CN113836806A - PHM model construction method, system, storage medium and electronic equipment - Google Patents

PHM model construction method, system, storage medium and electronic equipment Download PDF

Info

Publication number
CN113836806A
CN113836806A CN202111115053.4A CN202111115053A CN113836806A CN 113836806 A CN113836806 A CN 113836806A CN 202111115053 A CN202111115053 A CN 202111115053A CN 113836806 A CN113836806 A CN 113836806A
Authority
CN
China
Prior art keywords
data
algorithm
modeled
processing
parameter setting
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111115053.4A
Other languages
Chinese (zh)
Inventor
施建明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Technology and Engineering Center for Space Utilization of CAS
Original Assignee
Technology and Engineering Center for Space Utilization of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Technology and Engineering Center for Space Utilization of CAS filed Critical Technology and Engineering Center for Space Utilization of CAS
Priority to CN202111115053.4A priority Critical patent/CN113836806A/en
Publication of CN113836806A publication Critical patent/CN113836806A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/20Design optimisation, verification or simulation
    • G06F30/27Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/21Design, administration or maintenance of databases
    • G06F16/215Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/23Clustering techniques
    • G06F18/232Non-hierarchical techniques
    • G06F18/2321Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Quality & Reliability (AREA)
  • Probability & Statistics with Applications (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention relates to the field of model construction, in particular to a PHM model construction method, a PHM model construction system, a PHM model construction storage medium and an electronic device. The method comprises the following steps: step 1, obtaining data to be modeled; step 2, calling an intelligent parameter setting preset sub-process to process the data to be modeled; and 3, outputting the processing result as a PHM model. The invention provides a method for processing data through a standardized flow to solve the problem of inconvenient operation of modeling. The user can change the parameters according to the requirements, the basic algorithm does not need to be rewritten again in the process of establishing the model each time to obtain the required output result, the algorithm parameters only need to be adjusted, the problem that the coupling of the data and the algorithm is too tight is avoided, and the effect of improving the convenience in operation is achieved.

Description

PHM model construction method, system, storage medium and electronic equipment
Technical Field
The invention relates to the field of model construction, in particular to a PHM model construction method, a PHM model construction system, a PHM model construction storage medium and an electronic device.
Background
In the industrial fields of aerospace, aviation, nuclear power, energy, chemical engineering, ships, rail transit and the like, faults of industrial facilities and equipment can bring certain economic loss and even cause serious accidents. The traditional equipment maintenance strategy generally takes after-repair and regular maintenance as main points, the after-repair is difficult to prevent the occurrence of faults, and the regular maintenance has the defects of over-repair and under-repair. As the level of automation of industrial equipment increases, more and more sensors are integrated into industrial systems, and industrial monitoring data presents an explosive growth situation. The intelligent analysis and modeling are carried out on the equipment monitoring data, the health state of the equipment is sensed and predicted in real time, the abnormal condition of the equipment is found in time, and the fault diagnosis and prediction are carried out, so that the original monitoring data are converted into high-value information supporting maintenance decision.
There is currently a great deal of research into PHM modeling and intelligent algorithms, but most are limited to independent algorithms and models. For different data sets, the PHM modeling expert rewrites the basic algorithm to obtain the required output result every time the data is modeled and analyzed. The problem of tight coupling of data and algorithms exists, and when other algorithms need to be selected or parameters of the algorithms need to be adjusted, the operation is inconvenient.
Disclosure of Invention
The invention aims to provide a PHM model construction method, a PHM model construction system, a PHM model construction storage medium and an electronic device.
The technical scheme for solving the technical problems is as follows: a PHM model construction method comprises the following steps:
step 1, obtaining data to be modeled;
step 2, calling an intelligent parameter setting preset sub-process to process the data to be modeled;
and 3, outputting the processing result as a PHM model.
The invention has the beneficial effects that: the application provides a method for processing data through a standardized flow to solve the problem of inconvenient operation. The user can change the parameters according to the requirements, the basic algorithm does not need to be rewritten again in the process of establishing the model each time to obtain the required output result, the algorithm parameters only need to be adjusted, the problem that the data and the algorithm are coupled too tightly is avoided, and meanwhile the convenience of operation is improved.
On the basis of the technical scheme, the invention can be further improved as follows.
Further, step 2 specifically comprises:
and importing an algorithm parameter setting file according to the data to be modeled, judging whether the data to be modeled adopts a classification algorithm or a regression algorithm, if so, constructing a data set of the data to be modeled, adjusting algorithm parameters of the algorithm parameter setting file according to a judgment result, processing the data set according to the adjusted algorithm parameter setting file and generating a model corresponding to an algorithm name, and if not, processing the data to be modeled according to the algorithm parameter setting file and generating the model corresponding to the algorithm name.
Further, the algorithm comprises: clustering algorithms, classification algorithms, and regression algorithms.
Further, step 1 is preceded by:
step 0, acquiring monitoring data, judging whether the monitoring data needs to be preprocessed or not, if so, calling an intelligent parameter setting preset sub-process to generate a data preprocessing operator, inputting the monitoring data to the data preprocessing operator to obtain the data to be modeled, and if not, executing step 1, wherein the preprocessing comprises the following steps: normalization, normalization and dimension reduction.
The beneficial effect of adopting above-mentioned further scheme is that, to different processing demands, it can effectively improve the treatment effeciency to transfer intelligent parameter setting and predetermine the sub-flow, in addition, when needs adopt a plurality of algorithms to concatenate the processing data, adopt to transfer intelligent parameter setting and predetermine the sub-flow and also can improve the treatment effeciency.
Further, step 0 is preceded by:
acquiring read data, judging whether the read data needs to be subjected to data cleaning processing, if so, calling an intelligent parameter setting preset sub-process to generate a data cleaning operator, inputting the monitoring data to the data cleaning operator to obtain the monitoring data, and if not, executing the step 0, wherein the data cleaning processing comprises the following steps: and defective pixel elimination processing and steady state interception processing.
Another technical solution of the present invention for solving the above technical problems is as follows: a PHM model building system comprising:
the acquisition module is used for acquiring data to be modeled;
the calling module is used for calling an intelligent parameter setting preset sub-process to process the data to be modeled;
and the output module is used for outputting the processing result as the PHM model.
The invention has the beneficial effects that: the application provides a method for processing data through a standardized flow to solve the problem of inconvenient operation. The user can change the parameters according to the requirements, the basic algorithm does not need to be rewritten again in the process of establishing the model each time to obtain the required output result, the algorithm parameters only need to be adjusted, the problem that the data and the algorithm are coupled too tightly is avoided, and meanwhile the convenience of operation is improved.
Further, the retrieval module is specifically configured to:
and importing an algorithm parameter setting file according to the data to be modeled, judging whether the data to be modeled adopts a classification algorithm or a regression algorithm, if so, constructing a data set of the data to be modeled, adjusting algorithm parameters of the algorithm parameter setting file according to a judgment result, processing the data set according to the adjusted algorithm parameter setting file and generating a model corresponding to an algorithm name, and if not, processing the data to be modeled according to the algorithm parameter setting file and generating the model corresponding to the algorithm name.
Further, the algorithm comprises: clustering algorithms, classification algorithms, and regression algorithms.
Further, still include:
the preprocessing module is used for acquiring monitoring data, judging whether the monitoring data needs to be preprocessed or not, if so, calling an intelligent parameter setting preset sub-process to generate a data preprocessing operator, inputting the monitoring data to the data preprocessing operator to obtain the data to be modeled, and if not, executing the step 1, wherein the preprocessing comprises the following steps: normalization, normalization and dimension reduction.
Further, still include:
the data cleaning module is used for acquiring read data, judging whether the read data needs to be subjected to data cleaning processing, if so, calling an intelligent parameter setting preset sub-process to generate a data cleaning operator, inputting the monitoring data to the data cleaning operator to obtain the monitoring data, and if not, executing the step 0, wherein the data cleaning processing comprises: and defective pixel elimination processing and steady state interception processing.
Another technical solution of the present invention for solving the above technical problems is as follows: a storage medium having instructions stored therein, which when read by a computer, cause the computer to execute a PHM model construction method as in any one of the above.
The invention has the beneficial effects that: the application provides a method for processing data through a standardized flow to solve the problem of inconvenient operation. The user can change the parameters according to the requirements, the basic algorithm does not need to be rewritten again in the process of establishing the model each time to obtain the required output result, the algorithm parameters only need to be adjusted, the problem that the data and the algorithm are coupled too tightly is avoided, and meanwhile the convenience of operation is improved.
Another technical solution of the present invention for solving the above technical problems is as follows: an electronic device comprising a memory, a processor and a program stored in the memory and running on the processor, wherein the processor implements a PHM model construction method as described in any one of the above when executing the program.
The invention has the beneficial effects that: the application provides a method for processing data through a standardized flow to solve the problem of inconvenient operation. The user can change the parameters according to the requirements, the basic algorithm does not need to be rewritten again in the process of establishing the model each time to obtain the required output result, the algorithm parameters only need to be adjusted, the problem that the data and the algorithm are coupled too tightly is avoided, and meanwhile the convenience of operation is improved.
Drawings
FIG. 1 is a schematic flow chart of a PHM model construction method according to an embodiment of the present invention;
FIG. 2 is a system framework diagram provided by an embodiment of the PHM model construction system of the present invention;
FIG. 3 is a graph illustrating the performance degradation of a fuel cell according to an embodiment of the PHM modeling method of the present invention;
FIG. 4p1 is a graph of total engine outlet temperature T30 according to an embodiment of the PHM model construction method of the present invention;
FIG. 4P2 is a graph of total outlet pressure P30 of the high-pressure compressor provided by the embodiment of the PHM model construction method of the invention;
FIG. 4p3 is a graph of the ratio phi of the fuel flow to the outlet static pressure of the high-pressure compressor provided by the embodiment of the PHM model construction method of the invention;
FIG. 5 is a graph of normalized characteristic parameters of an engine according to an embodiment of the PHM model construction method of the present invention;
fig. 6 is a point cloud chart obtained by normalizing engine sensing data according to an embodiment of the PHM model construction method.
Detailed Description
The principles and features of this invention are described below in conjunction with examples which are set forth to illustrate, but are not to be construed to limit the scope of the invention.
As shown in fig. 1, a PHM model construction method includes:
step 1, obtaining data to be modeled;
step 2, calling an intelligent parameter setting preset sub-process to process the data to be modeled;
and 3, outputting the processing result as a PHM model.
In some possible embodiments, the present application proposes a method for processing data through a standardized flow to solve the problem of inconvenient operation. The user can change the parameters according to the requirements, the basic algorithm does not need to be rewritten again in the process of establishing the model each time to obtain the required output result, the algorithm parameters only need to be adjusted, the problem that the data and the algorithm are coupled too tightly is avoided, and meanwhile the convenience of operation is improved.
It should be noted that the to-be-built module data is sensor data for reading the monitored object, which may also be referred to as device monitoring data, and the data is generally structured data and includes multiple rows and multiple columns, where each row represents 1 data sample point, and each column corresponds to a characteristic variable, such as temperature, pressure, flow rate, and the like. The specific processing procedure of the intelligent parameter setting preset sub-flow can refer to example 1. For the intelligent parameter setting, reference is made to example 2. The policy model used in the single algorithm processing and the comprehensive policy model and template model used in the multi-algorithm processing can be referred to in example 3.
Example 1, this example takes a classification algorithm as an example to explain:
1) acquiring data to be modeled, and importing an algorithm parameter setting file;
2) judging whether the modulus data to be built adopts a classification algorithm or a regression algorithm, wherein the judgment result is yes because the classification algorithm is taken as an example in the embodiment;
3) constructing a data set of the belt modeling data through K-Fold, and adjusting algorithm parameters of an algorithm parameter setting file according to the algorithm category;
4) and processing the data set based on the adjusted algorithm parameter setting file to obtain a classification model.
In the process, the equipment monitoring data, namely the to-be-built module data, is mainly time sequence data and is multidimensional data containing a timestamp and a characteristic name (column name), the input data and the output data of the class are set into a table format, data interfaces among different algorithms are unified, and data can flow among the algorithms conveniently. The algorithm parameter setting is imported in the form of an Excel table, and the algorithm parameters are adjusted by modifying the table, so that an algorithm parameter adjusting window can be opened, and batch processing, comparison and tracing of the algorithm parameter setting are facilitated. When a classification and regression algorithm is adopted to model data, input data is required to be constructed into a data set, for example, a K-Fold method is adopted to construct a cross validation data set; other algorithms can directly process input data without constructing a data set. And the core processing module calls a basic algorithm framework, processes the input data based on the set algorithm parameters and obtains the processed data or a machine learning model. And (3) result sorting and outputting, namely sorting the processed results as expected, wherein the expected results of data cleaning and data preprocessing are the cleaned data and the preprocessed data generally, the output results of cluster analysis are the labeled data, the cluster information and the cluster score, the output results of classification are the predicted value of classification, the classification model and the model score, and the output results of regression are the predicted value of regression, the regression model and the model score. The data cleaning and data preprocessing output is data in a standard table format, and the machine learning processing output is data in a structure (struct) form, including a machine learning model, a prediction result, a model score and the like.
Embodiment 2, training and testing of the PHM model are carried out with the machine learning algorithm as the core, and the models under different algorithm parameter settings are compared. A traceable model comparison mechanism is designed, so that a user can quickly finish model batch training and batch testing, and model scores under different parameter settings are saved. In the standard algorithm design link, the problem of algorithm parameter adjustment is considered, the parameter setting of the current algorithm is managed in the form of an Excel table, and the Excel table is automatically read to obtain corresponding algorithm parameters each time the algorithm is called. The specific process is as follows:
preparing an Excel table with algorithm parameter setting, wherein each algorithm corresponds to 1 Excel file, and a user fills in the algorithm parameters or automatically writes the algorithm parameters into the table through a program;
when the operation of the algorithm subclass is specifically executed, the program can automatically read an Excel parameter table corresponding to the algorithm;
suppose that a user needs to perform clustering analysis on data by using a KmeanCluster algorithm, and sets the value range of a parameter KValue of the clustering algorithm to be 2-8, and 7 algorithm parameter settings need to be compared. Then, the user repeatedly calls the intelligent parameter setting preset sub-process through 1 upper-layer circulation program, so that the modeling task under the 7 algorithm parameter settings can be quickly completed, and 7 models and corresponding scores are obtained;
and selecting the optimal model as a final result according to the model scores under all the parameter settings.
In addition, selecting the optimal model by comparing different algorithm combinations is also a typical requirement of PHM modeling, and the optimal algorithm combination is selected after all analysis tasks are completed by recording the model scores under each algorithm combination.
Embodiment 3, the user may have different targets for the device monitoring data to be processed by the user: only one algorithm may be needed to complete the task, such as removing outliers in the data, performing dimensionality reduction on the data, or directly performing cluster analysis on the data, and so on; it is also possible that two or more algorithms need to be concatenated to process the data, e.g. dimensionality reduction followed by clustering. In order to facilitate a user to flexibly call the algorithm in the algorithm library, a flexible data processing pipeline design is provided, and the user can flexibly use and configure the algorithm. Two programming design modes are mainly adopted to realize the monitoring data elastic processing mechanism, namely a strategy mode and a template mode. The data processing pipeline developed through the strategy mode is used for dealing with a scene of processing data by a single algorithm, and the strategy mode and the template mode are combined to deal with a scene of processing data by 2 or more than 2 algorithms.
Regarding the policy model, a DataSinglePro class needs to be defined for packing a single algorithm processing data flow, and the clustering algorithm processing data is taken as an example for explanation, here, the packing data processing flow is a key for realizing the separation of data and algorithm. The datasinglelepro class is only a name and can be artificially defined, and the class attributes include DataInput, DataOutput and strategyObj, wherein the strategyObj points to a specific policy object, and the policy object refers to an object formed after an algorithm subclass is instantiated. The class method includes an action method in addition to a class structure method. It should be noted that, in the construction method of the datasinglelepro class, a data variable to be processed must be imported and assigned to the DataInput attribute of the class. The clusteriger class is the parent class of the clustering algorithm, also called the Strategy parent class, and indicates what policy (i.e., algorithm) is used to process the data. The number of the attributes of the parent class is only 1, and the attributes are algorithm parameters; then 1 abstract method action is defined, which is abstract, so that a subclass that inherits the parent class must implement the method. The data processing is really carried out by inheriting subclasses of parent classes, and aiming at the clustering algorithm, the subclasses comprise HCClustering subclasses, Kmeans clustering subclasses and the like. And programming based on the intelligent parameter preset sub-process and placing the program in an action method of the subclass. The action method of DataSinglePro is only a packing method, and the action method of strategyObj is called, namely, the request for processing the equipment monitoring data is transferred to a specific clustering algorithm subclass object. After the DataSinglePro class is instantiated, the DataSinglePro class object takes the DataSinglePro class object as a parameter in the action method and transmits the parameter to the algorithm subclass object, so that the action method of the algorithm subclass can acquire the DataInput attribute in the DataSinglePro object, and the data to be processed can be taken by specifically executing the operation of the processing task. And returning the result after the operation method of the algorithm subclass object is processed to the DataOutput attribute of the DataSinglePro class object. What algorithmic processing is to be taken on the device monitoring data depends on which Strength object (i.e., the algorithmic subclass object) the Strategy Obj of the DataSinglePro object points to. After instantiating the DataSinglePro, the strategyObj may be reassigned from being selected by a switching algorithm, such as switching between the HCClustering algorithm and the KmeasClustering algorithm. The above process can be simply understood as: the way that the user processes the device monitoring data using a single algorithm is:
1) reading in a data file to be processed, and declaring 1 variable to store input data;
2) instantiating a DataSinglePro class while assigning a data variable to the DataInput attribute of the DataSinglePro class object;
3) instantiating a certain algorithm subclass and assigning the algorithm subclass to a policy object attribute strategyObj of a DataSinglePro class object;
4) executing the action method of the DataSinglePro class object to obtain a target result;
5) when the user needs to adopt other algorithms to process data, only the corresponding other algorithm subclasses need to be instantiated in 3).
The following example is given for a scenario of how to handle 2 or more algorithms processing data in combination with policy and template patterns:
when two or more algorithms are required to be serially connected to process equipment monitoring data, a data processing pipeline which not only accords with data processing flow logic, but also can flexibly configure the algorithms is required to be constructed. Regardless of the change of data and the change of data processing flow, the algorithm subclass remains the same, except that the selected algorithm and the algorithm organization order are different each time the data is processed. When PHM modeling is carried out based on equipment monitoring data, a set of fixed flow is adopted for processing the data by adopting multiple algorithms (2 and more than 2), the whole flow processing follows three backbone steps of data cleaning, data preprocessing and machine learning, the three backbone steps cannot be reversed in sequence, but the steps which are not required to be executed can be skipped if the steps are available. There may be 6 data processing "pipeline" templates designed:
1) selecting only a plurality of algorithms in the data cleansing step;
2) selecting only a plurality of algorithms in the data preprocessing step;
3) selecting a plurality of algorithms in two steps of data cleaning and data preprocessing;
4) selecting a plurality of algorithms in two steps of cross-data cleaning and machine learning;
5) selecting a plurality of algorithms in two steps of data preprocessing and machine learning;
6) an algorithm is selected that spans three backbone steps. It is assumed here that the machine learning algorithm employs only 1 algorithm at a time.
The above 6 templates cover all data processing requirements of the PHM user, and the user only needs to select a plurality of algorithms and concatenate them into a "pipeline" according to a certain sequence. Theoretically, the data processing task can be successfully completed by the algorithm arrangement as long as the sequence of the three backbone steps is not violated. However, based on the experience and knowledge of the inventor, some algorithms are not suitable for concatenation (although the result can be obtained) during modeling, so that a process detection program is added for the user to choose during modeling software development.
Firstly, a data processing base class DataPro is designed, fixed three major backbone steps are packaged in a template method of DataPro, and then 6 data processing subclasses DataConcretePro 1-DataConcretePro 6 are designed to cover the 6 data processing templates. In order to freely select the algorithm for the subclass, strategyObj is defined in the attribute of DataPro so that both the base class and the subclass have policy objects, exemplified by the clustering algorithm Clusteringer:
the DataPro class attributes include DataInput, DataOutput, strategyObj, Strategies, where a Strategies attribute is a set of policies that contains all the algorithm subclass objects. The class method comprises a class construction method, an action method, a template method and an abstract method corresponding to each backbone processing step. the template method is to be declared as a method of encapsulation (Sealed), a fixed data processing flow is agreed in a DataPro base class, subclasses strictly violate the sequence, but three steps of different subclasses are implemented differently. The method of three backbone steps is declared as Abstract and Access protected method (Abstract, Access) so as to fix the interface and enforce the subclass to have to implement the corresponding backbone steps. The dataconferenpro 1 and dataconferenpro 2 are all processing template subclasses having only 1 backbone step, and are described by taking dataconferenpro 1 as an example. The subclass inherits the parent class DataPro directly without defining new attributes. The construction method of the subclass rewrites the construction method of the parent class, and assigns the data variable and the algorithm set to be specifically processed to DataInput and Strategies respectively. Then there are three major backbone methods that the subclass must implement and cannot reverse order, although in reality dataconferenpro 1 only has a data cleansing step, in order to ensure that the procedure does not go wrong, the prepronstep method and MlStep method are also defined in this subclass, except that there is no substantial content inside the methods, here printed some illustrative information, respectively "this template no data preprocessing step", "this template no machine learning step". In a specific CleanStep method, a single algorithm processing mode provided by a strategy mode is used for finishing the processing tasks of the step one by one, N (N is more than or equal to 2) algorithms are supposed to be connected in series in the CleanStep, and the method of the backbone step is declared as accessed which is protected, so that the methods can only be called by an internal method, namely networked.
The four processing templates, namely the dataconferenpro 3-6, are all processing flows with steps across backbones, the design of the subclass is described by taking the most complicated dataconferenpro 6 as an example, the structure of the subclass is the same as that of the dataconferenetpro 1, the difference is that each specifically executed backbone step has actual processing operation, and the serial processing flow of the three backbone steps is shown in fig. 7. In the data cleaning and data preprocessing backbone steps, if only 1 algorithm is available, a strategy mode is adopted to finish single algorithm processing data; if there are 2 or more than 2 algorithms, the templatethod method is called to process data after instantiating DataConcretepro1 or DataConcretepro2, and the development of the backbone step processing program can be simplified through the nesting mode. The above process can be simplified as follows: the process of the user for developing the equipment monitoring data processing or modeling task by adopting 2 or more than 2 algorithms is as follows:
1) reading data to be processed;
2) instantiating a corresponding DataConcreteProX according to the algorithm type and the template partition criterion of the first section of the section, wherein X is 1,2,3,4,5 and 6 subclasses;
3) instantiating each algorithm subclass, arranging the algorithm subclasses according to a certain sequence to form an algorithm set, and assigning the set to Strategies attributes of the DataConcreteProX subclass object;
4) calling a templateMethod method of the DataConcreteProX subclass object to complete a modeling task;
5) if the user needs to change the algorithm type or the algorithm sequence, only 3) needs to be changed.
Preferably, in any of the above embodiments, step 2 is specifically:
and importing an algorithm parameter setting file according to the data to be modeled, judging whether the data to be modeled adopts a classification algorithm or a regression algorithm, if so, constructing a data set of the data to be modeled, adjusting algorithm parameters of the algorithm parameter setting file according to a judgment result, processing the data set according to the adjusted algorithm parameter setting file and generating a model corresponding to an algorithm name, and if not, processing the data to be modeled according to the algorithm parameter setting file and generating the model corresponding to the algorithm name.
It should be noted that, when the algorithm parameter is a classification algorithm or a regression algorithm, a cross validation data set is constructed by a K-Fold method; other algorithms, such as clustering algorithm, can directly process the input data without constructing a data set.
Preferably, in any of the above embodiments, the algorithm comprises: clustering algorithms, classification algorithms, and regression algorithms.
Preferably, in any of the above embodiments, step 1 further comprises, before:
step 0, acquiring monitoring data, judging whether the monitoring data needs to be preprocessed or not, if so, calling an intelligent parameter setting preset sub-process to generate a data preprocessing operator, inputting the monitoring data to the data preprocessing operator to obtain the data to be modeled, and if not, executing step 1, wherein the preprocessing comprises the following steps: normalization, normalization and dimension reduction.
It should be noted that the data preprocessing operator can be understood as a data preprocessing operation or a data preprocessing flow, and the main meaning of the data preprocessing operator is that a flow for data preprocessing is established by only setting a preset sub-flow for parameters.
In some possible embodiments, the intelligent parameter setting preset sub-process is called according to different processing requirements, so that the processing efficiency can be effectively improved.
Preferably, in any of the above embodiments, step 0 further comprises, before:
acquiring read data, judging whether the read data needs to be subjected to data cleaning processing, if so, calling an intelligent parameter setting preset sub-process to generate a data cleaning operator, inputting the monitoring data to the data cleaning operator to obtain the monitoring data, and if not, executing the step 0, wherein the data cleaning processing comprises the following steps: and defective pixel elimination processing and steady state interception processing.
It should be noted that the data cleaning operator can be understood as a data cleaning operation or a data cleaning process, and the main meaning of the data cleaning operator is that a process for cleaning data is established through an intelligent parameter setting preset sub-process.
It should be noted that the data cleansing includes, but is not limited to: missing points or wild points exist in the data, or data segments need to be intercepted, and the like.
Example 4, using as input the fuel cell life degradation test data disclosed in french FCLAB, where the total output voltage Utot of the fuel cell stack was chosen as an indicator for the health of the cell, Utot shows a decreasing trend with time mechanism, as shown in fig. 4 and also in fig. 5, indicating a gradual degradation of the performance or health level of the fuel cell. The historical data covers the whole process from the initial state of health to the final failure of the fuel cell, is 1-dimensional time sequence data, is free of state labels, and is clustered into a plurality of clusters based on an unsupervised clustering algorithm to construct a cell state of health recognition model. Common Clustering algorithms include a Kmeans algorithm, a Hierarchical Clustering (HC) algorithm, a MidFilter + Kmeans algorithm, a MidFilter + HC algorithm and the like, the two algorithms can have different parameter settings, and in addition, data dead pixels can be removed through a median filtering algorithm before Clustering, and the data can be cleaned. PHM model batch training and comparison are carried out based on fuel cell degradation data, algorithm selection and parameter setting shown in the following table 1 are covered, and 28 clustering models are generated in total, wherein MidFilter + Kmeans are Kmeans added with a filtering algorithm, and MidFilter + HC is HC added with the filtering algorithm.
TABLE 1 Intelligent modeling comparison of Fuel cell data
Figure BDA0003275276880000141
The method is characterized in that the method adopts the simulation of 'Run-to-Failure, RtF' of an engine provided by a NASA Ames fault prediction data center as a case, and forms multidimensional monitoring data by using three sensor data as input, wherein the multidimensional monitoring data are respectively an engine outlet total temperature value T30, an HPC (high pressure compressor) outlet total pressure P30 and a ratio phi of fuel flow to HPC outlet static pressure (Ps 30).
The raw data 1 had 20631 points and included RtF simulations (representing 100 engines), wherein the data curve of the sensor data of 1 simulation (#2 engine) is shown in fig. 4.
Since the three characteristic parameters have different physical dimensions, the normalization processing is a necessary link, and fig. 6 is a three-dimensional space point cloud chart of all points after the normalization processing.
The data covers the whole process from the initial health state to the final failure of 100 engines, is three-dimensional time sequence data without state labels, and based on a clustering machine learning algorithm, historical data are clustered into a plurality of clusters to construct an engine health state identification model.
Here, four data modeling approaches are compared: 1) normalizing + Kmeans clustering; 2) normalization + HC clustering; 3) normalization + KNN outlier rejection + Kmeans clustering; 4) normalization + KNN outlier rejection + HC clustering. Normalization and KNN outlier rejection are data preprocessing steps, wherein parameters of a KNN outlier rejection algorithm are K to 4000, and 10 points with the most outliers are rejected; the parameters of the clustering algorithm are the number of clusters, which are 3/4/5/6 respectively, and the scoring results of the final 16 clustering models are shown in table 2.
TABLE 2 Intelligent modeling comparison of Engine data
Figure BDA0003275276880000151
As shown in fig. 2, a PHM model construction system includes:
an obtaining module 100, configured to obtain data to be modeled;
the calling module 200 is used for calling an intelligent parameter setting preset sub-process to process the data to be modeled;
and an output module 300, configured to output the processing result as a PHM model.
In some possible embodiments, the present application proposes a method for processing data through a standardized flow to solve the problem of inconvenient operation. The user can change the parameters according to the requirements, the basic algorithm does not need to be rewritten again in the process of establishing the model each time to obtain the required output result, the algorithm parameters only need to be adjusted, the problem that the data and the algorithm are coupled too tightly is avoided, and meanwhile the convenience of operation is improved.
Preferably, in any of the above embodiments, the retrieving module 200 is specifically configured to:
and importing an algorithm parameter setting file according to the data to be modeled, judging whether the data to be modeled adopts a classification algorithm or a regression algorithm, if so, constructing a data set of the data to be modeled, adjusting algorithm parameters of the algorithm parameter setting file according to a judgment result, processing the data set according to the adjusted algorithm parameter setting file and generating a model corresponding to an algorithm name, and if not, processing the data to be modeled according to the algorithm parameter setting file and generating the model corresponding to the algorithm name.
Preferably, in any of the above embodiments, the algorithm comprises: clustering algorithms, classification algorithms, and regression algorithms.
Preferably, in any of the above embodiments, further comprising:
the preprocessing module is used for acquiring monitoring data, judging whether the monitoring data needs to be preprocessed or not, if so, calling an intelligent parameter setting preset sub-process to generate a data preprocessing operator, inputting the monitoring data to the data preprocessing operator to obtain the data to be modeled, and if not, executing the step 1, wherein the preprocessing comprises the following steps: normalization, normalization and dimension reduction.
Preferably, in any of the above embodiments, further comprising:
the data cleaning module is used for acquiring read data, judging whether the read data needs to be subjected to data cleaning processing, if so, calling an intelligent parameter setting preset sub-process to generate a data cleaning operator, inputting the monitoring data to the data cleaning operator to obtain the monitoring data, and if not, executing the step 0, wherein the data cleaning processing comprises: and defective pixel elimination processing and steady state interception processing.
Another technical solution of the present invention for solving the above technical problems is as follows: a storage medium having instructions stored therein, which when read by a computer, cause the computer to execute a PHM model construction method as in any one of the above.
In some possible embodiments, the present application proposes a method for processing data through a standardized flow to solve the problem of inconvenient operation. The user can change the parameters according to the requirements, the basic algorithm does not need to be rewritten again in the process of establishing the model each time to obtain the required output result, the algorithm parameters only need to be adjusted, the problem that the data and the algorithm are coupled too tightly is avoided, and meanwhile the convenience of operation is improved.
Another technical solution of the present invention for solving the above technical problems is as follows: an electronic device comprising a memory, a processor and a program stored in the memory and running on the processor, wherein the processor implements a PHM model construction method as described in any one of the above when executing the program.
In some possible embodiments, the present application proposes a method for processing data through a standardized flow to solve the problem of inconvenient operation. The user can change the parameters according to the requirements, the basic algorithm does not need to be rewritten again in the process of establishing the model each time to obtain the required output result, the algorithm parameters only need to be adjusted, the problem that the data and the algorithm are coupled too tightly is avoided, and meanwhile the convenience of operation is improved.
The reader should understand that in the description of this specification, reference to the description of the terms "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described method embodiments are merely illustrative, and for example, the division of steps into only one logical functional division may be implemented in practice in another way, for example, multiple steps may be combined or integrated into another step, or some features may be omitted, or not implemented.
The above method, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention essentially or partially contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-only memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
While the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (10)

1. A PHM model construction method is characterized by comprising the following steps:
step 1, obtaining data to be modeled;
step 2, calling an intelligent parameter setting preset sub-process to process the data to be modeled;
and 3, outputting the processing result as a PHM model.
2. The PHM model construction method according to claim 1, wherein the step 2 specifically comprises:
and importing an algorithm parameter setting file according to the data to be modeled, judging whether the data to be modeled adopts a classification algorithm or a regression algorithm, if so, constructing a data set of the data to be modeled, adjusting algorithm parameters of the algorithm parameter setting file according to a judgment result, processing the data set according to the adjusted algorithm parameter setting file and generating a model corresponding to an algorithm name, and if not, processing the data to be modeled according to the algorithm parameter setting file and generating the model corresponding to the algorithm name.
3. The PHM model construction method according to claim 2, wherein the algorithm comprises: clustering algorithms, classification algorithms, and regression algorithms.
4. The PHM model construction method according to claim 3, wherein step 1 is preceded by:
step 0, acquiring monitoring data, judging whether the monitoring data needs to be preprocessed or not, if so, calling an intelligent parameter setting preset sub-process to construct a data preprocessing operator, inputting the monitoring data to the data preprocessing operator to obtain the data to be modeled, and if not, executing step 1, wherein the preprocessing comprises the following steps: normalization, normalization and dimension reduction.
5. The PHM model construction method according to claim 4, wherein step 0 is preceded by:
acquiring read data, judging whether the read data needs to be subjected to data cleaning processing, if so, calling an intelligent parameter setting preset sub-process to construct a data cleaning operator, inputting the monitoring data to the data cleaning operator to obtain the monitoring data, and if not, executing the step 0, wherein the data cleaning processing comprises the following steps: and defective pixel elimination processing and steady state interception processing.
6. A PHM model building system, comprising:
the acquisition module is used for acquiring data to be modeled;
the calling module is used for calling an intelligent parameter setting preset sub-process to process the data to be modeled;
and the output module is used for outputting the processing result as the PHM model.
7. The PHM model building system according to claim 6, wherein the invoking module is specifically configured to:
and importing an algorithm parameter setting file according to the data to be modeled, judging whether the data to be modeled adopts a classification algorithm or a regression algorithm, if so, constructing a data set of the data to be modeled, adjusting algorithm parameters of the algorithm parameter setting file according to a judgment result, processing the data set according to the adjusted algorithm parameter setting file and generating a model corresponding to an algorithm name, and if not, processing the data to be modeled according to the algorithm parameter setting file and generating the model corresponding to the algorithm name.
8. The PHM model building system according to claim 7, wherein the algorithm comprises: clustering algorithms, classification algorithms, and regression algorithms.
9. A storage medium having stored therein instructions, which when read by a computer, cause the computer to execute a PHM model construction method according to any one of claims 1 to 5.
10. An electronic device comprising a memory, a processor and a program stored in the memory and running on the processor, wherein the processor implements a PHM model construction method according to any one of claims 1 to 5 when executing the program.
CN202111115053.4A 2021-09-23 2021-09-23 PHM model construction method, system, storage medium and electronic equipment Pending CN113836806A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111115053.4A CN113836806A (en) 2021-09-23 2021-09-23 PHM model construction method, system, storage medium and electronic equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111115053.4A CN113836806A (en) 2021-09-23 2021-09-23 PHM model construction method, system, storage medium and electronic equipment

Publications (1)

Publication Number Publication Date
CN113836806A true CN113836806A (en) 2021-12-24

Family

ID=78969399

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111115053.4A Pending CN113836806A (en) 2021-09-23 2021-09-23 PHM model construction method, system, storage medium and electronic equipment

Country Status (1)

Country Link
CN (1) CN113836806A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114662371A (en) * 2022-05-24 2022-06-24 深圳市信润富联数字科技有限公司 Knowledge distillation-based light PHM system implementation method, device and system
CN116627761A (en) * 2023-05-22 2023-08-22 中国科学院空间应用工程与技术中心 PHM modeling and modeling auxiliary system and method based on big data frame

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106779087A (en) * 2016-11-30 2017-05-31 福建亿榕信息技术有限公司 A kind of general-purpose machinery learning data analysis platform
CN111612132A (en) * 2020-05-20 2020-09-01 广东电网有限责任公司 Artificial intelligence algorithm development system, training method, device and medium
CN112784501A (en) * 2021-03-23 2021-05-11 中国核电工程有限公司 Modeling system and method for residual life prediction model of equipment and prediction system

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106779087A (en) * 2016-11-30 2017-05-31 福建亿榕信息技术有限公司 A kind of general-purpose machinery learning data analysis platform
CN111612132A (en) * 2020-05-20 2020-09-01 广东电网有限责任公司 Artificial intelligence algorithm development system, training method, device and medium
CN112784501A (en) * 2021-03-23 2021-05-11 中国核电工程有限公司 Modeling system and method for residual life prediction model of equipment and prediction system

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
施建明等: "空间站有效载荷预测性维护支持系统设计", 《载人航天》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114662371A (en) * 2022-05-24 2022-06-24 深圳市信润富联数字科技有限公司 Knowledge distillation-based light PHM system implementation method, device and system
CN116627761A (en) * 2023-05-22 2023-08-22 中国科学院空间应用工程与技术中心 PHM modeling and modeling auxiliary system and method based on big data frame
CN116627761B (en) * 2023-05-22 2024-04-05 中国科学院空间应用工程与技术中心 PHM modeling and modeling auxiliary system and method based on big data frame

Similar Documents

Publication Publication Date Title
Zhang et al. Manifold: A model-agnostic framework for interpretation and diagnosis of machine learning models
US5014327A (en) Parallel associative memory having improved selection and decision mechanisms for recognizing and sorting relevant patterns
CN113836806A (en) PHM model construction method, system, storage medium and electronic equipment
CN112084237A (en) Power system abnormity prediction method based on machine learning and big data analysis
CN111460728A (en) Method and device for predicting residual life of industrial equipment, storage medium and equipment
EP2410312A1 (en) A method for computer-assisted analyzing of a technical system
US20100223218A1 (en) Data processing apparatus and method for automatically generating a classification component
Ren et al. Surrogate model assisted cooperative coevolution for large scale optimization
CN102291392A (en) Hybrid intrusion detection method based on bagging algorithm
CN112149909A (en) Ship oil consumption prediction method and device, computer equipment and storage medium
CN114970715A (en) Variable working condition fault diagnosis method and system under small sample and unbalanced data constraint
EP3839680B1 (en) Method and device for controlling a machine using principal component analysis
CN106846170B (en) Generator set trip monitoring method and monitoring device thereof
CN114422262A (en) Industrial control network intrusion detection model construction method based on automatic machine learning
CN114021425A (en) Power system operation data modeling and feature selection method and device, electronic equipment and storage medium
CN113935413A (en) Distribution network wave recording file waveform identification method based on convolutional neural network
CN115936926A (en) SMOTE-GBDT-based unbalanced electricity stealing data classification method and device, computer equipment and storage medium
CN113378722B (en) Behavior identification method and system based on 3D convolution and multilevel semantic information fusion
CN111737319B (en) User cluster prediction method, device, computer equipment and storage medium
CN114528906A (en) Fault diagnosis method, device, equipment and medium for rotary machine
CN113535522A (en) Abnormal condition detection method, device and equipment
CN116776134B (en) Photovoltaic output prediction method based on PCA-SFFS-BiGRU
KR102489115B1 (en) Method and Apparatus for Deep Machine Learning for Vision Inspection of a Manufactured Product
CN112749750B (en) Search engine aggregation analysis method and system based on k-means clustering algorithm
Young et al. Semi-automated knowledge capture and representation for the development of knowledge based systems

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20211224

RJ01 Rejection of invention patent application after publication