RH vacuum exhaust system fault positioning method and system based on hybrid expert system

Abstract

The invention provides an RH vacuum exhaust system fault positioning method and system based on a hybrid expert system, and the method comprises the steps: obtaining the production process data of an RH vacuum exhaust system, and carrying out the preprocessing of the data; according to the preprocessed production process data of the RH vacuum exhaust system, recording a constructed time sequence K mean value based on dynamic time warping as DTKmeans algorithm training to obtain a fault threshold value, and calculating a fault flag bit; and according to the fault flag bit, summarizing various fault types and a judgment rule base by the hybrid expert system, and calculating a fault position. According to the method, the RH process vacuum exhaust system can provide fault point positioning as quickly as possible by using the artificial intelligence algorithm, and the fault point finding efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automation, in particular, to an RH vacuum exhaust system fault positioning method and system based on a hybrid expert system. BACKGROUND

[0002] The vacuum exhaust system of RH refining is one of the key equipment for steelmaking, which directly affects the degassing effect of molten steel. The main fault types are two types: faults caused by external input, such as condenser cooling water temperature, condenser cooling water flow, steam pressure, etc., and faults of RH components, such as water ring pump running capacity, booster pump switch, etc. Experienced main operators are often needed to determine the fault position of the RH vacuum exhaust system, and the training cycle of experienced main operators is long, the artificial judgment load is large, and excellent operation experience is difficult to deposit. The existing fault positioning technology of the RH vacuum exhaust system mainly relies on experienced main operators to repeatedly test under load during production gaps to roughly determine the position, which not only consumes time and effort, but also relies on artificial experience.

[0003] Therefore, there is a need in the market for an RH vacuum exhaust system fault positioning method and system based on a hybrid expert system that uses artificial intelligence algorithms to provide fault point positioning as quickly as possible and improve the efficiency of finding fault points. SUMMARY

[0004] In view of the defects in the prior art, the purpose of the present application is to provide an RH vacuum exhaust system fault positioning method and system based on a hybrid expert system.

[0005] According to the RH vacuum exhaust system fault positioning method based on a hybrid expert system provided by the present application, the method comprises the following steps:

[0006] Step S1: Obtain RH vacuum exhaust system production process data, and pre-process the data;

[0007] Step S2: According to the pre-processed RH vacuum exhaust system production process data, a fault threshold is obtained by training a time series K-means based on dynamic time warping, denoted as DTKmeans algorithm, and a fault flag is calculated.

[0008] Step S3: According to the fault flag, the hybrid expert system summarizes various fault types and judgment rule library, and calculates the fault position.

[0009] Preferably, the pre-processing includes missing value processing, outlier removal, duplicate value processing, data type conversion, standard deviation standardization and time alignment.

[0010] Preferably, for the features that can set the threshold τ, if the current feature is lower than the threshold, it is considered as the fault state flag = 1; if the current feature is higher than the threshold, it is the normal state flag = 0, where flag represents the fault flag bit;

[0011] For the features that cannot directly provide the threshold τ, a time series K-means algorithm based on dynamic time warping (DTKmeans) is constructed to assign labels of fault state and normal state.

[0012] Preferably, the label assignment by the DTKmeans algorithm includes the following steps:

[0013] Step S2.1: jointly process the expert's experience knowledge, and label the pre-processed RH vacuum exhaust system production process data with labels including normal state and fault state;

[0014] Step S2.2: initialize the cluster center, randomly select k data points as the initial cluster center C = {C1, C2,..., Ck}, where each Ck K ∈R i n n , R n represents an n-dimensional real vector space;

[0015] Step S2.3: calculate the DTW distance of each point to the cluster center, the formula is as follows:

[0016]

[0017] wherein, is the set of all possible alignment paths π = (π1, π2), P is the path length, represents the pre-processed time series data set;

[0018] Step S2.4: assign data points, for each data point calculate the distance of the data point to the current cluster center, and assign the data point to the nearest cluster center;

[0019] Step S2.5: update the cluster center, recalculate the average of all points in each cluster, and update the center point of each cluster, and then return to step S2.3 until the clustering result converges.

[0020] Preferably, the distance calculation formula in the step S2.4 is as follows:

[0021]

[0022] wherein, represents the data point currently to be assigned, represents data point to all cluster centers; K represents the RH vacuum exhaust system current label mark;

[0023] The allocation strategy is as follows:

[0024]

[0025] wherein, data DTW distance with normal cluster center, new data x new DTW distance with fault cluster center.

[0026] Preferably, the step S3 comprises:

[0027] Step S3.1: determining the corresponding fault flag according to the fault threshold obtained in step S2, summarizing the positioning rules of the fault through the fault flag, and constructing a fault knowledge base;

[0028] Step S3.2: constructing an inference engine, wherein the inference engine comprises rules for solving specific problems, i.e., knowledge obtained from the knowledge base.

[0029] Preferably, the strategy of the inference engine comprises forward reasoning and backward reasoning:

[0030] The forward reasoning is used for the inference engine to combine all facts and IF-THEN-ELSE rules, and finally deduce the result;

[0031] The backward reasoning is used for the inference engine to find the answer to the problem according to what has happened.

[0032] Preferably, the forward reasoning comprises:

[0033] Manually selecting the RH vacuum exhaust system features to generate a fact database, calculating the fault flag according to the features, and finally forming a knowledge base composed of fault rules;

[0034] Then it is judged whether the fact database input by the user contains the rule set of the knowledge base. If it contains, the corresponding fault position corresponding to the corresponding rule is output, otherwise the corresponding rule and the corresponding fault position are added to the fact database.

[0035] Preferably, the backward reasoning comprises:

[0036] Step S3.2.1: inputting the fact database and the target fault position;

[0037] Step S3.2.2: exporting the required facts for the corresponding fault position according to the knowledge base;

[0038] Step S3.3.3: Loop through the required fact list, determine whether each element exists in the fact database, if it exists, continue to loop, if it does not exist, reasoning fails, until the end of the required fact list is looped through and each fact can exist in the fact database, then reasoning succeeds.

[0039] According to the RH vacuum exhaust system fault positioning system based on the mixed expert system provided by the application, the following are included:

[0040] Module M1: obtaining RH vacuum exhaust system production process data, and pre-processing the data;

[0041] Module M2: according to the pre-processed RH vacuum exhaust system production process data, a fault threshold is obtained by training a dynamic time warping-based time sequence K-means algorithm, and a fault flag is calculated.

[0042] Module M3: according to the fault flag, the mixed expert system summarizes various fault types and judgment rule bases, and calculates the fault position.

[0043] Compared with the prior art, the application has the following beneficial effects:

[0044] According to the related production process data of the RH vacuum exhaust system, the application combines field knowledge and machine learning algorithms to calculate the fault flag, and constructs a mixed expert system to timely infer the fault occurrence point, thereby improving the efficiency of manually finding the fault point and reducing the fault processing cycle. BRIEF DESCRIPTION OF DRAWINGS

[0045] Other features, objects and advantages of the application will become more apparent from the following detailed description of non-limiting embodiments, made with reference to the accompanying drawings:

[0046] Figure 1 The application is a working method flowchart. DETAILED DESCRIPTION

[0047] The application will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the application, but do not limit the application in any form. It should be noted that for those skilled in the art, without departing from the concept of the application, a number of changes and improvements can be made. These all belong to the protection scope of the application.

[0048] This invention first calculates fault flag bits for some equipment in the RH vacuum exhaust system based on domain knowledge; then, for equipment without relevant knowledge, it calculates fault flag bits according to the Dynamic timewarping based TimeSeriesKMeans algorithm (DTKmeans); finally, it constructs a hybrid expert system based on fault flag bits and process experience to form various fault judgment rules and realize fault location.

[0049] According to the present invention, a fault location method for an RH vacuum exhaust system based on a hybrid expert system is provided, such as... Figure 1 As shown, it includes:

[0050] Step S1: Obtain the production process data of the RH vacuum exhaust system and preprocess the data. The production process data S = {x1, x2, ... x} for different furnace batches of the RH vacuum exhaust system. n}, where n is the total number of furnace cycles. Data preprocessing is performed to obtain the preprocessed data. The preprocessing includes missing value handling, outlier removal, duplicate value handling, data type conversion, standard deviation standardization, and time alignment.

[0051] Missing value handling involves detecting and processing missing values ​​in the data, including deleting records containing missing values, filling missing values ​​with default values, or estimating missing values ​​through interpolation or other methods. Outlier removal involves identifying and processing outliers in the data, including deleting outliers, replacing outliers using statistical methods, or adjusting outliers using smoothing techniques. Duplicate value handling involves detecting and removing duplicate records in the data. Data type conversion involves ensuring that data columns have the correct data type. For example, converting date columns to date and time formats, text columns to appropriate text types, and ensuring that numeric columns have the correct numeric types. Standard deviation standardization involves removing the mean from each feature and scaling it to unit variance, so that the transformed feature distribution has a mean μ of 0 and a standard deviation σ of 1.

[0052]

[0053] The time alignment is achieved by padding with zeros to convert data points at different time scales to the same time scale, ensuring that data at all time scales are compared within the same time window.

[0054] Step S2: Based on the preprocessed RH vacuum exhaust system production process data, the fault threshold is obtained by training the constructed time series K-means algorithm (DTKmeans) based on dynamic time warping, and the fault flag is calculated.

[0055] For each feature, the process expert sets a threshold τ. If a feature (e.g., 1B steam pressure) falls below a certain value, it is considered a fault state (flag=1); otherwise, it is a normal state (flag=0). The formula is as follows:

[0056]

[0057] for example:

[0058] When "jet pump distributor steam pressure < 1.1 kPa", flag = 1

[0059] When "jet pump distributor steam pressure > 1.1 kPa", flag = 0

[0060] For features where process experts cannot directly provide the threshold τ, this invention constructs a dynamic time warping based TimeSeriesKMeans (DTKmeans) algorithm to assign labels for fault states (flag=1) and normal states (flag=0), specifically including the following steps:

[0061] Step S2.1: In conjunction with the experience and knowledge of process experts, label the production process data of the pretreated RH vacuum exhaust system. The labels include normal status and fault status.

[0062] Step S2.2: Initialize cluster centers. The DTKmeans algorithm randomly selects k data points as initial cluster centers C = {C1, C2, ..., C...} K}, where each C i ∈R n .

[0063] Step S2.3: Calculate the DTW distance from each point to the cluster center. Considering the inconsistent RH process completion times for each furnace batch, traditional Euclidean distance may fail to match the remaining points of the longer time series when measuring the similarity distance between two time series of unequal lengths. For time series data with inconsistent time lengths, Dynamic Time Warping (DTW) distance can capture changes or pattern matches in the time series, and is therefore suitable for data with temporal relationships.

[0064] For a given preprocessed time series dataset Each of them Given a time series, define the DTW distance matrix D∈R between all pairs of time series. n×n ,in The formula is as follows:

[0065]

[0066] in, Let P be the set of all possible alignment paths π = (π1, π2), where P is the path length, and D ∈ R. n×n It is a symmetric distance matrix that satisfies D ij =D ji .

[0067] Step S2.4: Assign data points. The data with the smallest DTW distance to a cluster center usually means that the new data is most likely to belong to that cluster. For each data point... The DTKmeans algorithm calculates the distance (based on values ​​in the distance matrix) from the data point to the current cluster center and assigns the data point to the nearest cluster center. Specifically, it compares new data... Distance to all cluster centers Where k∈{1,2,...,K), find the cluster center C with the smallest distance. k The formula is as follows:

[0068]

[0069] Since the RH vacuum exhaust system only distinguishes between normal operating conditions and fault conditions, K=2, meaning the label for the normal operating condition is 0 (C0=0, flag=0), and the label for the fault condition is 1 (C1=1, flag=1). Simultaneously, data... The DTW distance from the "normal" cluster center is New data x new The DTW distance from the "fault" cluster center is Therefore, the labeling rules for the new data are as follows:

[0070]

[0071] Step S2.5: Update cluster centers. The DTKmeans algorithm recalculates the average of all points in each cluster and updates the center point of each cluster, then returns to step S2.3 until the clustering results converge.

[0072] Step S3: Based on the fault flag, the hybrid expert system summarizes various fault types and judgment rule bases to deduce the fault location. Step S3 includes:

[0073] Step S3.1: Constructing a knowledge base. Knowledge representation is a method for organizing and formalizing knowledge in a knowledge base. This invention uses production rules to simulate the thinking of RH refining experts. Specifically, if the preconditions are met, the production rule generates a corresponding action or conclusion, in the form of an IF-THEN-ELSE rule. According to step S2, all equipment data of the RH vacuum exhaust system have obtained fault thresholds and formed fault flags. Based on process experience, these fault flags can summarize the fault location rules, forming a fault knowledge base. For example:

[0074] If flag 2C冷凝器出水温度 =1, and flag 2C真空度 =1

[0075] Then the 5EA steam valve leaked.

[0076] Common fault locations are shown in the table below:

[0077]

[0078]

[0079] Step S3.2: Construct an inference engine, which includes rules for solving specific problems, i.e., knowledge obtained from a knowledge base. When attempting to answer a user's query, the inference engine selects facts and rules to apply, providing reasoning based on information from the knowledge base. The inference engine of the RH fault location hybrid expert system uses the following strategy:

[0080] Forward reasoning: The inference engine considers all facts and IF-THEN-ELSE rules, ultimately inferring a result to answer the question "What will happen next?". The steps include manually selecting RH vacuum exhaust system features to generate a fact database, deriving fault flags based on these features, and finally forming fault rules to create a knowledge base. Then, it determines whether the user-input fact database contains the rule set of the knowledge base. If it does, it outputs the fault location corresponding to the rule; otherwise, it adds the corresponding rule and fault location to the fact database.

[0081] Backward reasoning: The inference engine finds the answer to the question, i.e., "Why did this happen?", based on what has already occurred. The steps include the following:

[0082] Step S3.2.1: Input the fact database and the target fault location.

[0083] Step S3.2.2: Export the facts required for the corresponding fault location from the knowledge base. For example, if the 5EA steam valve leak is selected, the required facts are: [flag] 2C冷凝器出水温度 =1, flag 2C真空度 =1).

[0084] Step S3.3.3: Loop through the required list of facts, such as [flag] 2C冷凝器出水温度 =1, flag 2C真空度 =1]. Check if each element exists in the fact database. If it exists, continue iterating; otherwise, the reasoning fails. Continue iterating until the required list of facts is exhausted and every fact exists in the fact database. In this case, the reasoning succeeds.

[0085] This invention aims to determine fault thresholds based on relevant high-frequency data of the RH vacuum system using domain knowledge and big data analysis, and to construct a hybrid expert system for RH fault location. By summarizing various fault types and judgment criteria, the fault occurrence point can be inferred, thereby enabling faster and more accurate fault point inference for the entire production process.

[0086] Example 2

[0087] The present invention also provides a fault location system for an RH vacuum exhaust system based on a hybrid expert system. The fault location system for an RH vacuum exhaust system based on a hybrid expert system can be implemented by executing the process steps of the fault location method for an RH vacuum exhaust system based on a hybrid expert system. That is, those skilled in the art can understand the fault location method for an RH vacuum exhaust system based on a hybrid expert system as a preferred embodiment of the fault location system for an RH vacuum exhaust system based on a hybrid expert system.

[0088] A fault location system for an RH vacuum exhaust system based on a hybrid expert system, provided by the present invention, includes:

[0089] Module M1: Acquires production process data from the RH vacuum exhaust system and preprocesses the data. The preprocessing includes missing value handling, outlier removal, duplicate value handling, data type conversion, standard deviation standardization, and time alignment.

[0090] Module M2: Based on the preprocessed RH vacuum exhaust system production process data, a fault threshold is obtained through training the DTKmeans algorithm using a constructed time-series K-means algorithm based on dynamic time warping, and fault flag bits are calculated. For features where a threshold τ can be set, if the current feature is below the threshold, it is considered a fault state (flag = 1). If the current feature is above the threshold, it is considered a normal state (flag = 0), where flag represents the fault flag bit. For features where a threshold τ cannot be directly provided, a DTKmeans algorithm based on dynamic time warping is constructed to assign labels for fault and normal states. The label assignment using the DTKmeans algorithm includes the following modules: Module M2.1: Combining the experience and knowledge of process experts, the preprocessed RH vacuum exhaust system production process data is labeled, including normal and fault states. Module M2.2: Initialize cluster centers, randomly selecting k data points as initial cluster centers C = {C1, C2, ..., C...} K}, where each C i ∈R n R n Let represent an n-dimensional real vector space.

[0091] Module M2.3: Calculates the DTW distance from each point to the cluster center, using the following formula:

[0092]

[0093] in, Let P be the set of all possible alignment paths π = (π1, π2), where P is the path length. This represents the preprocessed time series dataset. Module M2.4: Assigns data points; for each data point... Calculate the distance from the data point to the current cluster center, and assign the data point to the nearest cluster center. The distance calculation formula in module M2.4 is as follows:

[0094]

[0095] in, This indicates the data points that need to be allocated. express The distance from each data point to all cluster centers. K represents the current label of the RH vacuum exhaust system. The allocation strategy is as follows:

[0096]

[0097] in, Representing data DTW distance from normal cluster centers Represents the new data xnew DTW distance to faulty cluster centers. Module M2.5: Update cluster centers, recalculate the average of all points in each cluster, update the center point of each cluster, and then return to module M2.3 until the clustering results converge.

[0098] Module M3: Based on the fault flag, the hybrid expert system summarizes various fault types and a judgment rule base to deduce the fault location. Module M3 includes: Module M3.1: Determines the corresponding fault flag based on the fault threshold obtained by Module M2, summarizes the fault location rules through the fault flag, and constructs a fault knowledge base. Module M3.2: Constructs an inference engine, which includes rules for solving specific problems, i.e., knowledge obtained from the knowledge base. The inference engine's strategy includes forward inference and backward inference: Forward inference is used by the inference engine to combine all facts and IF-THEN-ELSE rules to ultimately deduce the result. Backward inference is used by the inference engine to find the answer to the problem based on what has already happened. Forward inference includes: manually selecting RH vacuum exhaust system features to generate a fact database, calculating the fault flag based on the features, and finally forming a fault rule knowledge base. Then, it is determined whether the user-input fact database contains the rule set of the knowledge base. If it does, the fault location corresponding to the corresponding rule is output; otherwise, the corresponding rule and the corresponding fault location are added to the fact database. Backward inference includes: inputting the fact database and the target fault location. Export the facts required for the corresponding fault location from the knowledge base. Iterate through the list of required facts, checking if each element exists in the fact database. If it does, continue iterating. If it doesn't exist, the reasoning fails. Continue until the entire list of required facts has been traversed and every fact exists in the fact database; in this case, the reasoning succeeds.

[0099] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0100] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for fault location in a RH vacuum exhaust system based on a hybrid expert system, characterized in that, The method comprises the following steps: Step S1: obtaining RH vacuum exhaust system production process data and preprocessing the data; Step S2: according to the preprocessed RH vacuum exhaust system production process data, a fault threshold is trained by constructing a time series K-means based on dynamic time warping, denoted as DTKmeans algorithm, and a fault flag is calculated; Step S3: according to the fault flag, a mixed expert system summarizes various fault types and judgment rule base to calculate the fault position.

2. The RH vacuum exhaust system fault locating method based on a hybrid expert system according to claim 1, characterized in that, The preprocessing includes missing value processing, outlier removal, duplicate value processing, data type conversion, standard deviation standardization and time alignment.

3. The RH vacuum exhaust system fault locating method based on a hybrid expert system according to claim 1, characterized in that, For features that can set a threshold τ, if the current feature is lower than the threshold, it is considered to be a fault state flag = 1; if the current feature is higher than the threshold, it is a normal state flag = 0, wherein flag represents the fault flag; For features that cannot directly provide a threshold τ, a time series K-means based on dynamic time warping DTKmeans algorithm is constructed to assign labels of fault state and normal state.

4. The RH vacuum exhaust system fault locating method based on a hybrid expert system according to claim 3, characterized in that, Label assignment by DTKmeans algorithm includes the following steps: Step S2.1: jointly process expert experience knowledge to label the preprocessed RH vacuum exhaust system production process data, and the label includes normal state and fault state; Step S2.2: Initialization of cluster centers, randomly select k data points as initial cluster centers C = {C1, C2,..., Ck} K}, where each C i ∈ R n , R n denotes an n-dimensional real vector space; Step S2.3: calculate the DTW distance of each point to the cluster center, and the formula is as follows: wherein, is the set of all possible alignment paths π = (π1, π2), P is the path length, denotes the pre-processed time series dataset; Step S2.4: Assign data points, for each data point Calculate the distance of the data point to the current cluster centers, assign the data point to the nearest cluster center; Step S2.5: update the cluster center, recalculate the average value of all points of each cluster, and update the center point of each cluster, and then return to step S2.3 until the clustering result converges.

5. The RH vacuum exhaust system fault locating method based on a hybrid expert system according to claim 4, characterized in that, The distance calculation formula in step S2.4 is as follows: wherein, represents the data point currently requiring allocation, represents the distance of the data point to all cluster centers; K represents the current label annotation of the RH vacuum exhaust system. The assignment strategy is as follows: wherein, representing data DTW distance to normal cluster center, representing new data x new DTW distance to fault cluster center.

6. The RH vacuum exhaust system fault locating method based on a hybrid expert system according to claim 1, characterized in that, The step S3 includes: Step S3.1: determine the corresponding fault flag according to the fault threshold obtained in step S2, summarize the positioning rules of the fault occurring through the fault flag, and construct a fault knowledge base; Step S3.2: build a reasoning machine, and the reasoning machine includes rules for solving specific problems, i.e. knowledge obtained from the knowledge base.

7. The RH vacuum exhaust system fault locating method based on a hybrid expert system according to claim 6, characterized in that, The strategy of the reasoning machine includes forward reasoning and backward reasoning: The forward reasoning is used for the reasoning machine to combine all facts and IF-THEN-ELSE rules to finally deduce the result; The backward reasoning is used for the reasoning machine to find the answer to the problem according to what has happened.

8. The RH vacuum exhaust system fault locating method based on a hybrid expert system according to claim 7, characterized in that, The forward reasoning includes: Manually select RH vacuum exhaust system features to generate a fact database, calculate the fault flag according to the features, and finally form a fault rule to constitute a knowledge base; Then judge whether the fact base input by the user contains the rule set of the knowledge base, if yes, output the fault position corresponding to the corresponding rule, otherwise add the corresponding rule and the corresponding fault position to the fact database.

9. The RH vacuum exhaust system fault locating method based on a hybrid expert system according to claim 7, characterized in that, The backward reasoning includes: Step S3.2.1: input the fact database and the target fault position; Step S3.2.2: export the facts required for the corresponding fault position according to the knowledge base; Step S3.3.3: Loop through the required fact list, determine whether each element exists in the fact database, if it exists, continue to traverse; if it does not exist, the reasoning fails, until the traversal of the required fact list ends and each fact can exist in the fact database, then the reasoning succeeds.

10. A hybrid expert system based RH vacuum exhaust system fault location system characterized in that, Comprise: Module M1: Obtain the RH vacuum exhaust system production process data, and pre-process the data; Module M2: According to the pre-processed RH vacuum exhaust system production process data, the fault threshold is trained by constructing a time series K-means based on dynamic time warping, denoted as DTKmeans algorithm, and the fault flag bit is calculated; Module M3: According to the fault flag bit, the mixed expert system summarizes various fault types and judgment rule base, and calculates the fault position.

Images