, no asset failure warning is issued; If the predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset status evaluation value threshold and the first decline rate is greater than the first preset decline rate threshold, or if the revised predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset comprehensive status evaluation value threshold and the second decline rate is greater than the preset second decline rate threshold, a first-level asset failure warning is issued; If the predicted comprehensive status evaluation value of the future preset period is less than the preset comprehensive status evaluation value threshold, or the revised predicted comprehensive status evaluation value of the future preset period is less than the preset comprehensive status evaluation value threshold, a secondary asset failure warning is issued.

[0014] Compared with the prior art, the asset failure early warning system based on machine learning in the embodiment of the present application has the following advantages: By setting collection time nodes at different time intervals according to the important evaluation values ​​of the assets, and screening the characteristic state parameters of the status evaluation indicators of each asset, a state prediction model is constructed according to the characteristic state parameters and the corresponding historical state evaluation values, and an impact prediction model is constructed according to the characteristic state parameters and the corresponding impact mapping table. Based on the state prediction model and the impact prediction model, the real-time characteristic state parameters are analyzed to obtain the corresponding predicted comprehensive state evaluation values ​​and generate asset fault warning signals of different levels, thereby improving the prediction accuracy of asset fault warning and reducing the probability of factory failure and operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a schematic diagram of an asset failure early warning system based on machine learning in a preferred embodiment of the present application. DETAILED DESCRIPTION

[0016] The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0017] In the description of this application, it should be understood that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on this application.

[0018] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature specified as "first" or "second" may explicitly or implicitly include one or more of such features. Throughout this application, unless otherwise specified, "plurality" means two or more.

[0019] In the description of this application, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in this application based on the specific circumstances.

[0020] like Figure 1 As shown, a machine learning-based asset failure early warning system according to a preferred embodiment of the present application includes: An acquisition module is used to obtain the historical fault logs of each asset in the asset monitoring system and generate an important evaluation value of the corresponding asset based on the historical fault logs; The setting module is used to set the time interval between adjacent collection time nodes of each asset according to the important evaluation value, and obtain the historical status parameters of the corresponding asset in the previous monitoring cycle according to the collection time node; A determination module is used to perform correlation analysis between the historical status parameters of each asset and the corresponding status evaluation indicators, determine the characteristic status parameters based on the analysis results, and determine the influence mapping table of the characteristic status parameters based on the influence relationship between the characteristic status parameters of different assets; A construction module is used to construct a state prediction model for the corresponding asset based on the characteristic state parameters of each asset and the corresponding comprehensive state evaluation value, and to construct an impact prediction model for the corresponding asset based on the characteristic state parameters of each asset and the corresponding impact mapping table; The early warning module is used to obtain real-time characteristic status parameters and determine whether the mutation value of the real-time characteristic status parameters is greater than the standard mutation value. If it is greater, a predicted impact mapping table and a predicted comprehensive status evaluation value are generated based on the impact prediction model and the status prediction model. If it is not greater, a predicted comprehensive status evaluation value is generated based on the status prediction model and it is determined whether to issue an asset failure early warning.

[0021] In this embodiment, the assets include rotors, blades, detection pumps, gears, bearings, etc. The asset monitoring system refers to amsasset monitor, which performs online monitoring and analysis of multiple assets in the factory to achieve fault warning for each asset.

[0022] In some embodiments of the present application, generating an important evaluation value of a corresponding asset based on historical fault logs includes: Obtain several historical fault logs for each asset in the asset monitoring system, analyze each historical fault log, and extract the historical fault level, historical impact, and historical operation and maintenance cost of the corresponding historical fault log; Generate an initial evaluation value for the corresponding asset based on the historical fault level and corresponding historical impact of all historical fault logs of each asset; Generate the compensation coefficient for the corresponding asset based on the historical operation and maintenance costs of all historical fault logs of each asset; The corresponding initial evaluation value is modified according to the compensation coefficient of each asset to obtain the important evaluation value of the corresponding asset.

[0023] In some embodiments of the present application, the time interval between adjacent collection time nodes of each asset is set according to the important evaluation value, including: Presetting a first preset important evaluation value interval, a second preset important evaluation value interval, a third preset important evaluation value interval and a fourth preset important evaluation value interval; When the important evaluation value of the asset is within the first preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to a fourth preset time interval; When the important evaluation value of the asset is within the second preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to a third preset time interval; When the important evaluation value of the asset is within the third preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to the second preset time interval; When the important evaluation value of an asset is within the fourth preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to the first preset time interval.

[0024] In this embodiment, the first preset important evaluation value interval < the second preset important evaluation value interval < the third preset important evaluation value interval < the fourth preset important evaluation value interval, and the first preset time interval < the second preset time interval < the third preset time interval < the fourth preset time interval.

[0025] In this embodiment, the larger the preset important evaluation value interval of the important evaluation value is, the more important the corresponding asset is to the factory and operation and maintenance. Therefore, a shorter time interval should be set to achieve continuous monitoring of the corresponding assets to avoid failure to detect asset failures in time, which may cause serious losses to the factory.

[0026] In some embodiments of the present application, a correlation analysis is performed between historical status parameters and multiple asset evaluation indicators, and characteristic status parameters of the asset evaluation indicators are determined based on the analysis results, including: Set multiple status evaluation indicators for each asset based on its historical parameters; Taking the time length of the previous monitoring cycle as the time reference line, obtain the historical state parameters of each asset in the previous monitoring cycle and the historical state evaluation value of each state evaluation indicator according to the collection time nodes of the corresponding time interval, and map them to the time reference line to obtain the state parameter-state coefficient relationship diagram of each asset; Perform correlation analysis on each historical state parameter in the state parameter-state coefficient relationship diagram of each asset and the historical state evaluation values ​​of all state evaluation indicators to obtain the dependency coefficient between the historical state parameter at each acquisition time node in the state parameter-state coefficient relationship diagram and the historical state evaluation values ​​of all state evaluation indicators; The collection time nodes whose dependency coefficients between the historical state parameters and the historical state evaluation values ​​of each state evaluation indicator are greater than the preset dependency coefficients are selected and set as the association time nodes between the historical state parameters and each state evaluation indicator; Compare the associated time nodes of the historical state parameters and each state evaluation index with all the acquisition time nodes in the time reference line to generate the ratio of the associated nodes of the historical state parameters and the corresponding state evaluation index; If the ratio of associated nodes between the historical state parameter and the corresponding state evaluation index is greater than the preset associated node ratio, it is determined that the correlation between the historical state parameter and the corresponding state evaluation index is greater than the preset correlation threshold, and the historical state parameter is set as the characteristic state parameter of the corresponding state evaluation index.

[0027] In this embodiment, the status evaluation index is set according to the specific fault status in the historical fault log of each asset, for example, the imbalance state and misalignment state of the rotor, the tooth wear state, gear misalignment state, tooth fracture state, etc. The historical status parameter refers to the parameter representing the operating status of the corresponding asset.

[0028] In this embodiment, the historical status evaluation value refers to the evaluation value of the corresponding status evaluation indicator. Each historical status evaluation value is generated based on human judgment combined with the judgment results of the asset monitoring system. The larger the historical status evaluation value, the better the status of the corresponding status evaluation indicator. The dependence coefficient refers to the degree to which the historical status evaluation value of each status evaluation indicator changes with the change of the current historical status parameter. The greater the degree of change, the greater the corresponding dependence coefficient, and vice versa. The proportion of associated nodes = associated time nodes / total collection time nodes in the time reference line. The preset associated node ratio is 2 / 3.

[0029] In this embodiment, by constructing a state parameter-state coefficient relationship diagram in the previous monitoring period according to the collection time node of each asset, the synchronous change relationship between the historical state parameters and the historical state evaluation values ​​is obtained, so as to analyze and obtain the dependency coefficient and filter out the associated time nodes, and calculate the proportion of associated nodes, so as to judge whether the correlation between the historical state parameters and the corresponding state evaluation indicators is greater than the preset correlation threshold. If so, the historical state parameters are determined to be the characteristic state parameters of the state evaluation indicators, thereby improving the accuracy of the characteristic state parameters of each state evaluation indicator, laying the foundation for subsequent judgment of whether the asset is faulty, and ensuring the accuracy of asset fault warning.

[0030] In some embodiments of the present application, determining an influence mapping table of characteristic state parameters based on the influence relationship between characteristic state parameters of different assets includes: Obtain the mutation value of the characteristic state parameter of each state evaluation indicator at each acquisition time node in the previous monitoring cycle, and compare it with the corresponding standard mutation value. If the mutation value is greater than the standard mutation value, set the corresponding acquisition time node as the mutation time node of the current characteristic state parameter; Set the preset period immediately following the mutation time node of the characteristic state parameter as the focus period, obtain the characteristic state parameters of other assets within the focus period, and generate a parameter change curve of the characteristic state parameters of other assets within the focus period; Analyze all parameter change curves to obtain change characteristics of the characteristic state parameters corresponding to the corresponding parameter change curves, wherein the change characteristics include the change trend, change degree and change rate of the corresponding characteristic state parameters; Generate a change evaluation value of the corresponding characteristic state parameter based on the degree of change and change rate of the characteristic state parameters of other assets during the period of interest; The characteristic state parameters of other assets whose change evaluation values ​​are greater than the preset change evaluation value threshold are set as the impact state parameters of the mutation value corresponding to the current characteristic state parameters, and the impact characteristics of the mutation value corresponding to the current characteristic state parameters are set according to the change characteristics of the impact state parameters; An influence mapping table of the corresponding characteristic state parameter is constructed according to the influence state parameters and influence characteristics corresponding to multiple mutation values ​​of each characteristic state parameter, wherein each mutation value, influence state parameter, and influence characteristic in the influence mapping table of each characteristic state parameter is mapped one by one.

[0031] In this embodiment, the change trend includes a normal trend and an abnormal trend. When the characteristic state parameter changes toward the corresponding standard state parameter, it is a normal trend, and vice versa. The degree of change refers to the fluctuation of the characteristic state parameter in a preset time period. The greater the fluctuation, the greater the degree of change. The rate of change refers to the speed of change of the characteristic state parameter in a preset time period. The greater the degree of change or the greater the rate of change, the greater the change evaluation value.

[0032] In this embodiment, by analyzing the mutation values ​​of the characteristic state parameters, the influencing state parameters and the influencing characteristics of the mutation values ​​of each characteristic state parameter in the focus period are determined, and the influence relationship of the characteristic state parameters of each asset on the characteristic state parameters of other assets is accurately obtained, thereby improving the accuracy of asset fault warning.

[0033] In some embodiments of the present application, before constructing a state prediction model for a corresponding asset based on the characteristic state parameters of each asset and the corresponding historical state evaluation value, the following steps are included: Generate the actual state evaluation value of the corresponding state evaluation indicator at each collection time node in the previous monitoring period based on the characteristic state parameters of the same state evaluation indicator of each asset at each collection time node in the previous monitoring period; The calculation formula of the actual state evaluation value is: ; Where Z is the actual state evaluation value, z0 is the state evaluation conversion coefficient, n is the total number of characteristic state parameters of the state evaluation indicator at the corresponding acquisition time node in the previous monitoring period, t0i is the i-th characteristic state parameter, Ti is the standard state parameter corresponding to the i-th characteristic state parameter, and ai is the weight coefficient of the i-th characteristic state parameter; Compare the actual state evaluation value of each acquisition time node in the previous monitoring cycle with the corresponding historical state evaluation value to obtain the state evaluation value difference, and generate the credibility of the historical state evaluation value of each acquisition time node in the previous monitoring cycle based on the state evaluation value difference; If the credibility is greater than the preset credibility threshold, the historical status evaluation value will not be corrected; If the credibility is less than the preset credibility threshold, the state evaluation value mean is generated based on the historical state evaluation value of the corresponding collection time node in the previous monitoring cycle and the actual state evaluation value, and the state evaluation value difference replaces the historical state evaluation value at the corresponding collection time node.

[0034] In this embodiment, the operating status of each characteristic state parameter is determined based on the difference between the characteristic state parameter and the standard state parameter of each state evaluation indicator of each asset at the same collection time node in the previous monitoring cycle, and the actual state evaluation value is generated based on the operating status of all characteristic state parameters of the same state evaluation indicator, and the state evaluation value difference with the historical state evaluation value is obtained. When the state evaluation value difference is smaller, the accuracy of the historical state evaluation value is higher, that is, the credibility of the historical state evaluation value is higher, and vice versa, and the evaluation mode or method of the historical state evaluation value in the corresponding asset monitoring system is updated.

[0035] In this embodiment, the accuracy of the historical status evaluation values ​​is improved by judging the credibility of the historical status evaluation values, thereby laying a foundation for the subsequent construction of a status prediction model for each asset.

[0036] In some embodiments of the present application, a state prediction model for the corresponding asset is constructed based on the characteristic state parameters of each asset and the corresponding comprehensive state evaluation value, and an impact prediction model for the corresponding asset is constructed based on the characteristic state parameters of each asset and the corresponding impact mapping table, including: All characteristic state parameters of each asset at the same acquisition time node in the previous monitoring cycle are used as the first training input data set, and the comprehensive state evaluation value of the next adjacent preset time period at the same acquisition time node is used as the corresponding first training output data set; Performing neural network training based on the first training input data set and the first training output data set to obtain a state prediction model for the corresponding asset; The mutation value of each asset's characteristic state parameter at the mutation time node of the previous monitoring cycle is used as the second training input dataset, and the influencing state parameters mapped by the mutation value at the same mutation time node and the influencing features corresponding to the influencing state parameters are used as the second training output dataset; A neural network is trained based on the second training input data set and the second training output data set to obtain an impact prediction model for the corresponding asset.

[0037] In this embodiment, a comprehensive status evaluation value is generated according to the historical status evaluation value of the status evaluation indicator of each asset and the corresponding weight coefficient.

[0038] In some embodiments of the present application, a real-time feature state parameter is obtained and it is determined whether the mutation value of the real-time feature state parameter is greater than the standard mutation value. If so, a predicted impact mapping table and a predicted comprehensive state evaluation value are generated based on the impact prediction model and the state prediction model. If not, a predicted comprehensive state evaluation value is generated based on the state prediction model, including: Obtain the real-time characteristic state parameters of each asset at the same collection time node within the current monitoring cycle, and calculate the mutation value of each real-time characteristic state parameter; If the mutation values ​​of all real-time characteristic state parameters of the same asset are less than the standard mutation value, the real-time characteristic state parameters of the same acquisition time node are input into the state prediction model of the corresponding asset to obtain the predicted comprehensive state evaluation value of the next adjacent future preset time period of the current acquisition time node. Based on the predicted comprehensive state evaluation value of the future preset time period, it is determined whether to issue an asset fault warning; If the mutation values ​​of all real-time characteristic status parameters of the same asset are not all less than the standard mutation value, the mutation values ​​of the real-time characteristic status parameters that are greater than the standard mutation value are input into the impact prediction model of the corresponding asset to obtain the predicted impact status parameters of the next adjacent future preset time period of the current acquisition time node and the predicted impact characteristics of the predicted impact status parameters. According to the predicted impact characteristics of the predicted impact status parameters of the future preset time period and combined with the predicted comprehensive status evaluation value, it is determined whether to issue an asset fault warning.

[0039] In this embodiment, the change trend, change value and change rate of the corresponding real-time characteristic status parameters of other assets in a future preset time period are predicted based on the predicted impact characteristics of the predicted impact status parameters, so as to predict whether the corresponding real-time characteristic status parameters of other assets will fail, and obtain the predicted failure probability. According to the influence relationship of the sudden change of the real-time characteristic status parameters of the asset on the real-time characteristic status parameters of other assets, the failure probability is accurately predicted, thereby improving the accuracy of asset failure warning.

[0040] In some embodiments of the present application, determining whether to perform an asset failure warning includes: When the mutation values ​​of all real-time characteristic state parameters of the same asset are less than the standard mutation value, the predicted comprehensive state evaluation value of the corresponding asset in the future preset period is obtained, and a first decline rate of the predicted state evaluation value is generated; If the mutation values ​​of all real-time characteristic state parameters of the same asset are not all less than the standard mutation value, the predicted failure probability of the predicted influencing state parameter in the future preset period is generated based on the predicted influencing characteristics of the predicted influencing state parameter in the future preset period; Generating a plurality of correction coefficients based on the predicted failure probabilities of all predicted influencing state parameters in a future preset period, and correcting the corresponding predicted state evaluation values ​​in the future preset period to obtain a corrected predicted comprehensive state evaluation value in the future preset period, and generating a second rate of decrease of the corrected predicted comprehensive state evaluation value; If the predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset comprehensive status evaluation value threshold and the first decline rate is less than the preset first decline rate threshold, or if the revised predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset comprehensive status evaluation value threshold and the second decline rate is less than the preset second decline rate threshold, no asset failure warning is issued; If the predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset status evaluation value threshold and the first decline rate is greater than the first preset decline rate threshold, or if the revised predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset comprehensive status evaluation value threshold and the second decline rate is greater than the preset second decline rate threshold, a first-level asset failure warning is issued; If the predicted comprehensive status evaluation value of the future preset period is less than the preset comprehensive status evaluation value threshold, or the revised predicted comprehensive status evaluation value of the future preset period is less than the preset comprehensive status evaluation value threshold, a secondary asset failure warning is issued.

[0041] In this embodiment, the preset comprehensive status evaluation value threshold is the minimum status evaluation value to ensure the normal operation of each asset. The preset first decline rate threshold and the preset second decline rate threshold are set in advance based on the historical warning log. First, it is determined whether there is a sudden change in the real-time characteristic status parameter. If so, the predicted failure probability of other assets is obtained based on the impact of the sudden change value of the real-time characteristic status parameter on the real-time characteristic status parameters of other assets, and a correction coefficient is generated. If the predicted failure probability is greater, the correction coefficient is smaller, and vice versa. The value range of the correction coefficient is (0,1).

[0042] In this embodiment, asset failure warning is performed on the corresponding assets through the predicted comprehensive status evaluation value or the revised predicted comprehensive status evaluation value in the future preset time period, thereby improving the prediction accuracy of the asset failure warning and reducing the failure probability and operation and maintenance costs of the factory.

[0043] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and replacements can be made without departing from the technical principles of the present application. These improvements and replacements should also be regarded as the scope of protection of the present application.

Claims

1. An asset failure early warning system based on machine learning, characterized in that: include: An acquisition module is used to obtain the historical fault logs of each asset in the asset monitoring system and generate an important evaluation value of the corresponding asset based on the historical fault logs; The setting module is used to set the time interval between adjacent collection time nodes of each asset according to the important evaluation value, and obtain the historical status parameters of the corresponding asset in the previous monitoring cycle according to the collection time node; A determination module is used to perform correlation analysis between the historical status parameters of each asset and the corresponding status evaluation indicators, determine the characteristic status parameters based on the analysis results, and determine the influence mapping table of the characteristic status parameters based on the influence relationship between the characteristic status parameters of different assets; A construction module is used to construct a state prediction model for the corresponding asset based on the characteristic state parameters of each asset and the corresponding comprehensive state evaluation value, and to construct an impact prediction model for the corresponding asset based on the characteristic state parameters of each asset and the corresponding impact mapping table; The early warning module is used to obtain real-time characteristic status parameters and determine whether the mutation value of the real-time characteristic status parameters is greater than the standard mutation value. If it is greater, a predicted impact mapping table and a predicted comprehensive status evaluation value are generated based on the impact prediction model and the status prediction model. If it is not greater, a predicted comprehensive status evaluation value is generated based on the status prediction model and it is determined whether to issue an asset failure early warning.

2. The asset failure early warning system based on machine learning according to claim 1, characterized in that: Generate important evaluation values ​​for corresponding assets based on historical fault logs, including: Obtain several historical fault logs for each asset in the asset monitoring system, analyze each historical fault log, and extract the historical fault level, historical impact, and historical operation and maintenance cost of the corresponding historical fault log; Generate an initial evaluation value for the corresponding asset based on the historical fault level and corresponding historical impact of all historical fault logs of each asset; Generate the compensation coefficient for the corresponding asset based on the historical operation and maintenance costs of all historical fault logs of each asset; The corresponding initial evaluation value is modified according to the compensation coefficient of each asset to obtain the important evaluation value of the corresponding asset.

3. The asset failure early warning system based on machine learning according to claim 2, characterized in that: The time intervals between adjacent collection time nodes for each asset are set based on important evaluation values, including: Presetting a first preset important evaluation value interval, a second preset important evaluation value interval, a third preset important evaluation value interval and a fourth preset important evaluation value interval; When the important evaluation value of the asset is within the first preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to a fourth preset time interval; When the important evaluation value of the asset is within the second preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to a third preset time interval; When the important evaluation value of the asset is within the third preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to the second preset time interval; When the important evaluation value of an asset is within the fourth preset important evaluation value interval, the time interval between adjacent collection time nodes of the corresponding asset is set to the first preset time interval.

4. The asset failure early warning system based on machine learning according to claim 3, characterized in that: Conduct correlation analysis between historical status parameters and multiple asset evaluation indicators, and determine characteristic status parameters of asset evaluation indicators based on the analysis results, including: Set multiple status evaluation indicators for each asset based on its historical parameters; Taking the time length of the previous monitoring cycle as the time reference line, obtain the historical state parameters of each asset in the previous monitoring cycle and the historical state evaluation value of each state evaluation indicator according to the collection time nodes of the corresponding time interval, and map them to the time reference line to obtain the state parameter-state coefficient relationship diagram of each asset; Perform correlation analysis on each historical state parameter in the state parameter-state coefficient relationship diagram of each asset and the historical state evaluation values ​​of all state evaluation indicators to obtain the dependency coefficient between the historical state parameter at each acquisition time node in the state parameter-state coefficient relationship diagram and the historical state evaluation values ​​of all state evaluation indicators; The collection time nodes whose dependency coefficients between the historical state parameters and the historical state evaluation values ​​of each state evaluation indicator are greater than the preset dependency coefficients are selected and set as the association time nodes between the historical state parameters and each state evaluation indicator; Compare the associated time nodes of the historical state parameters and each state evaluation index with all the acquisition time nodes in the time reference line to generate the ratio of the associated nodes of the historical state parameters and the corresponding state evaluation index; If the ratio of associated nodes between the historical state parameter and the corresponding state evaluation index is greater than the preset associated node ratio, it is determined that the correlation between the historical state parameter and the corresponding state evaluation index is greater than the preset correlation threshold, and the historical state parameter is set as the characteristic state parameter of the corresponding state evaluation index.

5. The asset failure early warning system based on machine learning according to claim 4, characterized in that: Based on the influence relationship between the characteristic state parameters of different assets, an influence mapping table of the characteristic state parameters is determined, including: Obtain the mutation value of the characteristic state parameter of each state evaluation indicator at each acquisition time node in the previous monitoring cycle, and compare it with the corresponding standard mutation value. If the mutation value is greater than the standard mutation value, set the corresponding acquisition time node as the mutation time node of the current characteristic state parameter; Set the preset period immediately following the mutation time node of the characteristic state parameter as the focus period, obtain the characteristic state parameters of other assets within the focus period, and generate a parameter change curve of the characteristic state parameters of other assets within the focus period; Analyze all parameter change curves to obtain change characteristics of the characteristic state parameters corresponding to the corresponding parameter change curves, wherein the change characteristics include the change trend, change degree and change rate of the corresponding characteristic state parameters; Generate a change evaluation value of the corresponding characteristic state parameter based on the degree of change and change rate of the characteristic state parameters of other assets during the period of interest; The characteristic state parameters of other assets whose change evaluation values ​​are greater than the preset change evaluation value threshold are set as the impact state parameters of the mutation value corresponding to the current characteristic state parameters, and the impact characteristics of the mutation value corresponding to the current characteristic state parameters are set according to the change characteristics of the impact state parameters; An influence mapping table of the corresponding characteristic state parameter is constructed according to the influence state parameters and influence characteristics corresponding to multiple mutation values ​​of each characteristic state parameter, wherein each mutation value, influence state parameter, and influence characteristic in the influence mapping table of each characteristic state parameter is mapped one by one.

6. The asset failure early warning system based on machine learning according to claim 5, characterized in that: Before building a state prediction model for the corresponding asset based on the characteristic state parameters of each asset and the corresponding historical state evaluation values, the following steps are included: Generate the actual state evaluation value of the corresponding state evaluation indicator at each collection time node in the previous monitoring period based on the characteristic state parameters of the same state evaluation indicator of each asset at each collection time node in the previous monitoring period; The calculation formula of the actual state evaluation value is: ； Where Z is the actual state evaluation value, z0 is the state evaluation conversion coefficient, n is the total number of characteristic state parameters of the state evaluation indicator at the corresponding acquisition time node in the previous monitoring period, t0i is the i-th characteristic state parameter, Ti is the standard state parameter corresponding to the i-th characteristic state parameter, and ai is the weight coefficient of the i-th characteristic state parameter; Compare the actual state evaluation value of each acquisition time node in the previous monitoring cycle with the corresponding historical state evaluation value to obtain the state evaluation value difference, and generate the credibility of the historical state evaluation value of each acquisition time node in the previous monitoring cycle based on the state evaluation value difference; If the credibility is greater than the preset credibility threshold, the historical status evaluation value will not be corrected; If the credibility is less than the preset credibility threshold, the state evaluation value mean is generated based on the historical state evaluation value of the corresponding collection time node in the previous monitoring cycle and the actual state evaluation value, and the state evaluation value difference replaces the historical state evaluation value at the corresponding collection time node.

7. The asset failure early warning system based on machine learning according to claim 6, characterized in that: A state prediction model for the corresponding asset is constructed based on the characteristic state parameters of each asset and the corresponding comprehensive state evaluation value. An impact prediction model for the corresponding asset is constructed based on the characteristic state parameters of each asset and the corresponding impact mapping table, including: All characteristic state parameters of each asset at the same acquisition time node in the previous monitoring cycle are used as the first training input data set, and the comprehensive state evaluation value of the next adjacent preset time period at the same acquisition time node is used as the corresponding first training output data set; Performing neural network training based on the first training input data set and the first training output data set to obtain a state prediction model for the corresponding asset; The mutation value of each asset's characteristic state parameter at the mutation time node of the previous monitoring cycle is used as the second training input dataset, and the influencing state parameters mapped by the mutation value at the same mutation time node and the influencing features corresponding to the influencing state parameters are used as the second training output dataset; A neural network is trained based on the second training input data set and the second training output data set to obtain an impact prediction model for the corresponding asset.

8. The asset failure early warning system based on machine learning according to claim 7, characterized in that: Obtain the real-time feature state parameter and determine whether the mutation value of the real-time feature state parameter is greater than the standard mutation value. If so, generate a predicted impact mapping table and a predicted comprehensive state evaluation value based on the impact prediction model and the state prediction model. If not, generate a predicted state evaluation value based on the state prediction model, including: Obtain the real-time characteristic state parameters of each asset at the same collection time node within the current monitoring cycle, and calculate the mutation value of each real-time characteristic state parameter; If the mutation values ​​of all real-time characteristic state parameters of the same asset are less than the standard mutation value, the real-time characteristic state parameters of the same acquisition time node are input into the state prediction model of the corresponding asset to obtain the predicted comprehensive state evaluation value of the next adjacent future preset time period of the current acquisition time node. Based on the predicted comprehensive state evaluation value of the future preset time period, it is determined whether to issue an asset fault warning; If the mutation values ​​of all real-time characteristic status parameters of the same asset are not all less than the standard mutation value, the mutation values ​​of the real-time characteristic status parameters that are greater than the standard mutation value are input into the impact prediction model of the corresponding asset to obtain the predicted impact status parameters of the next adjacent future preset time period of the current acquisition time node and the predicted impact characteristics of the predicted impact status parameters. According to the predicted impact characteristics of the predicted impact status parameters of the future preset time period and combined with the predicted comprehensive status evaluation value, it is determined whether to issue an asset fault warning.

9. The asset failure early warning system based on machine learning according to claim 8, characterized in that: Determine whether to issue an asset failure warning, including: When the mutation values ​​of all real-time characteristic state parameters of the same asset are less than the standard mutation value, the predicted comprehensive state evaluation value of the corresponding asset in the future preset period is obtained, and a first decline rate of the predicted state evaluation value is generated; If the mutation values ​​of all real-time characteristic state parameters of the same asset are not all less than the standard mutation value, the predicted failure probability of the predicted influencing state parameter in the future preset period is generated based on the predicted influencing characteristics of the predicted influencing state parameter in the future preset period; Generating a plurality of correction coefficients based on the predicted failure probabilities of all predicted influencing state parameters in a future preset period, and correcting the corresponding predicted state evaluation values ​​in the future preset period to obtain a corrected predicted comprehensive state evaluation value in the future preset period, and generating a second rate of decrease of the corrected predicted comprehensive state evaluation value; If the predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset comprehensive status evaluation value threshold and the first decline rate is less than the preset first decline rate threshold, or if the revised predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset comprehensive status evaluation value threshold and the second decline rate is less than the preset second decline rate threshold, no asset failure warning is issued; If the predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset status evaluation value threshold and the first decline rate is greater than the first preset decline rate threshold, or if the revised predicted comprehensive status evaluation values ​​for the future preset time period are all greater than the preset comprehensive status evaluation value threshold and the second decline rate is greater than the preset second decline rate threshold, a first-level asset failure warning is issued; If the predicted comprehensive status evaluation value of the future preset period is less than the preset comprehensive status evaluation value threshold, or the revised predicted comprehensive status evaluation value of the future preset period is less than the preset comprehensive status evaluation value threshold, a secondary asset failure warning is issued.

Images