CN108126987B - Remote online monitoring and intelligent diagnosis method for high-speed wire rolling mill - Google Patents
Remote online monitoring and intelligent diagnosis method for high-speed wire rolling mill Download PDFInfo
- Publication number
- CN108126987B CN108126987B CN201710307331.3A CN201710307331A CN108126987B CN 108126987 B CN108126987 B CN 108126987B CN 201710307331 A CN201710307331 A CN 201710307331A CN 108126987 B CN108126987 B CN 108126987B
- Authority
- CN
- China
- Prior art keywords
- fault
- data
- diagnosis
- training
- samples
- 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.)
- Active
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
- B21B—ROLLING OF METAL
- B21B38/00—Methods or devices for measuring, detecting or monitoring specially adapted for metal-rolling mills, e.g. position detection, inspection of the product
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
- B21B—ROLLING OF METAL
- B21B33/00—Safety devices not otherwise provided for; Breaker blocks; Devices for freeing jammed rolls for handling cobbles; Overload safety devices
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
- B21B—ROLLING OF METAL
- B21B38/00—Methods or devices for measuring, detecting or monitoring specially adapted for metal-rolling mills, e.g. position detection, inspection of the product
- B21B38/006—Methods or devices for measuring, detecting or monitoring specially adapted for metal-rolling mills, e.g. position detection, inspection of the product for measuring temperature
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
- B21B—ROLLING OF METAL
- B21B38/00—Methods or devices for measuring, detecting or monitoring specially adapted for metal-rolling mills, e.g. position detection, inspection of the product
- B21B38/008—Monitoring or detecting vibration, chatter or chatter marks
Landscapes
- Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
A remote online monitoring and intelligent diagnosis system for a high-speed wire rod mill comprises a sensor network, a central node, an industrial personal computer (field processor) and a monitoring and diagnosis center, wherein the sensor network is a subnet consisting of data acquisition equipment in a mill unit, data acquired by the data acquisition equipment is transmitted to the industrial personal computer, the industrial personal computer carries out drying and conversion processing while storing the data, and the industrial personal computer carries out judgment on whether the fault of the high-speed wire rod mill exists or not, namely pre-diagnosis and outputs a result; if the structure output by the industrial personal computer judges that the high-speed wire rolling mill is in fault, the data is transmitted to the monitoring and diagnosis center, otherwise, the industrial personal computer does not transmit the data to the monitoring and diagnosis center, the monitoring and diagnosis center receives the data with the fault and then carries out further analysis on the pre-diagnosis of the industrial personal computer, and the result is output. The method can judge the type and the fault position of the current fault of the rolling mill on line in real time, and prolong the service life of equipment.
Description
Technical Field
The invention relates to the technical field of diagnosis of wire rod rolling mills, in particular to a remote online monitoring and intelligent diagnosis method for a high-speed wire rod rolling mill.
Background
A rolling mill is a widely used device for realizing a metal rolling process, and a high-speed wire rolling mill is a commonly used rolling mill. Various faults can occur along with the operation of the rolling mill, including bearing seat abrasion, box body deformation, shaft cracking and the like. For determining and locating the fault, a professional usually acquires various parameters, especially parameters, by means of various testing instruments, analyzes and determines the fault type according to the parameters, and further determines which component sends the fault. The process is determined by analysis of professional technicians when the rolling mill stops running, and online analysis and diagnosis are difficult to perform.
Disclosure of Invention
The invention aims to solve the problems in the prior art and provides a remote online monitoring and intelligent diagnosis method for a high-speed wire rod mill, which is used for determining whether a fault exists in the high-speed wire rod mill and diagnosing and positioning the fault type.
The technical scheme of the invention is as follows:
a remote online monitoring and intelligent diagnosis method for a high-speed wire rolling mill comprises data acquisition equipment, a central node, an industrial personal computer and a monitoring and diagnosis center, and is characterized in that: the data collected by the data collecting equipment are collected to a central node and then transmitted to the industrial personal computer through the local area network or the wireless network, and the industrial personal computer transmits the data to the monitoring and diagnosis center through the local area network or the wireless network.
A remote online monitoring and intelligent diagnosis method for a high-speed wire rod rolling mill is characterized by comprising the following steps:
and S1 data acquisition step: the data collected by each data collection device is collected to a central node and then transmitted to an industrial personal computer by adopting a local area network or a wireless network, and the data collection device comprises a vibration sensor, a speed sensor, a three-dimensional acceleration sensor, a displacement sensor, a rotary encoder, a temperature sensor, a power meter and a power sensor in a rolling mill set;
s2 pre-diagnosis step: the industrial personal computer stores data, simultaneously carries out noise removal, data transformation and feature extraction processing on the data, simultaneously judges whether the processed data has a fault of the high-speed wire mill, and outputs a result, if the structure output by the industrial personal computer judges that the high-speed wire mill has the fault, the data is transmitted to the monitoring and diagnosis center, otherwise, the industrial personal computer does not transmit the data to the monitoring and diagnosis center, so as to avoid transmitting a large amount of fault-free data to the diagnosis center, thereby avoiding network congestion and increasing the burden of a monitoring and diagnosis center server;
s3 precision diagnosis step: and the monitoring and diagnosis center further analyzes the pre-diagnosis result of the industrial personal computer after receiving the data with the fault so as to determine the type and the position of the fault.
In the data acquisition step S1, the vibration sensor is installed on the box, the shafting, the gear and the bearing part and acquires the three-dimensional vibration waveform comprising the rolling bearing, the transmission shaft, the pinch roll and the box, the rotary encoder is installed on the rotating shaft and acquires the rotation angle and the rotation speed comprising the rotating shaft, the temperature sensor is installed on the bearing and the motor and acquires the working temperature comprising the bearing part and the motor, and the power sensor is installed on the motor and acquires the power consumption comprising the motor.
In the data acquisition step S1, the rotary encoder performs sampling according to the following rule: the sampling period is divided into two time periods Twork and Twoit, the sampling is carried out 16 times when the rotating shaft rotates for one circle, the Twork time period is the time spent on rotating the rotating shaft for 64 circles, the sampling is carried out 1024 times in the Twok time period, the sampling is not carried out in the Twoit time period, and the time spent on rotating the rotating shaft for 8-512 circles.
The process of the pre-diagnosis step S2 is: the data of the data acquisition equipment are stored, the stored original data can be directly called to the cloud, meanwhile, transformation and feature extraction processing are carried out on the data, a feature vector containing a fault is selected through a two-classifier (HSSVM) and uploaded to a cloud interface, the feature vector containing the fault is added into an HSSVM parameter base again after the reliability is confirmed through an accurate diagnosis step to carry out HSSVM learning training so as to optimize the two-classifier, and the two-classifier (HSSVM) carries out sampling control on each sensor of the data acquisition equipment according to the fault judgment condition.
In the pre-diagnosis step S2, the data transformation and feature extraction method includes:
(1) data transformation: the data series output by one sensor in Ttwork time period is recorded as data [ k]K is 0,1, …, if data [ k [ ]]From the vibration sensor, performing one-dimensional multi-scale wavelet transform to extract singular points in each scale, and recording the transform result as data [ k ]](s)Wherein s is a scale number, s ═ 1,2, …, N; the invention takes 8 scales, namely N is 8;
(2) feature extraction: arranged in a sensor group, having MAnVibration sensor, correspondingly, there are
Assuming that the time taken for the shaft to rotate 1 turn (i.e., the rotation period) is Tc, the position of each Tc is obtained from the output data of the rotary encoder, and an attempt is made to determine the position of the shaft within one Tc periodSearch to
N × M singular points p existing onsmIf a singular point does not exist, the singular point is set to 0, otherwise, the singular point is set to be in the period
A modulus maximum of;
setting a sensor group with Q temperature sensors and 1 power sensor; defining a d-dimensional feature vector P:
p＝(〖psm|〗S＝1...N，m＝1...M，〖pq|〗q＝1...Q，pw)
d＝N×M+Q+1
wherein: m is 2-5; p is a radical ofq|q＝1…QThe average value of sampling of each temperature sensor in the period is Q, and Q is 1-3; pw is the average of the power sensor samples over the period in question, 1.
In the pre-diagnosis step S2, a two-classifier based on HSSVM is adopted, each component in the rolling mill production line corresponds to one two-classifier, and the state of each component is divided into two types: giving out fault reliability R at the same time when there is or is no fault, wherein the given fault reliability R represents the severity of the fault and is used for controlling the size of the sampling pause period Twait; the Ttwork period contains 64 Tc cycles, each Tc extracts a characteristic vector Pj, j is 1,2, …, 64, and the Pj is sent to a classifier to judge whether a fault exists; if the number of times of faults is majority in 64 judgments in the Ttwork period, judging the component to be faulty; when a certain component is judged to be in fault, data selection is started, and the current 64 Pj are transmitted to the cloud for accurate diagnosis; at any time, the unit operation maintenance personnel can also transmit 64 Pj or original data series of the current Ttwork time period to the cloud for diagnosis.
In the pre-diagnosis step S2, the process of the HSSVM learning training includes: selecting a kernel function as a Gauss function, and establishing a characteristic space of the HSSVM; firstly, collecting a batch of positive training samples P + with good cohesiveness and storing the positive training samples P + in an HSSVM training sample library; then, through iteration, a minimum hypersphere capable of wrapping all P + is found, so that parameters such as the radius, the sphere center and the like of the hypersphere are determined, and the HSSVM with the initial diagnosis capability is established; in practical application, a reward and punishment mechanism of reinforcement learning is applied, and a training sample library is optimized according to a current training result, so that the diagnostic performance of the system is continuously improved; adding the collected samples into a training sample library after accurate diagnosis is carried out on the collected samples to obtain reliable labels (including positive and negative samples), and retraining the HSSVM to determine a new hypersphere; if the new HSSVM has better classification performance, the added sample is valid, otherwise, the addition is cancelled, and the training is cancelled.
The procedure of the precise diagnosis step S3 is: the method comprises the steps of firstly establishing a fault type sample library storing training samples, then establishing a fault diagnosis and classification model, receiving a characteristic vector containing a fault from pre-diagnosis, then comparing and analyzing data and the training samples in the database according to the stored training samples, determining the fault type according to the fault diagnosis and classification model in fault classification, updating an analysis result into the training samples in the fault type sample library, and optimizing through reinforcement learning.
Alternatively, the procedure of the precise diagnosis step S3 is: the method comprises the steps of firstly establishing a fault type sample base and a fault grading knowledge base which store training samples, then establishing a fault diagnosis and classification model and a fault fuzzy evaluation model, receiving a characteristic vector containing a fault from pre-diagnosis, comparing and analyzing data and the training samples in the database according to the stored training samples, determining the fault type according to the fault diagnosis and classification model for fault classification, grading the severity of the fault according to the fuzzy evaluation model for fault grading, predicting the development trend of the fault according to the fuzzy state model, and updating the analysis result into the training samples of the fault type sample base and the fault grading knowledge base to optimize through reinforcement learning.
In the step S3 of accurate diagnosis, the preliminarily determined feature vectors with possible faults are further subjected to fault classification, where the fault types include bearing defects, bearing seat wear, bearing seat hole wear, shaft cracking, shaft deformation, gear tooth breakage, and box deformation.
In the accurate diagnosis step S3, the SOM is adopted to realize fault classification, each component corresponds to one SOM, the node number of an input layer of the SOM structure is d (the dimension of a characteristic vector P), the node number of an output layer is the number of training samples, and is 16 × 16, and the training process of the SOM is as follows:
(1) the method comprises the following steps of selecting an initial training sample, wherein the initial training sample is a sample obtained by analyzing and fitting various typical faults by a field expert by using a dynamics theory and experience, b, setting simulated faults on a rolling mill component and collecting the sample, c, the sample collected when the actual fault occurs, obtaining 256 samples through the above way, and marking the samples as Pl, l as 1,2, … and 256, wherein the samples can represent the fault types listed in the table 1, the number of the samples of each fault type is approximately the same, then carrying out normalization processing on N × M components of the Pl, which are derived from a vibration sensor, for training of an SOM (sequence of order) and the result is that output node positions corresponding to the similar input samples are adjacent on an output layer plane, and one type of fault is distributed in a sub-region;
(2) optimization of training samples: in practical application, when a true negative sample is received, the true negative sample is added into a training sample library and the SOM is retrained; optimizing the training sample library according to the current training result by applying a reward and punishment mechanism of reinforcement learning, and eliminating the inferior samples; thereby increasing the diagnostic performance of the system.
In the step S3 of precisely diagnosing, the fault classification is to divide the fault into 5 grades according to the severity; the level 1 fault is the lowest and the level 5 fault is the highest.
The precise diagnosis step S3 further includes a function of performing auxiliary diagnosis on the raw data from the pre-diagnosis by an auxiliary analysis system.
The invention has the beneficial effects that: the various sensors are reasonably arranged at each part of the rolling mill, so that the operation data of the rolling mill is collected, the data is pre-diagnosed and accurately diagnosed, the type and the fault position of the current fault of the rolling mill can be judged on line in real time, the fault of the rolling mill can be found in time, the fault of the rolling mill is prevented from going to the stage with poor performance, and the service life of equipment is prolonged.
Drawings
FIG. 1 is a diagram of the networking architecture of the present invention;
FIG. 2 is an overall process flow diagram of the present invention;
FIG. 3 is a flow chart of the fault pre-diagnosis process of the present invention;
FIG. 4 is a flow chart of the fault pinpoint diagnostic process of the present invention;
FIG. 5 is a structural diagram of the SOM of the present invention.
Detailed Description
The following further describes embodiments of the present invention with reference to the accompanying drawings:
as shown in fig. 1, the present invention mainly comprises a fault pre-diagnosis subsystem located at each rolling mill production site (client), and a fault accurate diagnosis subsystem located at a diagnosis service center (cloud).
The fault pre-diagnosis subsystem is contained in the dashed boxes M1, M2, … Mn in fig. 1. Each subsystem is composed of a sensor network and a field processor. A rolling mill production line is composed of a plurality of key components, and a sensor network is responsible for acquiring physical parameters of each key component in the rolling mill production line during operation. The sensors mounted on one component are referred to as a sensor group, denoted by Gi, i ═ 0,1, … m, see fig. 1. The fault pre-diagnosis is mainly carried out by taking key components as independent objects. The sensor network type is RS485 and Zigbee mixed network, and one-to-many star-shaped structure networking is adopted. Wherein the RS485 subnet collects the static point, and the Zigbee subnet collects the moving point. The power supply of the terminal node of the Zigbee adopts electromagnetic coupling wireless feed. The sensor network consists of a central node and a plurality of terminal nodes. Each terminal in the sensor network is composed of a sensor and a control circuit thereof.
(1) Sensor types and uses: the types of sensors used in the present invention and their uses are shown in table 1:
TABLE 1 sensor types and uses
(2) Control circuit of the sensor: the system consists of an AVR series industrial single chip microcomputer, a Zigbee communication module, an RS485 interface and the like, and controls the sampling rate, the data bit width, the sampling mode, the network communication and the like. Each sensor is synchronously sampled under the control of information provided by the rotary encoder, and the acquired data are gathered at the central node. The sampling rate is: sampling is carried out 16 times every time the rotating shaft rotates 1 circle. The data bit width is 16 bits. The sampling mode is of an equal intermittent burst type, namely: one acquisition cycle includes a sampling period Twork and a rest period Twait. The time taken by the invention for rotating the rotating shaft 64 turns is Twork and is about 1800 ms. The time spent by the rotating shaft rotating 8-512 turns is about 200 ms-16000 ms for Twait. The size of Twait is related to the severity of the fault, with higher severity of the fault yielding less Twait.
(3) Central node of sensor network: the Zigbee communication system is composed of an RAM processor, a Zigbee communication module, an RS485 interface, an Ethernet interface and the like. The functions are as follows: zigbee networking; managing and coordinating each terminal node to orderly finish data acquisition and transmission; and the Ethernet interface is utilized to collect and transmit the original data collected by each terminal to a field processor and the like. And the central node of the sensor network and the field processor transmit data by using an Ethernet interface. The original data is also stored in a local memory, the data in the latest period is stored and used as a history file, and the data can be remotely called if the cloud end needs the data. The local storage is a solid state disk, the capacity is 1T byte, and the original data of about 12 months can be buffered.
As shown in fig. 2, a remote online monitoring and intelligent diagnosis method for a high-speed wire rolling mill is characterized by comprising the following steps:
and S1 data acquisition step: data acquired by each data acquisition device is collected to a central node and then transmitted to an industrial personal computer by adopting a local area network or a wireless network;
s2 pre-diagnosis step: the industrial personal computer stores data, simultaneously carries out noise removal, data transformation and feature extraction processing on the data, simultaneously judges whether the processed data has a fault of the high-speed wire mill, and outputs a result, if the structure output by the industrial personal computer judges that the high-speed wire mill has the fault, the data is transmitted to the monitoring and diagnosis center, otherwise, the industrial personal computer does not transmit the data to the monitoring and diagnosis center, so as to avoid transmitting a large amount of fault-free data to the diagnosis center, thereby avoiding network congestion and increasing the burden of a monitoring and diagnosis center server;
s3 precision diagnosis step: and the monitoring and diagnosis center further analyzes the pre-diagnosis result of the industrial personal computer after receiving the data with the fault so as to determine the type and the position of the fault.
Furthermore, the vibration sensor is arranged on the box body, the shaft system, the gear and the bearing part and collects three-dimensional vibration waveforms comprising a rolling bearing, a transmission shaft, a pinch roll and the box body; the rotary encoder is arranged on the rotating shaft and acquires the rotating angle and the rotating speed of the rotating shaft; the temperature sensor is arranged on the bearing and the motor and used for collecting the working temperature of the bearing part and the motor; the power sensor is arranged on the motor and used for collecting power consumption of the motor; further, the vibration waveforms include rolling bearing vibration waveforms and rolling mill vibration waveforms; the working temperature comprises the temperature of the rolling mill, the temperature of a bearing part and the temperature of a rolling bearing.
Further, the rotary encoder performs sampling according to the following rule: the sampling period is divided into two time periods Twork and Twoit, the sampling is carried out for 16 times when the rotating shaft rotates for one circle, and the bit width of the sampling data is 16 bits; the Tway time period is the time spent on rotating the rotating shaft for 64 circles, as the best embodiment, the value is 1800ms, sampling is carried out for 1024 times in the Tway time period, sampling is not carried out in the Tway time period, and the time is the time spent on rotating the rotating shaft for 8-512 circles, as the best embodiment, the value range is 200 ms-16000 ms, specific time can be set, and the optimal duty cycle of the sampling period is 50%.
As shown in fig. 3, the specific process of the pre-diagnosis step S2 is as follows: the data of the data acquisition equipment are stored, the stored original data can be directly called to the cloud, meanwhile, transformation and feature extraction processing are carried out on the data, a feature vector containing a fault is selected through a two-classifier (HSSVM) and uploaded to a cloud interface, the feature vector containing the fault is added into an HSSVM parameter base again after the reliability is confirmed through an accurate diagnosis step to carry out HSSVM learning training so as to optimize the two-classifier, and the two-classifier (HSSVM) carries out sampling control on each sensor of the data acquisition equipment according to the fault judgment condition.
Further, in the pre-diagnosis step S2, the data transformation employs a one-dimensional eight-scale wavelet transformation, and a two-classifier of a support vector machine is employed, in this embodiment, a selection tree classifier is employed, and whether there is a fault in the acquired data is determined, and if it is determined that there is a fault, the fault data is transmitted to the monitoring and diagnosis center, and the next step is performed for analysis.
Further, in the pre-diagnosis step S2, the method for data transformation and feature extraction includes:
(1) data transformation: the data series output by one sensor in Ttwork time period is recorded as data [ k]K is 0,1, …, if data [ k [ ]]From the vibration sensor, the one-dimensional multi-scale wavelet transform is carried out to extract singular points at each scale, and the transform result is recorded as wal [ k ]](s)Wherein s is a scale number, s ═ 1,2, … N; the invention takes 8 scales, namely N is 8;
(2) feature extraction: in a sensor group, there are M vibration sensors, and correspondingly, there are
Assuming that the time taken for the shaft to rotate 1 turn (i.e., the rotation period) is Tc, the position of each Tc is obtained from the output data of the rotary encoder, and an attempt is made to search for a Tc period
N × M singular points p existing onsmIf a singular point does not exist, the singular point is set to be 0, otherwise, the singular point is set to be a modulus maximum value in the period;
setting a sensor group with Q temperature sensors and 1 power sensor; defining a d-dimensional feature vector P:
P＝(〖psm|〗s＝1...N，m＝1...M，〖pq|〗q＝1…Q，pw)
d＝N×M+Q+1
wherein: p is a radical ofq|q＝1…QThe average value of sampling of each temperature sensor in the period in question is Q; p is a radical ofw1 being the average value of the power sensor samples over the period in question;
in the invention, M is 2-5, and Q is 1-3; that is, the dimension d of the feature vector is generally different for different components.
Further, in the pre-diagnosis step S2, a two-classifier based on HSSVM is adopted, each component in the rolling mill production line corresponds to one two-classifier, and the state of each component is divided into two types: giving out fault reliability R at the same time when there is or is no fault, wherein the given fault reliability R represents the severity of the fault and is used for controlling the size of the sampling pause period Twait; the Ttwork period contains 64 Tc cycles, each Tc extracts a characteristic vector Pj, j is 1,2, …, 64, and the Pj is sent to a classifier to judge whether a fault exists; if the number of times of faults is majority in 64 judgments in the Ttwork period, judging the component to be faulty; when a certain component is judged to be in fault, data selection is started, and the current 64 Pj are transmitted to the cloud for accurate diagnosis; at any time, the unit operation maintenance personnel can also transmit 64 Pj or original data series data [ k ] of the current Ttwork time period to the cloud for diagnosis.
In the invention, the two classifiers are realized by a Hyper-Sphere Support vector machine (HSSVM), and the feature vector P is a sample of the HSSVM.
Further, in the pre-diagnosis step S2, an HSSVM learning training system is adopted, but usually, a positive sample and a negative sample are required for training one SVM; the positive sample is a characteristic vector of the rolling mill component in normal operation and can be collected in a large quantity; the negative sample is a characteristic vector when the rolling mill component strip is in fault operation, and is difficult to collect; the method for solving the problem of the invention is as follows:
(1) because the pre-diagnosis is only a prophase diagnosis and only judges whether the fault exists or not, and the positive sample has good cohesiveness, the HSSVM is adopted to realize the two classifications; the SVM has strong generalization capability, and can control the risk of fault division into no fault to be very low;
(2) training of HSSVM: selecting a kernel function as a Gauss function, and establishing a characteristic space of the HSSVM; firstly, collecting a batch of positive training samples P + with good cohesiveness and storing the positive training samples P + in an HSSVM training sample library; then, through iteration, a minimum hypersphere capable of wrapping all P + is found, so that parameters such as the radius, the sphere center and the like of the hypersphere are determined, and the HSSVM with the initial diagnosis capability is established;
in practical application, a reward and punishment mechanism of reinforcement learning is applied, and a training sample library is optimized according to a current training result, so that the diagnostic performance of the system is continuously improved; adding the collected samples into a training sample library after accurate diagnosis is carried out on the collected samples to obtain reliable labels (including positive and negative samples), and retraining the HSSVM to determine a new hypersphere; if the new HSSVM has better classification performance, the added sample is valid, otherwise, the addition is cancelled, and the training is cancelled.
As shown in fig. 4, the accurate fault diagnosis is performed by a series of servers located in the cloud, and the functions of the server are as follows: and performing further accurate diagnosis on the pre-diagnosis result to obtain a fault type and a fault level, wherein the specific process of the accurate diagnosis step S3 is as follows: the method comprises the steps of firstly establishing a fault type sample library storing training samples, then establishing a fault diagnosis and classification model, receiving a characteristic vector containing a fault from pre-diagnosis, then comparing and analyzing data and the training samples in the database according to the stored training samples, determining the fault type according to the fault diagnosis and classification model in fault classification, updating an analysis result into the training samples in the fault type sample library, and optimizing through reinforcement learning.
In the step S3 of accurate diagnosis, a fault type sample library and a fault classification knowledge library in which training samples are stored are first established, then a confirmed diagnosis and classification model and a fuzzy evaluation model of faults are established, after a feature vector containing faults from pre-diagnosis is received, data and the training samples in the database are compared and analyzed according to the stored training samples, wherein fault classification determines the fault type according to the confirmed diagnosis and classification model of faults, fault classification classifies the severity of the faults according to the fuzzy evaluation model, meanwhile, the development trend of the faults is predicted according to the fuzzy state model, and the analysis results are updated into the training samples in the fault type sample library and the fault classification knowledge library for optimization through reinforcement learning.
In the step S3, the accurate diagnosis is used for data and command transmission between the cloud and the user side, and specifically includes: (1) receiving the feature vector uploaded by a user side for accurate diagnosis; feeding back an accurate diagnosis result; (2) when needed, the user terminal can be directly instructed to upload the specified original fault data segment as auxiliary and supplementary analysis data for diagnosis of a certain time; (3) and the system is interacted with a user side to realize the transmission of various control and state information.
In the accurate diagnosis step S3, the preliminarily diagnosed and preliminarily determined feature vectors with possible faults are further subjected to fault classification, and the fault types include bearing defects, bearing seat wear, bearing seat hole wear, shaft cracking, shaft deformation, gear tooth breakage and box body deformation; the main key parts of the high-speed wire rolling mill production line are as follows: the device comprises a horizontal transmission case, a vertical transmission case, a speed increasing case, a conical case, a reducing diameter speed increasing case, a conical case, a pinch roll and a laying head body; each component may experience a failure of the type listed in table 2.
TABLE 2 Fault types Table
| Failure point | Type of failure |
| Box body | Wear of bearing housing bore and deformation of housing |
| Shaft system | Shaft cracking, shaft deformation, abnormal idle shaft engagement |
| Gear wheel | Tooth breakage, tooth cracking |
| Bearing assembly | Damage to the inner and outer races or cages of the bearings, rolling element defects |
In the step S3 of accurately diagnosing, a fault classification method based on SOM is adopted:
because Self-Organizing mapping neural networks (SOMs) have large tolerance to sample distortion and noise, the SOMs are adopted to realize fault classification, and each component corresponds to one SOM;
the SOM has a structure as shown in fig. 5, in which the number of nodes in the input layer is d (the dimension of the feature vector P), and the number of nodes in the output layer is 16 × 16, which is the number of training samples.
In the step S3 of accurately diagnosing, the training process for the SOM includes:
(1) selecting an initial training sample: the initial training samples were derived from: a. the method comprises the following steps that a field expert analyzes and fits samples of various typical faults by using a kinetic theory and experience; b. setting a simulation fault and a collected sample on a rolling mill component; c. samples collected when an actual fault occurs;
256 samples were obtained from the above route, denoted as Pl, l ═ 1,2, …, 256; these samples should be representative of the fault types listed in table 1, with approximately the same number of samples for each fault type;
after the training is finished, the output node positions corresponding to the similar input samples are adjacent on the plane of an output layer, and one type of fault is distributed in a subarea;
(2) optimization of training samples: in practical application, when a true negative sample is received, the true negative sample is added into a training sample library and the SOM is retrained; optimizing the training sample library according to the current training result by applying a reward and punishment mechanism of reinforcement learning, and eliminating the inferior samples; thereby increasing the diagnostic performance of the system.
In the step S3 of precisely diagnosing, the fault classification is to divide the fault into 5 grades according to the severity; the level 1 fault is the lowest and the level 5 fault is the highest.
In the accurate diagnosis step S3, the initial data from the pre-diagnosis is also subjected to auxiliary diagnosis by an auxiliary analysis system, and the auxiliary diagnosis system provides an interface for intervention diagnosis for a domain expert; if necessary, the domain expert can do the following:
(1) evaluating the automatic diagnosis result to help a machine learning system to complete the optimization of a sample base or a knowledge base;
(2) the field expert can analyze the difficult and complicated faults, and the result can be used as expert knowledge to optimize the system;
(3) and the functions of a software analysis platform, a common professional measuring and analyzing instrument interface and the like are provided for field experts.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be construed as being within the scope of the present invention if they are within the spirit and principle of the present invention.
Claims (6)
1. A remote online monitoring and intelligent diagnosis method for a high-speed wire rod rolling mill comprises the following steps:
and S1 data acquisition step: the data collected by each data collecting device is collected to a central node and then transmitted to an industrial personal computer by adopting a local area network, and the data collecting device comprises a vibration sensor, a speed sensor, a three-dimensional acceleration sensor, a displacement sensor, a rotary encoder, a temperature sensor, a power meter and a power sensor in a rolling mill set;
s2 pre-diagnosis step: the industrial personal computer stores data, simultaneously carries out noise removal, data transformation and feature extraction processing on the data, simultaneously judges whether the processed data has a fault of the high-speed wire mill, and outputs a result, if the structure output by the industrial personal computer judges that the high-speed wire mill has the fault, the data is transmitted to the monitoring and diagnosis center, otherwise, the industrial personal computer does not transmit the data to the monitoring and diagnosis center, so as to avoid transmitting a large amount of fault-free data to the diagnosis center, thereby avoiding network congestion and increasing the burden of a monitoring and diagnosis center server;
s3 precision diagnosis step: the monitoring and diagnosing center further analyzes the pre-diagnosing result of the industrial personal computer after receiving the data with faults so as to determine the type and the position of the faults,
in the data acquisition step S1, the rotary encoder performs sampling according to the following rule: the sampling period is divided into two time periods Twork and Twoit, the sampling is carried out 16 times when the rotating shaft rotates for one circle, the Twork time period is the time spent on 64 circles of rotation of the rotating shaft, the sampling is carried out 1024 times in the Twoit time period, the sampling is not carried out in the Twoit time period, and the time spent on 8-512 circles of rotation of the rotating shaft is the time;
the process of the pre-diagnosis step S2 is: the data of the data acquisition equipment are stored, the stored original data can be directly called to the cloud end, meanwhile, transformation and feature extraction processing are carried out on the data, a feature vector containing a fault is selected through a two-classifier of the HSSVM and is uploaded to a cloud interface, the feature vector containing the fault is added into an HSSVM parameter base again after the reliability is confirmed through an accurate diagnosis step to carry out HSSVM learning training so as to optimize the two-classifier, and the two-classifier of the HSSVM carries out sampling control on each sensor of the data acquisition equipment according to the fault judgment condition;
in the pre-diagnosis step S2, the data transformation and feature extraction method includes:
(1) data transformation: the data series output by one sensor in Ttwork time period is recorded as data [ k]K is 0, 1.., if data [ k ]]From the vibration sensor, the one-dimensional multi-scale wavelet transform is carried out to extract singular points under each scale, and the transform result is recordedIs wal [ k ]](s)Wherein s is a scale number, s ═ 1,2, … N; the diagnostic method takes 8 scales, namely N is 8;
(2) feature extraction: in a sensor group, there are M vibration sensors, and correspondingly wal [ k ]]m (s)M is 1, 2.. M, where Tc is the time taken for the rotating shaft to rotate 1 turn (i.e., the rotation period), the location of each Tc is obtained from the output data of the rotary encoder, and wal [ k ] is searched in an attempt within one Tc period]m (s)N × M singular points P existing abovesmIf a singular point does not exist, the singular point is set to 0, otherwise the singular point is set to wal [ k ] in the period]m (s)A modulus maximum of;
setting a sensor group with Q temperature sensors and 1 power sensor; defining a d-dimensional feature vector P:
d＝N×M+Q+1
wherein: m is 2-5; pq|q＝1...QThe average value of sampling of each temperature sensor in a rotation period is Q, and Q is 1-3; p is a radical ofwIs the average of the power sensor samples over the rotation period, 1.
2. The high-speed wire rolling mill remote on-line monitoring and intelligent diagnosis method according to claim 1, characterized in that: in the pre-diagnosis step S2, a two-classifier based on HSSVM is adopted, each component in the rolling mill production line corresponds to one two-classifier, and the state of each component is divided into two types: giving out fault reliability R at the same time when there is or is no fault, wherein the given fault reliability R represents the severity of the fault and is used for controlling the size of the sampling pause period Twait; the Ttwork period contains 64 Tc cycles, each Tc extracts a characteristic vector Pj, j is 1,2, …, 64, and the Pj is sent to a classifier to judge whether a fault exists; if the number of times of faults is majority in 64 judgments in the Ttwork period, judging the component to be faulty; when a certain component is judged to be in fault, data selection is started, and the current 64 Pj are transmitted to the cloud for accurate diagnosis; at any time, the unit operation maintenance personnel can also transmit 64 Pj or original data series data [ k ] of the current Ttwork time period to the cloud for diagnosis.
3. The high-speed wire mill remote online monitoring and intelligent diagnosis method according to claim 1 or 2, characterized in that: in the pre-diagnosis step S2, the process of the HSSVM learning training includes: selecting a kernel function as a Gauss function, and establishing a characteristic space of the HSSVM; firstly, collecting a batch of positive training samples P + with good cohesiveness and storing the positive training samples P + in an HSSVM training sample library; then, through iteration, a minimum hypersphere capable of wrapping all P + is found, so that the parameters of the radius and the center of the hypersphere are determined, and the HSSVM with the initial diagnosis capability is established; in practical application, a reward and punishment mechanism of reinforcement learning is applied, and a training sample library is optimized according to a current training result, so that the diagnostic performance of the system is continuously improved; adding the collected sample into a training sample library after accurate diagnosis is carried out on the collected sample to obtain a reliable label, and retraining the HSSVM to determine a new hypersphere; if the new HSSVM has better classification performance, the added sample is valid, otherwise, the addition is cancelled, and the training is cancelled.
4. The high-speed wire rolling mill remote on-line monitoring and intelligent diagnosis method according to claim 1, characterized in that: the procedure of the precise diagnosis step S3 is: the method comprises the steps of firstly establishing a fault type sample library storing training samples, then establishing a fault diagnosis and classification model, receiving a characteristic vector containing a fault from pre-diagnosis, then comparing and analyzing data and the training samples in the database according to the stored training samples, determining the fault type according to the fault diagnosis and classification model in fault classification, updating an analysis result into the training samples in the fault type sample library, and optimizing through reinforcement learning.
5. The high-speed wire rolling mill remote on-line monitoring and intelligent diagnosis method according to claim 1, characterized in that: the procedure of the precise diagnosis step S3 is: the method comprises the steps of firstly establishing a fault type sample base and a fault grading knowledge base which store training samples, then establishing a fault diagnosis and classification model and a fault fuzzy evaluation model, receiving a characteristic vector containing a fault from pre-diagnosis, comparing and analyzing data and the training samples in the database according to the stored training samples, determining the fault type according to the fault diagnosis and classification model for fault classification, grading the severity of the fault according to the fuzzy evaluation model for fault grading, predicting the development trend of the fault according to the fuzzy state model, and updating the analysis result into the training samples of the fault type sample base and the fault grading knowledge base to optimize through reinforcement learning.
6. The remote online monitoring and intelligent diagnosis method for the high-speed wire rod rolling mill according to claim 4 or 5, wherein in the accurate diagnosis step S3, SOM is adopted to realize fault classification, each component corresponds to an SOM, the node number of an input layer of an SOM structure is d, the node number of an output layer is the number of training samples, and is 16 × 16, and the training process for the SOM is as follows:
(1) selecting an initial training sample: the initial training samples were derived from: a. the method comprises the following steps that a field expert analyzes and fits samples of various typical faults by using a kinetic theory and experience; b. setting a simulation fault and a collected sample on a rolling mill component; c. samples collected when an actual fault occurs; 256 samples, denoted P, were obtained from the above routel1,2, …, 256; for P againlAfter the training is finished, the positions of output nodes corresponding to similar input samples are adjacent on the plane of an output layer, and faults of one type are distributed in a subarea;
(2) optimization of training samples: in practical application, when a true negative sample is received, the true negative sample is added into a training sample library and the SOM is retrained; optimizing the training sample library according to the current training result by applying a reward and punishment mechanism of reinforcement learning, and eliminating the inferior samples; thereby increasing the diagnostic performance of the system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710307331.3A CN108126987B (en) | 2017-05-04 | 2017-05-04 | Remote online monitoring and intelligent diagnosis method for high-speed wire rolling mill |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710307331.3A CN108126987B (en) | 2017-05-04 | 2017-05-04 | Remote online monitoring and intelligent diagnosis method for high-speed wire rolling mill |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108126987A CN108126987A (en) | 2018-06-08 |
| CN108126987B true CN108126987B (en) | 2020-07-28 |
Family
ID=62388440
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710307331.3A Active CN108126987B (en) | 2017-05-04 | 2017-05-04 | Remote online monitoring and intelligent diagnosis method for high-speed wire rolling mill |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108126987B (en) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108667927A (en) * | 2018-05-07 | 2018-10-16 | 山推工程机械股份有限公司 | Method for remote management, device, system and the storage medium of engineering machinery failure |
| CN108994090B (en) * | 2018-07-06 | 2020-03-27 | 南京钢铁股份有限公司 | Method for diagnosing faults of rolling mill pressing system |
| CN109143851B (en) * | 2018-07-11 | 2021-06-01 | 佛山科学技术学院 | Method for recognizing multi-mark fault deep learning and intelligently expressing result thereof |
| CN109002672A (en) * | 2018-09-30 | 2018-12-14 | 珠海市运泰利自动化设备有限公司 | A kind of design method of the intelligent producing line based on optic test |
| CN109683565A (en) * | 2018-12-12 | 2019-04-26 | 电子科技大学 | A kind of instrument and meter fault detection method based on multi-method fusion |
| CN109598309B (en) * | 2018-12-14 | 2022-02-15 | 广东寰球智能科技有限公司 | Detection system and monitoring method of metal packaging punching machine |
| CN109894476B (en) * | 2019-02-28 | 2020-06-26 | 北京首钢股份有限公司 | Fault diagnosis method and device for hydraulic equipment of cold-rolled silicon steel production line |
| CN109909306B (en) * | 2019-03-11 | 2020-08-21 | 北京首钢自动化信息技术有限公司 | Monitoring method for on-line monitoring and intelligent degradation trend early warning of acid rolling equipment |
| CN112857797B (en) * | 2019-11-27 | 2023-08-08 | 睿捷国际股份有限公司 | Torsional vibration detection system and method for rolling mill equipment |
| CN110825059A (en) * | 2019-11-29 | 2020-02-21 | 浙江工业大学 | Intelligent monitoring system and monitoring method for elevator wire rod production based on MQTT intelligent gateway |
| CN111122155B (en) * | 2019-12-31 | 2021-10-12 | 湖南大学 | Gear fault diagnosis method based on telescopic shifting super-disc |
| CN111438184B (en) * | 2020-03-20 | 2024-05-17 | 中冶赛迪工程技术股份有限公司 | Accident handling system and method between high-speed wire rod module units |
| CN111652227B (en) * | 2020-05-21 | 2021-01-22 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting damage fault of bottom floor of railway wagon |
| CN111922095A (en) * | 2020-07-14 | 2020-11-13 | 上海数深智能科技有限公司 | Vibration diagnosis method for abnormal torsional vibration fault of roller of cold rolling mill |
| CN113280910A (en) * | 2021-04-27 | 2021-08-20 | 圣名科技(广州)有限责任公司 | Real-time monitoring method and system for long product production line equipment |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100900676B1 (en) * | 2002-11-14 | 2009-06-01 | 주식회사 포스코 | Apparatus for diagnosing fault sucsessive vernier in mill |
| CN102059256A (en) * | 2010-11-26 | 2011-05-18 | 首钢总公司 | Equipment operating and process quality state monitoring system for high-speed rolling mill |
| CN202951709U (en) * | 2012-09-21 | 2013-05-29 | 浙江联鑫板材科技有限公司 | Continuous rolling system state remote monitoring device |
| CN203343182U (en) * | 2013-07-13 | 2013-12-18 | 吉林大学 | Magnesium alloy sheet rolling parameter monitoring and fault diagnosis system |
| CN203380190U (en) * | 2013-05-08 | 2014-01-08 | 浙江富春江环保热电股份有限公司 | Device for monitoring operating state of rolling mill |
| CN103521531A (en) * | 2013-11-07 | 2014-01-22 | 天津理工大学 | Fault diagnosis and feedback system according to third octave flutter of high-speed cold rolling mill |
| CN106292591A (en) * | 2016-08-20 | 2017-01-04 | 广州明珞汽车装备有限公司 | A kind of equipment big data platform monitoring diagnosis system |
-
2017
- 2017-05-04 CN CN201710307331.3A patent/CN108126987B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100900676B1 (en) * | 2002-11-14 | 2009-06-01 | 주식회사 포스코 | Apparatus for diagnosing fault sucsessive vernier in mill |
| CN102059256A (en) * | 2010-11-26 | 2011-05-18 | 首钢总公司 | Equipment operating and process quality state monitoring system for high-speed rolling mill |
| CN202951709U (en) * | 2012-09-21 | 2013-05-29 | 浙江联鑫板材科技有限公司 | Continuous rolling system state remote monitoring device |
| CN203380190U (en) * | 2013-05-08 | 2014-01-08 | 浙江富春江环保热电股份有限公司 | Device for monitoring operating state of rolling mill |
| CN203343182U (en) * | 2013-07-13 | 2013-12-18 | 吉林大学 | Magnesium alloy sheet rolling parameter monitoring and fault diagnosis system |
| CN103521531A (en) * | 2013-11-07 | 2014-01-22 | 天津理工大学 | Fault diagnosis and feedback system according to third octave flutter of high-speed cold rolling mill |
| CN106292591A (en) * | 2016-08-20 | 2017-01-04 | 广州明珞汽车装备有限公司 | A kind of equipment big data platform monitoring diagnosis system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108126987A (en) | 2018-06-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108126987B (en) | Remote online monitoring and intelligent diagnosis method for high-speed wire rolling mill | |
| CN109822399B (en) | Numerical control machine tool cutter wear state prediction method based on parallel deep neural network | |
| CN110554657B (en) | Health diagnosis system and diagnosis method for operation state of numerical control machine tool | |
| CN111504676B (en) | Equipment fault diagnosis method, device and system based on multi-source monitoring data fusion | |
| CN107247230B (en) | Rotary motor state monitoring method based on support vector machine and data driving | |
| CN111637964B (en) | Rail corrugation identification method | |
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| JP2022524244A (en) | Predictive classification of future behavior | |
| CN111046945A (en) | Fault type and damage degree diagnosis method based on combined convolutional neural network | |
| CN102054179A (en) | Online state monitoring and fault diagnosis device and method for rotary machine | |
| JPH06170696A (en) | System and method for using real time expert system for diagnosing tool life and predicted tool wear | |
| CN112633339A (en) | Bearing fault intelligent diagnosis method, bearing fault intelligent diagnosis system, computer equipment and medium | |
| WO2011073613A1 (en) | A method, apparatus and computer program for diagnosing a mode of operation of a machine | |
| CN116629627A (en) | Intelligent detection system of power transmission on-line monitoring device | |
| CN112692646A (en) | Intelligent assessment method and device for tool wear state | |
| CN115438726A (en) | Device life and fault type prediction method and system based on digital twin technology | |
| CN113762344A (en) | Fault identification method of machine tool spindle and fault identification model training method and device | |
| CN117786373B (en) | Equipment operation diagnosis system based on big data corrugated paper processing | |
| CN112766200A (en) | Planetary gear box small sample fault diagnosis method based on one-dimensional convolutional neural network | |
| CN115655717A (en) | Bearing fault diagnosis method based on depth domain generalization network | |
| CN111707458A (en) | Rotor monitoring method based on deep learning signal reconstruction | |
| CN112763215B (en) | Multi-working-condition online fault diagnosis method based on modular federal deep learning | |
| CN117633688A (en) | Large-scale power data anomaly detection method based on ridge regression-k-means clustering-LOF-LSTM fusion algorithm | |
| JP7451854B2 (en) | Methods, systems and computer program products for diagnosing energy consumption in industrial environments | |
| CN117009841A (en) | Model training method, motor fault diagnosis method and microcontroller |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |