FCM-based hot rolling production line finishing mill group fault rapid positioning method

Abstract

A hot rolling production line finishing mill group fault rapid positioning method based on an FCM comprises the steps that multi-source signals and working condition parameter data of a speed reducer and a distribution box of a finishing mill group are acquired through a data acquisition system, and a data set of effective variables is obtained through standardized operation; constructing a fuzzy neural network based on the FCM nonparametric membership degree; outputting state fuzzy results and corresponding probabilities of different devices through the fuzzy neural network, and performing defuzzification according to a maximum probability value principle to obtain a final device state; according to the state of each device, the state of the rolling mill and the state of the production line, rapid fault positioning is achieved, then the purposes of rapid production recovery and fault loss reduction are achieved, and stable operation of the devices and continuous production are guaranteed. According to the method, the coupling and multi-scale characteristics of multi-sensor signals of complex equipment are considered, high-accuracy fault positioning suitable for the hot rolling production line can be achieved through the three-layer fault positioning strategy of the production line layer, the rolling mill layer and the equipment layer, and the steel rolling production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to a technology in the field of hot rolling technology, in particular to a method for quickly locating faults of a finishing mill of a hot rolling production line based on fuzzy clustering (FCM) non-parametric membership. Background Art

[0002] The finishing mill is a typical equipment of the hot rolling production line in steel rolling production. Its structure is extremely complex, and there are multiple coupling relationships between energy flow, material flow, and information flow in the production process. How to quickly locate and even diagnose the faults in the finishing mill through sensor monitoring data to achieve the purpose of rapid resumption of production and reduce failure losses has always been a technical problem that enterprises are concerned about. At the same time, efficient and rapid fault location is also a key component of building an enterprise physical information fusion system. Research on rapid fault location for finishing mills is of great significance for improving the accuracy of fault location in a multiple coupling data environment, ensuring efficient and safe operation of rolling line production, and improving product quality. Summary of the invention

[0003] In view of the shortcomings of the prior art, such as the need to manually extract the fault features of equipment and component sensors, reliance on expert knowledge in the field of signal processing, the need for different feature extraction methods for different sensor signals, poor versatility and portability of fault location algorithms, and the tendency to ignore the correlation between multi-sensor signals of complex equipment, the present invention proposes a method for rapid fault location of a hot rolling production line finishing mill based on FCM. Taking into account the coupling and multi-scale characteristics of multi-sensor signals of complex equipment, a three-level fault location strategy of production line level, rolling mill level and equipment level is used to achieve high-accuracy fault location suitable for hot rolling production lines, thereby improving steel rolling production efficiency.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to a method for quickly locating faults of a hot rolling production line finishing mill unit based on FCM, comprising:

[0006] In the first step, the multi-source signals and operating parameter data of the reducer and distribution box of the finishing mill are collected through the data acquisition system, and the data set of effective variables is obtained through standardized operation;

[0007] The second step is to construct a fuzzy neural network based on FCM nonparametric membership;

[0008] The fuzzy neural network includes: a feature fuzzification unit based on FCM, a state fuzzification unit based on expert knowledge, and a neural network unit, wherein: the feature fuzzification unit based on FCM uses the FCM algorithm to iterate the clustering center and membership degree according to the input data set containing valid variables, and obtains the clustering center and membership degree expression of the final data set sample; the state fuzzification unit based on expert knowledge performs device state fuzzification processing according to the definition of different device states by expert knowledge, and obtains the corresponding codes of different device states; the neural network unit establishes the association relationship between input and output according to the feature fuzzification results and the state fuzzification results.

[0009] The third step is to output the fuzzy results of the states of different devices and their corresponding probabilities through the fuzzy neural network, and defuzzify them according to the maximum probability value principle to obtain the final device state;

[0010] The fourth step is to quickly locate the fault according to the status of each device, rolling mill and production line, so as to achieve the purpose of rapid resumption of production and reduce the loss caused by the fault, and ensure the stable operation of the equipment and continuous production.

[0011] The present invention relates to a system for realizing the above method, comprising: a data preprocessing unit, a membership function construction unit, a state coding unit, a neural network construction unit and a fault rapid positioning unit, wherein: the data preprocessing unit screens and standardizes the data according to the original rail motor monitoring data of the hot rolling production line, and extracts effective variables and features; the membership function construction unit initializes the data membership function based on the input characteristic variables and working condition parameters, based on the FCM algorithm, corrects the data set, and continuously updates the clustering center and the membership function, and converges to obtain the final membership function; the state coding unit combines the three equipment classification states under expert knowledge and the definition of the equipment state, and uses coding to realize the conversion of the equipment state to the fuzzy index; the model construction unit takes the final membership that meets the convergence condition as input, and takes the equipment fuzzy index as output to construct a BP neural network; the fault rapid positioning unit obtains different equipment states by defuzzification, and then combines the rolling mill and production line states to realize the association between the system, rolling mill state and specific equipment state, and then accurately and quickly locates the fault, so as to achieve the purpose of rapid resumption of production and reduction of fault losses, and ensure stable operation of equipment and continuous production. Technical Effects

[0012] Aiming at the multi-coupling and multi-scale characteristics of hot rolling production line sensor data, the present invention adopts FCM to perform fuzzy processing on different data distribution types, combines expert knowledge to define and classify different fault states, applies fuzzy logic to process the fuzzy transition states between different fault states, introduces neural network for adaptive adjustment and optimization, and finally obtains the final equipment, rolling mill and production line status through defuzzification operation.

[0013] Compared with the existing technology, the present invention can achieve accurate and rapid fault location. Compared with four classic membership functions based on distribution assumptions, the present invention occupies 60% of the optimal solution ratio, the lowest average error rate is 0.6858%, and the fault location accuracy is improved by 2.8270%, with better positioning accuracy and classification effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the present invention;

[0015] Figure 2 It is a schematic diagram of classification of data sets in the embodiment;

[0016] Figure 3 It is a schematic diagram of the effect of the embodiment. DETAILED DESCRIPTION

[0017] like Figure 1 As shown, this embodiment involves a method for quickly locating high-coupling faults of a hot rolling production line finishing mill based on FCM, combined with the field data of a 1580 hot rolling production line of a steel company, specifically including:

[0018] In the first step, the multi-source signals and operating parameter data of the reducer and distribution box of the finishing mill are collected through the data acquisition system, and invalid data are removed. Through standardization operations, a data set of valid variables is obtained;

[0019] The standardization operation is to calculate the standardized value Where: x ij is the element in the i-th row and j-th column of the data set, x i is the mean of all elements in the i-th row, s i is the standard deviation of the elements in the i-th row.

[0020] The second step is to construct a fuzzy neural network based on FCM non-parametric membership, which includes:

[0021] 2.1) By constructing a non-parametric membership function based on FCM, the fuzzification of features is achieved, including:

[0022] 2.1.1) Set the number of clusters c, convergence constant m, and convergence threshold ε;

[0023] 2.1.2) Initialize the membership matrix U (0) , the membership matrix stores the membership of each sample, and the membership of the sample is randomly initialized based on satisfying the constraints;

[0024] 2.1.3) Set the iteration count a=0;

[0025] 2.1.4) According to U (a)Calculate the cluster center v of the sample j ;

[0026] 2.1.5) Calculate the membership matrix U (a+1) ;

[0027] 2.1.6) When the loss function of two iterations is less than the convergence threshold of the objective function, that is, Or the data label lab calculated by FCM FCM and the true label lab true If they are consistent, stop the iteration, otherwise set a=a+1 and return to 2.1.4) to continue the iteration;

[0028] In this embodiment: the convergence threshold ε of the objective function is 0.001;

[0029] 2.1.7) When the model converges, correct the historical data set: remove the real label lab in the data set true And the data label lab obtained by the FCM algorithm FCM Inconsistent samples, that is, samples with abnormal attributes in historical data are eliminated;

[0030] 2.1.8) Use the corrected data set to calculate the cluster center v′ of the final sample j , and then calculate the final membership function expression U′, where: The membership is obtained by finding the objective function The minimum value of is obtained, where: N is the total number of historical data, c is the number of clustering groups, which is a positive integer greater than 2, and m is a constant greater than 1, which is used to control the convergence speed. The larger the m, the faster the convergence. Generally speaking, m is between 1.5 and 2.5. All historical data constitute a set X = {x1, x2, ..., x N}, x i is the value of the i-th sample, u ji For sample x i The degree of belonging to the jth category, v j is the value of the cluster center of the jth category, d(x i , v j ) is the sample x i With cluster center v j The Euclidean distance between ||x i -v j ||, in order to obtain the minimum value of the objective function under the constraints, the Lagrange multiplier method is used to obtain the cluster center v j And the membership function u ji The update formula is to iteratively calculate the cluster center and membership function until convergence. Specifically, the updated cluster center Updated membership function

[0031] 2.1.9) Based on the final membership degree obtained in step 2.1.8, determine the category of the data and achieve feature fuzzification;

[0032] 2.2) State fuzzification based on expert knowledge;

[0033] According to expert experience, the device status is fuzzy processed. For device A, there are three states at a certain moment: normal, warning, and alarm. Experts assign different coefficients to the degree to which device A belongs to the three states of normal, warning, and alarm based on experience values;

[0034] For example, if the device A corresponding to sample 1 is judged to be completely normal based on expert knowledge, then the one-hot encoding of normal, warning, and alarm corresponds to [1,0,0];

[0035] For example, if the expert believes that the device A corresponding to sample 2 is between normal and warning, then the one-hot encoding of normal, warning, and alarm corresponds to [0.5, 0.5, 0];

[0036] Through the above coding method based on expert experience, the equipment status is converted into fuzzy indicators;

[0037] 2.3) Using the feature fuzzification result obtained in step 2.1 as input and the state fuzzy index obtained in step 2.2 as output, a fuzzy neural network is constructed and trained;

[0038] The fuzzy neural network includes: a feature fuzzification unit based on FCM, a state fuzzification unit based on expert knowledge, and a neural network unit, wherein: the feature fuzzification unit based on FCM uses the FCM algorithm to iterate the clustering center and membership degree according to the input data set containing valid variables, and obtains the clustering center and membership degree expression of the final data set sample; the state fuzzification unit based on expert knowledge performs device state fuzzification processing according to the definition of different device states by expert knowledge, and obtains the corresponding codes of different device states; the neural network unit establishes the association relationship between input and output according to the feature fuzzification results and the state fuzzification results, and forms a neural network structure.

[0039] The third step is to output the fuzzy results of the states of different devices and their corresponding probabilities through the fuzzy neural network, obtain the results of the states of different devices and their corresponding probabilities, and defuzzify them according to the maximum probability value principle, that is, take the state corresponding to the maximum probability as the final state of the device;

[0040] For example, the probabilities of normal, warning, and alarm of device A are [0.8, 0.1, 0.1] respectively, so the final status of device A is normal.

[0041] In the fourth step, according to the status of each equipment, rolling mill and production line, the correspondence between the equipment status and the rolling mill status and the final production line status can be achieved, and then the production line-rolling mill-equipment can be checked layer by layer to achieve production line early warning and alarm, and rapid location of faults in the rolling mill and specific equipment, so as to achieve the purpose of rapid resumption of production and reduction of failure losses, and ensure stable operation of equipment and continuous production.

[0042] After specific practical experiments, the data of the roller motor data set of 1580 in a steel plant and the data of the rolling line of 2250 were collected. Combined with expert knowledge, the hot rolling production line data set was divided into normal data, minor faults, and serious faults according to the kurtosis index, such as Figure 2 As shown, the system status is divided into stable operation, early warning status, and alarm status. At the same time, we summarized relevant expert knowledge through exchanges with steel mills, defined the inference engine of fuzzy neural network, and designed a three-level fault location method at the production line level, rolling mill level, and equipment level.

[0043] Under the same training set and validation set conditions, the proposed FCM non-parametric membership is compared with four classic membership functions based on distribution assumptions. The results in Table 1 show that the average error rate of the present invention is 0.6858%, accounting for 60% of the optimal solution ratio among the five algorithms involved in the comparison. The present invention can solve the multi-coupling and multi-scale characteristics of hot rolling production line faults, and is better suitable for hot rolling production line fault location, with higher positioning accuracy.

[0044] Table 1 Membership function Gauss triangle Trapezoid Generalized bell shape Based on FCM non-parametric The number of optimal values 3 1 2 3 12 Average error rate 1.3777％ 4.4221％ 3.6202％ 5.0302％ 0.6858％

[0045] According to the hot rolling production line failure -> F1 ~ F7 rolling mill failure -> main reducer, distribution box failure, the fault location is carried out, forming a three-level fault location method of production line level - rolling mill level - equipment level. The location accuracy is as follows: Figure 3 As shown, the results show that the average positioning error rate of the proposed invention is 0.6858%, accounting for 60% of the optimal solution ratio among the five algorithms involved in the comparison and is not limited by the distribution assumption, and can better adapt to complex and changing environments.

[0046] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. A method for quickly locating high-coupling faults of a hot rolling production line finishing mill based on FCM, characterized in that: include: In the first step, the multi-source signals and operating parameter data of the reducer and distribution box of the finishing mill are collected through the data acquisition system, and invalid data are removed. Through standardization operations, a data set of valid variables is obtained; The second step is to construct a fuzzy neural network based on FCM nonparametric membership; The third step is to output the fuzzy results of the states of different devices and their corresponding probabilities through the fuzzy neural network, and defuzzify them according to the maximum probability value principle to obtain the final device state; The fourth step is to quickly locate the fault according to the status of each device, rolling mill and production line, so as to quickly resume production and reduce the loss caused by the fault, thus ensuring stable operation of the equipment and continuous production; The fuzzy neural network includes: a feature fuzzification unit based on FCM, a state fuzzification unit based on expert knowledge, and a neural network unit, wherein: the feature fuzzification unit based on FCM uses the FCM algorithm to iterate the clustering center and membership degree according to the input data set containing valid variables, and obtains the clustering center and membership degree expression of the final data set sample; the state fuzzification unit based on expert knowledge performs device state fuzzification processing according to the definition of different device states by expert knowledge, and obtains the corresponding codes of different device states; the neural network unit establishes the association relationship between input and output according to the feature fuzzification results and the state fuzzification results.

2. The method for quickly locating high-coupling faults of a hot rolling production line finishing mill unit based on FCM according to claim 1 is characterized in that: The standardization operation is to calculate the standardized value Where: x ij is the element in the i-th row and j-th column of the data set, x i is the mean of all elements in the i-th row, s i is the standard deviation of the elements in the i-th row.

3. The method for quickly locating high-coupling faults of a hot rolling production line finishing mill unit based on FCM according to claim 1 is characterized in that: The construction of the fuzzy neural network based on FCM non-parametric membership specifically includes: a) Fuzzification of features is achieved by constructing a non-parametric membership function based on FCM; b) State fuzzification based on expert knowledge: Based on expert experience, the equipment state is fuzzified. For equipment A, there are three states: normal, warning, and alarm at a certain moment. Experts assign different coefficients to the degree to which equipment A belongs to the three states of normal, warning, and alarm based on their experience values; c) Using the feature fuzzification result obtained in step a as input and the state fuzzy index obtained in step b as output, a fuzzy neural network is constructed and trained.

4. The method for quickly locating high-coupling faults of a hot rolling production line finishing mill unit based on FCM according to claim 3 is characterized in that: The fuzzification of the features specifically includes: i) Set the number of clusters c, convergence constant m, and convergence threshold ε; ii) Initialize the membership matrix U (0) , the membership matrix stores the membership of each sample, and the membership of the sample is randomly initialized based on satisfying the constraints; iii) set the iteration count a=0; iv) According to U (a) Calculate the cluster center v of the sample j ; v) Calculate the membership matrix U (a+1) ; vi) When the loss function of two iterations is less than the convergence threshold of the objective function, that is, Or the data label lab calculated by FCM FCM and the true label lab true If they are consistent, stop the iteration, otherwise set a=a+1 and return to step iv to continue the iteration; vii) When the model converges, correct the historical data set: remove the real label lab in the data set true And the data label lab obtained by the FCM algorithm FCM Inconsistent samples, that is, samples with abnormal attributes in historical data are eliminated; viii) Use the corrected data set to calculate the cluster center v' of the final sample j , and then calculate the final membership function expression U', where: the membership is obtained by finding the objective function The minimum value is obtained, where: N is the total number of historical data, c is the number of clustering groups, which is a positive integer greater than 2, and m is a constant greater than 1, which is used to control the convergence speed. The larger the m, the faster the convergence. All historical data constitute the set X = {x1, x2, ..., x N }, x i is the value of the i-th sample, u ji For sample x i The degree of belonging to the jth category, v j is the value of the cluster center of the jth category, d(x i ,v j ) is the sample x i With cluster center x j The Euclidean distance between ||x i -v j ||, in order to obtain the minimum value of the objective function under the constraints, the Lagrange multiplier method is used to obtain the cluster center v j And the membership function u ji The update formula is to iteratively calculate the cluster center and membership function until convergence, specifically: the updated cluster center Updated membership function ix) Based on the final membership degree obtained in step viii, the category of the data is determined to achieve feature fuzzification.

5. A system for realizing the FCM-based hot rolling production line finishing mill high coupling fault rapid positioning method as described in any one of claims 1 to 4, characterized in that: include: Data preprocessing unit, membership function building unit, state encoding unit, neural network building unit and fault fast positioning unit, wherein: the data preprocessing unit screens and standardizes the data according to the original rail motor monitoring data of the hot rolling production line, and extracts effective variables and features; The membership function construction unit initializes the data membership function based on the FCM algorithm, corrects the data set, and continuously updates the cluster center and membership function according to the input characteristic variables and operating parameters, and converges to the final membership function; the state coding unit combines the three equipment classification states and the definition of equipment state under expert knowledge, and uses coding to realize the conversion of equipment state to fuzzy indicators; the model construction unit takes the final membership that meets the convergence conditions as input and the equipment fuzzy indicators as output to construct a BP neural network; the fault rapid location unit obtains different equipment states through defuzzification, and then combines the rolling mill and production line states to realize the association between the system, rolling mill state and specific equipment state, and then accurately and quickly locates the fault, so as to achieve the purpose of rapid resumption of production and reduce fault losses, and ensure stable operation of equipment and continuous production.

Images