CN113515158A - Equipment state monitoring method based on probability hybrid finite state machine - Google Patents
Equipment state monitoring method based on probability hybrid finite state machine Download PDFInfo
- Publication number
- CN113515158A CN113515158A CN202111071624.9A CN202111071624A CN113515158A CN 113515158 A CN113515158 A CN 113515158A CN 202111071624 A CN202111071624 A CN 202111071624A CN 113515158 A CN113515158 A CN 113515158A
- Authority
- CN
- China
- Prior art keywords
- state
- data
- variable data
- initial
- discrete
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 89
- 238000012544 monitoring process Methods 0.000 title claims abstract description 37
- 230000008569 process Effects 0.000 claims abstract description 64
- 238000012549 training Methods 0.000 claims abstract description 11
- 230000007704 transition Effects 0.000 claims description 73
- 230000008859 change Effects 0.000 claims description 36
- 238000006243 chemical reaction Methods 0.000 claims description 35
- 239000013598 vector Substances 0.000 claims description 32
- 230000006870 function Effects 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 5
- 238000012986 modification Methods 0.000 claims description 4
- 230000004048 modification Effects 0.000 claims description 4
- 238000004886 process control Methods 0.000 claims description 4
- 230000001960 triggered effect Effects 0.000 claims description 3
- 238000012806 monitoring device Methods 0.000 claims 1
- 239000007788 liquid Substances 0.000 description 11
- 239000000047 product Substances 0.000 description 6
- 239000002994 raw material Substances 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 230000001276 controlling effect Effects 0.000 description 2
- 238000007599 discharging Methods 0.000 description 2
- 238000010438 heat treatment Methods 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000012467 final product Substances 0.000 description 1
- 239000007789 gas Substances 0.000 description 1
- 238000011031 large-scale manufacturing process Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 239000000123 paper Substances 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 238000003756 stirring Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D27/00—Simultaneous control of variables covered by two or more of main groups G05D1/00 - G05D25/00
- G05D27/02—Simultaneous control of variables covered by two or more of main groups G05D1/00 - G05D25/00 characterised by the use of electric means
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a device state monitoring method based on a probability hybrid finite state machine, which comprises the steps of acquiring flow data of a plurality of process periods when a device normally operates to obtain a data set X; dividing the data set X into a discrete variable data group and a continuous variable data group, and training a finite state machine by using the discrete variable data group to obtain a working state set Pn of the equipment; calculating the function relation of continuous variable data groups in each state in the state set Pn continuously changing along with time; and then monitoring the real-time working state of the equipment by using the state set Pn and the functional relation. By utilizing the method and the device, the fault or error in the equipment flow operation can be found in time, and the accuracy and the efficiency of the equipment work are improved.
Description
Technical Field
The invention relates to the technical field of equipment state monitoring, in particular to an equipment state monitoring method based on a probability hybrid finite state machine.
Background
The process industry mainly adds value to raw materials by physical or chemical method operations such as mixing, separating, crushing, heating and the like. It includes chemical, paper, steel, food, pharmaceutical and other industries, and the final product types are solid, liquid, gas and various forms of energy, and are usually produced in batch or continuous mode. Most of the equipment used in the process industry is large-scale production equipment, and variables such as temperature, flow, pressure, concentration and the like in the reaction equipment must be accurately regulated and controlled in different stages of product preparation so as to meet the requirements of the production process and ensure that the quality of the product meets corresponding standards.
With the continuous increase of the complexity of the product process, a plurality of preparation processes are required to be completed in each large-scale device. Because the process parameters of each flow are different, signals such as temperature, flow, pressure and the like in the same equipment have larger difference in different time periods and need to be monitored.
At present, there are two main methods for monitoring the state of equipment: one is to manually configure the process flow in advance, set the maximum and minimum (boundary values) of each sensor signal to be monitored according to the experience of the process engineer aiming at different stages of the flow, and judge whether the equipment runs abnormally by detecting whether the current sensor signal is out of range in real time. This approach often requires a large amount of manpower and material resources, and if the process parameters need to be adjusted subsequently, all configurations must be manually modified accordingly. In addition, the error probability of manual configuration is high, and the accuracy of abnormal diagnosis and prediction can be greatly reduced by a small amount of deviation. Therefore, the monitoring precision of the whole process flow cannot meet the requirement, and the product quality has flaws. And the other method is that the process flow information is not considered, only the key sensor signals are subjected to a time continuous signal modeling method, and a unified model which has a high probability of meeting the requirements of different process stages is tried to be excavated for anomaly monitoring and early warning. Because the parameters of the same signal in different process flows are generally different, the use of a unified model can generate excessive false reports and false negative reports. In addition, a globally unified data model is not very effective in diagnosing and predicting process state transition junctions.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the method aims to solve the technical problem that in the prior art, the monitoring precision of a large-scale equipment state monitoring method in the process industry is low, so that the product is flawed. The invention provides a device state monitoring method based on a probability hybrid finite state machine, which comprises the steps of acquiring flow data of a plurality of process periods when a device normally operates to obtain a data set X; dividing the data set X into a discrete variable data group and a continuous variable data group, and training a finite state machine by using the discrete variable data group to obtain a working state set Pn of the equipment; calculating the function relation of continuous variable data groups in each state in the state set Pn continuously changing along with time; and then, the real-time working state of the equipment is monitored by utilizing the state set Pn and the functional relation, so that faults or errors can be found in time, and the working accuracy and efficiency of the equipment are improved.
The technical scheme adopted by the invention for solving the technical problems is as follows: a device state monitoring method based on a probability hybrid finite state machine comprises the following steps:
s1: the method comprises the steps of circularly acquiring process data related to process control in a plurality of process periods when equipment A normally operates, wherein the process data in each process period comprise sensor data required for controlling the process flow when the equipment A works and switching value data in a PLC (programmable logic controller), and forming a data set X by the process data in all the process periods according to a time stamp sequence.
S2: dividing data in the data set X into a discrete variable data group Xd and a continuous variable data group Xc, training a finite state machine by using the discrete variable data group Xd, and learning a plurality of different states corresponding to different stages of the equipment A in normal operation from the discrete variable data group Xd, wherein the different states form a state set Pn, and each state P in the state set Pn has a corresponding starting time t1 and an ending time t 2; the discrete variable data and the continuous variable data acquired at the same time have the same timestamp; the combination of the current values of the discrete variables forms the current state P of the device A, and according to the timestamp information, a plurality of continuous variable data in a continuous variable data group Xc are divided into a plurality of data packets Z, each data packet Z corresponds to each state P in the state set Pn one by one, and the data in each data packet Z is acquired in the continuous time period of the corresponding state P.
S3: and calculating the function relation of continuous variable data in the data packet Z corresponding to each state P in the state set Pn continuously changing along with time.
S4: monitoring a working device A, collecting flow data of the device in real time, and assuming that discrete variable data collected under a certain time stamp are Xdn and continuous variable data are Xcn; current discrete variable data Xdn constitutes a current state vector Pn, matching Pn with state P in state set Pn when device a is operating normally,
if pn cannot find a perfect match, P, the feedback device a operates abnormally,
if pn can find a matching P and the change over time at Xcn meets the functional relationship in step S3, then device a is operating normally, continues to monitor device a,
if pn can find a matching P, but the change over time at Xcn does not satisfy the functional relationship in step S3, the feedback device a operates abnormally.
Further, the functional relationship in step S3 includes a unitary functional relationship and a multivariate functional relationship, and when the continuous variable data group Xc only includes the change data of one continuous variable, the calculated functional relationship is a unitary functional relationship; when the continuous variable data group Xc includes the variable data of a plurality of continuous variables, the calculated functional relationship is a multivariate functional relationship, and the multivariate functional relationship represents the correlation relationship between the variable data of the plurality of continuous variables.
Further, when the finite state machine is trained by using the discrete variable data set Xd, a transition condition set Tn between any two states that can be transitioned between each other in the state set Pn can also be obtained, where each transition condition T includes a trigger condition of state transition, a probability of state transition, and a time interval during which a previous state stays before transition.
Further, the discrete variable data refers to variable data whose value is a natural number or an integer, and the discrete variable data group Xd contains change data of at least one discrete variable; the continuous variable data refers to variable data with any value in a defined interval, and the continuous variable data group Xc contains change data of at least one continuous variable. For example, the discrete variable data may be 0,1,3, etc., and the continuous variable data may be data that continuously changes with a time stamp.
Further, the process of training the finite-state machine by using the discrete variable data set Xd in step S2 specifically includes:
s21: forming a signal vector V by all discrete variables in a discrete variable data set Xd, initializing a state set and a conversion condition set in a finite state machine to obtain an initial state set P ' and an initial conversion condition set T ', setting the state formed by the current signal vector V as an initial state P0, and storing the initial state P0 in the initial state set P ';
s22: monitoring whether the value in the signal vector V changes every other sampling period, wherein the change of the median value in the signal vector V is the trigger condition of state conversion;
if the value in the signal vector V changes, first checking whether the trigger condition of the state transition caused by the current change tc is already recorded in the initial transition condition set T ', and if the trigger condition of the state transition caused by the change tc is not recorded in the initial transition condition set T ', writing the trigger condition of the state transition caused by the change tc into the initial transition condition set T '; if the record exists in the initial conversion condition set T', the record is not modified;
then checking whether the next state st triggered by the change tc is recorded in the initial state set P ', and if the initial state set P ' is not recorded, writing the state st into the initial state set P '; if the record exists in the initial state set P', no modification is made;
calculating the time interval of the stay of the last state before the tc trigger state is converted according to the timestamp information;
s23: if the value in the signal vector V has not changed, repeating step S22 until all the data in the discrete variable data group Xd are used up, writing all the states st into the initial state set P' to obtain a state set Pn in which the device a normally operates; counting the total times of converting the same state into other states, and calculating the probability of converting the same state into other states, namely the probability of state conversion; and writing all the trigger conditions for all the state transitions, the probability of the state transitions and the time interval of the last state stay before the state transitions into an initial transition condition set T' to obtain a transition condition set Tn.
Further, the step S4 further includes:
monitoring a working device A, collecting flow data of the device in real time, and assuming that discrete variable data collected under a certain time stamp are Xdn and continuous variable data are Xcn; current discrete variable data Xdn form a current state vector Pn, Pn is matched with a state P in a state set Pn when the device a normally works, and a transition condition Tn caused by the last state transition to the state Pn is matched with a transition condition T in a transition condition set Tn, if Pn cannot find a completely matched P or Tn cannot find a completely matched T, the device a is fed back to be abnormally operated, if Pn can find a completely matched P and Tn can find a completely matched T, and Xcn changes with time to satisfy the functional relationship in step S3, the device a works normally, monitoring of the device a is continued, if Pn can find a completely matched P and Tn can find a completely matched T, but Xcn changes with time do not satisfy the functional relationship in step S3, the device a is fed back to be abnormally operated.
Compared with the existing solution, the equipment state monitoring method based on the probability hybrid finite-state machine fully considers the parameter details of the process flow control, can better accurately monitor the process flow change, and realizes higher monitoring and early warning accuracy. In addition, the monitoring method of the invention does not need an engineer to manually set a warning value, can automatically carry out monitoring and early warning, and can automatically update corresponding parameter configuration when an abnormal condition is found, thereby improving the monitoring efficiency and precision.
Drawings
The invention is further illustrated with reference to the following figures and examples.
Fig. 1 is a flow chart of the device state monitoring method based on the probability hybrid finite state machine of the present invention.
FIG. 2 is a flow chart of the present invention for training a finite state machine with a discrete variable data set.
FIG. 3 is a probabilistic mixing finite state machine model of an example reaction vessel according to the present invention.
Detailed Description
The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic views illustrating only the basic structure of the present invention in a schematic manner, and thus show only the constitution related to the present invention.
Finite state machines are a computing model abstracted for studying finite memory computing processes and certain language classes. The finite state machine has a finite number of states, each of which can be migrated to zero or more states, and the input string determines which state migration is to be performed. The finite state automaton can be represented as a directed graph. The finite-state machine in the prior art can only process discrete variable data, and the probability mixing finite-state machine of the invention can simultaneously process the discrete variable data and the continuous variable data.
As shown in fig. 1, a device status monitoring method based on a probabilistic hybrid finite state machine includes the following steps:
s1: the method comprises the steps of circularly acquiring process data related to process control in a plurality of process periods when the equipment A normally operates, wherein the process data in each process period comprise sensor data (comprising continuous variable data and/or discrete variable data) required for controlling a process flow when the equipment A works and switching value data (being discrete variable data) in a PLC (programmable logic controller), and forming a data set X by the process data in all the process periods according to a time stamp sequence.
It should be noted that the process data can be obtained from the PLC of the device a or obtained through a communication middleware (e.g., OPC). The data acquisition is carried out when the equipment A produces qualified products according to the requirements of process standards, so that the acquired data can reflect the process flow of the standard more accurately. The process data includes sensor data and PLC switch control quantity data related to process flow control in the device a, and may be selected according to specific process flow control requirements. In this embodiment, 35 qualified production periods or production lot flow data during normal operation of the device a may be collected in a loop, and the data amount may satisfy training use of the finite state machine. And (4) forming a data set X by the process data in all the process periods according to the time stamp sequence.
S2: dividing data in a data set X into a discrete variable data group Xd and a continuous variable data group Xc, training a finite state machine by using the discrete variable data group Xd, and learning a plurality of different states corresponding to different stages of equipment A in normal operation from the discrete variable data group Xd, wherein the different states form a state set Pn, and each state P in the state set Pn has corresponding starting time t1 and ending time t 2; the discrete variable data and the continuous variable data acquired at the same time have the same timestamp; the combination of the current values of the discrete variables forms the current state P of the device A, and according to the timestamp information, a plurality of continuous variable data in a continuous variable data group Xc are divided into a plurality of data packets Z, each data packet Z corresponds to each state P in the state set Pn one by one, and the data in each data packet Z is acquired in the continuous time period of the corresponding state P.
It should be noted that the discrete variable data refers to variable data whose value is a natural number or an integer, and the discrete variable data group Xd includes change data of at least one discrete variable, such as switching value data, binary signal data, start/stop signal data of the motor, limit switch process data, and the like. The continuous variable data refers to variable data of any value in a defined interval, the continuous variable data group Xc includes change data of at least one continuous variable, such as temperature data, pressure data, flow data, and the like, and the continuous data may be data that continuously changes with time.
Whether the working state of the equipment changes can be known from the discrete variable data, so that the finite-state machine is trained by utilizing the discrete variable data set. After the training is completed, a plurality of different states corresponding to different stages of the device a during normal operation can be obtained, the plurality of different states form a state set Pn, the state set Pn includes a plurality of different states P, each state P in the state set Pn has a corresponding start time t1 and end time t2, that is, each state lasts for a period of time. For example, the discrete variable data set Xd includes 4 discrete variables v1, v2, v3 and v4, where the 4 discrete variables may constitute a signal vector (v 1, v2, v3, v 4), values of the variables v1, v2, v3 and v4 may all be changed, different values of the 4 variables may form different states P, and each state P may exist continuously for a certain period of time (may be regarded as a process sub-flow). According to the timestamp information, the data in the continuous variable data group Xc may be divided into a plurality of data packets Z, each data packet Z may contain one or more continuous variable data, and the data in each data packet Z is collected during a time period during which its corresponding state P lasts. For example, the data packet Z may contain a plurality of data of temperature variation over a period of time, or the data packet Z may contain a plurality of data of temperature variation over a period of time and a plurality of data of pressure variation over a period of time. Therefore, the state P and the packet Z can be matched one by one according to the time stamp information.
When the finite-state machine is trained by using the discrete variable data set Xd, a transition condition set Tn between any two states that can be transitioned between each other in the state set Pn can also be obtained, and each transition condition T includes a trigger condition of state transition, a probability of state transition occurrence, and a time interval during which the last state stays before transition.
As shown in fig. 2, the process of training the finite-state machine by using the discrete variable data set Xd specifically includes:
s21: forming a signal vector V by all discrete variables in the discrete variable data set Xd, initializing a state set and a conversion condition set in a finite state machine to obtain an initial state set P ' and an initial conversion condition set T ', setting the state formed by the current signal vector V as an initial state P0, and storing the initial state P0 in the initial state set P '.
S22: monitoring whether the value in the signal vector V changes every other sampling period (for example, 1ms or 1 s), wherein the change of the median value in the signal vector V is a trigger condition of state transition; if the value in the signal vector V changes, first checking whether the trigger condition of the state transition caused by the current change tc is already recorded in the initial transition condition set T ', and if the trigger condition of the state transition caused by the change tc is not recorded in the initial transition condition set T ', writing the trigger condition of the state transition caused by the change tc into the initial transition condition set T '; if the record exists in the initial conversion condition set T', the record is not modified; then checking whether the next state st triggered by the change tc is recorded in the initial state set P ', and if the initial state set P ' is not recorded, writing the state st into the initial state set P '; if the record exists in the initial state set P', no modification is made; and calculating the time interval of the last state staying before the change of the value in the signal vector V according to the timestamp information.
S23: if the value in the signal vector V has not changed, repeating step S22 until all the data in the discrete variable data group Xd are used up, writing all the states st into the initial state set P' to obtain a state set Pn in which the device a normally works; counting the total times of converting the same state into other states, and calculating the probability of converting the same state into other states, namely the probability of state conversion; and writing all the trigger conditions for all the state transitions, the probability of the state transitions and the time interval of the last state stay before the state transitions into an initial transition condition set T' to obtain a transition condition set Tn.
S3: and calculating the function relation of continuous variable data in the data packet Z corresponding to each state P in the state set Pn continuously changing along with time.
It should be noted that the functional relationship includes a univariate functional relationship and a multivariate functional relationship. When the continuous variable data group Xc only contains the variation data of one continuous variable, the calculated functional relationship is a unitary functional relationship. When the continuous variable data group Xc includes variation data of a plurality of continuous variables, the calculated functional relationship is a multivariate functional relationship, and the multivariate functional relationship represents a correlation relationship between the plurality of continuous variables. For example, when the continuous variable data group Xc only includes a plurality of data of temperature variable changes with time, which are data of temperature variable changes in different process cycles when the apparatus a is operating normally (for example, temperature change data of 35 process cycles are collected). In order to monitor temperature variable data more accurately and conveniently during real-time monitoring, data in Xc is required to correspond to a specific state P in Xc and Pn according to a discrete variable value acquired by Xc in the same time period, so as to obtain a plurality of subsets x1, x 2. Fitting is performed on a plurality of temperature change data in the same state P to obtain a functional relationship between the temperature variable data and time, wherein the functional relationship is a unitary functional relationship, for example, x = at + b, x is the temperature variable data, and t is the time. For example, when three continuous variable data of temperature, pressure and flow are included in the continuous variable data group Xc, several subsets x1, x2,.. gtoreq, xn; f1, f 2.. fn; k1, k 2. Linear regression can be used to linearly fit the three variable data for multiple temperature, pressure and flow data in the same state P, respectively, resulting in a temperature variable data function x = a1t + b1, a pressure variable data function f = a1t + b1 and a flow variable data function k = a3t + b 3. Other statistical or data mining algorithms (e.g., a proximity algorithm, a support vector regression algorithm) can also be used for fitting to obtain a multivariate functional relationship, which represents the correlation relationship of the three variable data of temperature, pressure and flow when the equipment a works normally.
Through steps S1-S3, the state set Pn, the conversion condition set Tn, and a plurality of functional relationships (which may be referred to as a function set Ω) of the device a during normal operation can be obtained, a combination of the state set Pn, the conversion condition set Tn, and the function set Ω is referred to as a probability mixed finite state set model, and the probability mixed finite state set model is stored in the local computer. The probability mixed finite state set model can be directly read later to carry out real-time monitoring on the working state of other equipment A.
S4: monitoring a working device A, collecting flow data of the device in real time, and assuming that discrete variable data collected under a certain time stamp are Xdn and continuous variable data are Xcn; current discrete variable data Xdn form a current state vector Pn, and Pn is matched with a state P in a state set Pn when the device A works normally; if the pn can not find the completely matched P, the feedback equipment A works abnormally; if the pn can find the matched P and the change of Xcn along with time meets the functional relation in the step S3, the operation of the equipment A is normal, and the equipment A is continuously monitored; if pn can find a matching P, but the change over time at Xcn does not satisfy the functional relationship in step S3, the feedback device a operates abnormally.
Step S4 may further include: monitoring a working device A, collecting flow data of the device in real time, and assuming that discrete variable data collected under a certain time stamp are Xdn and continuous variable data are Xcn; current discrete variable data Xdn constitutes a current state vector Pn, matching Pn with state P in state set Pn when device a is operating normally, and matching transition condition Tn caused by the transition from the previous state to state Pn with transition condition T in transition condition set Tn; if pn cannot find completely matched P or tn cannot find completely matched T, the feedback device A works abnormally, if pn can find completely matched P and tn can find completely matched T, and Xcn meets the functional relationship in the step S3 along with the change of time, the device A works normally, the monitoring of the device A is continued, if pn can find completely matched P and tn can find completely matched T, but Xcn does not meet the functional relationship in the step S3 along with the change of time, the feedback device A works abnormally.
When tn is matched with T, if any one of the trigger condition of the state transition in tn and the time interval of the last state stay before the transition is not matched with T, the device A is considered to be abnormal in operation.
The specific process for obtaining the probabilistic hybrid finite state machine is described below by taking the process flow of the reaction kettle as an example.
For example, the reaction vessel includes two feed valves, one discharge valve, two level switches, and one temperature sensor. The process carried out in this reactor is as follows: and stopping adding the liquid raw material A into the reaction kettle until the liquid level is one. Then adding the liquid raw material B into the reaction kettle to a liquid level II and stopping. And continuously heating the reaction kettle in the process of adding the raw materials until the temperature of the mixed liquid in the reaction kettle reaches 35 ℃, and keeping the temperature constant. And (3) stirring the mixed liquid at the constant temperature of 35 ℃ for 290 seconds, and discharging the mixed liquid from the reaction kettle.
Part of the flow data that may be collected for the reactor run during this process cycle is shown in table 1. The process data of 35 process cycles of the reaction kettle can be collected circularly, namely 35 pieces of data in the table 1 can be obtained.
TABLE 1
| Timestamp (t) | Feed valve 1 (v 1) | Feed valve 2 (v 2) | Discharging valve (v 3) | Liquid level switch 1 (h 1) | Liquid level switch 2 (h 2) | Temperature (tem1) |
| 0 | 1 | 0 | 0 | 0 | 0 | 15.10 |
| 11.5 | 0 | 0 | 0 | 1 | 0 | 18.30 |
| 12 | 0 | 1 | 0 | 1 | 0 | 19.10 |
| 17 | 0 | 0 | 0 | 1 | 1 | 20.50 |
| 27 | 0 | 0 | 0 | 1 | 1 | 32.2 |
| 30 | 0 | 0 | 0 | 1 | 1 | 35 |
| 317.5 | 0 | 0 | 1 | 1 | 1 | 35 |
| 329 | 0 | 0 | 1 | 1 | 0 | 35 |
| 346 | 0 | 0 | 1 | 0 | 0 | 35 |
| 346.5 | 0 | 0 | 0 | 0 | 0 | 35 |
As can be seen from table 1, the data of the two feed valve variables, the two discharge valve variables and the two liquid level switch variables are natural numbers and are discrete variable data, where a value of 1 indicates that the valve or switch is in an open state and a value of 0 indicates that the valve or switch is in a closed state. The temperature variable data is continuous variable data, and the temperature value continuously changes along with time and finally tends to be balanced. The variable data for three valves and two switches may constitute a vector (v 1, v2, v3, h1, h 2). The finite state machine can be trained by using variable data of three valves and two switches, and a state set Pn of the reaction kettle at different time stages can be obtained. As can be seen from table 1, the vectors (v 1, v2, v3, h1, h 2) have 8 value combinations, so the state set Pn includes 8 states P1-P8 and 1 initial state P0. States P1-P8 are P1 (1, 0,0,0, 0), P2 (0, 0,0,1, 0), P3 (0, 1,0,1, 0), P4 (0, 0,0,1, 1), P5 (0, 0,1,1, 1), P6 (0, 0,1,1, 0), P7 (0, 0,1,0, 0), P8 (0, 0,0, 0), respectively, and initial state P0 is a state when the reactor is not operating. In table 1, the data collected by each horizontal bar includes values of the discrete variable and the continuous variable at the same time point, and the sampling time point is stored in the timestamp. Due to space constraints, the timestamps in table 1 are skipped, the timestamps are equidistant in the real data, and the time difference between two adjacent timestamps is always equal to one sampling period. According to the last state and the current value situation of the discrete variable, the continuous variable data in each piece of collected data can be corresponding to one state. The continuous variable data belonging to the same state is regarded as one data packet. Here the temperature variable data is divided into 8 data packets z1, z 2. States P1-P8 and packets z1-z8 are in one-to-one correspondence, e.g., state P1 for packet z1, state P2 for packet z2, and so on. Calculating a functional relationship between the temperature in the data packet corresponding to each state and the time variation, where the functional relationship in this embodiment is a unitary functional relationship, for example, the functional relationship corresponding to the state P1 is y = t +6, the functional relationship corresponding to the state P4 is y =1.2t +1, and so on, where t represents a time stamp and y represents the temperature. In this embodiment, a transition condition set Tn between any two states that can be transitioned in the state set Pn of the reaction kettle can also be obtained, where the transition condition includes a trigger condition of state transition, a probability of state transition occurrence, and a time interval during which the last state stays before the transition. The trigger condition for the state transition of the present embodiment is that 1 or more discrete variable values are changed.
FIG. 3 is a probabilistic mixing finite state machine model of an example reaction vessel according to the present invention. P1 to P8 represent 8 states, respectively, and correspond to Table 1. Each state contains a vector and a functional relation which are composed of discrete variable data. For example, the vector of state P1 is (1, 0,0,0, 0), and the functional relationship is y = t + 6. The vector of state P2 is (0, 0,0,1, 0), and the functional relationship is y = t + 6. When the state P1 is changed to the state P2, the variable v1 of the feed valve 1 is changed from 1 to 0, and at the same time, the variable h1 of the level switch 1 is changed from 0 to 1, so that the trigger condition for the state change caused by the state P1 being changed to the state P2 is (-1, 0,0,1, 0). From the state P1 to the state P2, the present embodiment collects the flow data of 35 process cycles, and the time of the stay of the state P1 is calculated to be between 10.5s and 11.5s according to the timestamp, so that the stay time interval (from the minimum value to the maximum value) of the state P1 is [10.5, 11.5 ]. Since the process performed by the reaction vessel of this example was performed in order, the probability of occurrence of a transition between two adjacent states was considered to be 100%. Thus, the set of transition conditions Tn for transitioning from state P1 to state P2 are (-1, 0,0,1, 0), [10.5, 11.5], and 100%. Similarly, the transition condition sets Tn from state P2 to state P3 are (0, 1,0,0, 0), [0.5, 0.5] and 100%, the transition condition set Tn from state P3 to state P4 is (0, -1,0,0, 1), [4.5, 5] and 100%, the transition condition set Tn from state P4 to state P5 is (0, 0,1,0, 0), [300, 301] and 100%, the transition condition set Tn from state P5 to state P6 is (0, 0,0,0, -1), [11, 12] and 100%, the transition condition set Tn from state P6 to state P7 is (0, 0,0, -1, 0), [17, 18] and 100%, and the transition condition set Tn from state P7 to state P8 is (0, 0, -1,0,0, 0, 5] and 100%. The probabilistic hybrid finite state machine model is saved to a local computer. The staff can utilize this probability to mix finite state machine and monitor and the early warning to reation kettle's process flow.
In light of the foregoing description of the preferred embodiment of the present invention, many modifications and variations will be apparent to those skilled in the art without departing from the spirit and scope of the invention. The technical scope of the present invention is not limited to the contents of the specification, and must be determined by the scope of the claims.
Claims (6)
1. A device state monitoring method based on a probability hybrid finite state machine is characterized by comprising the following steps:
s1: circularly acquiring process data related to process control in a plurality of process periods when the equipment A normally operates, wherein the process data in each process period comprises sensor data required by the process control during the working of the equipment A and switching value data in a PLC (programmable logic controller), and forming a data set X by the process data in all the process periods according to a time stamp sequence;
s2: dividing data in the data set X into a discrete variable data group Xd and a continuous variable data group Xc, training a finite state machine by using the discrete variable data group Xd, and learning a plurality of different states corresponding to different stages of the equipment A in normal operation from the discrete variable data group Xd, wherein the different states form a state set Pn, and each state P in the state set Pn has a corresponding starting time t1 and an ending time t 2; the discrete variable data and the continuous variable data acquired at the same time have the same timestamp; the combination of the current values of the discrete variables forms the current state P of the equipment A, a plurality of continuous variable data in a continuous variable data group Xc are divided into a plurality of data packets Zn according to the timestamp information, each data packet Z corresponds to each state P in a state set Pn one by one, and the data in each data packet Z is acquired in the continuous time period of the corresponding state P;
s3: calculating the function relation of continuous variable data in the data packet Z corresponding to each state P in the state set Pn, which continuously changes along with time;
s4: monitoring a working device A, collecting flow data of the device in real time, and assuming that discrete variable data collected under a certain time stamp are Xdn and continuous variable data are Xcn; current discrete variable data Xdn constitutes a current state vector Pn, matching Pn with state P in state set Pn when device a is operating normally,
if pn cannot find a perfect match, P, the feedback device a operates abnormally,
if pn can find a matching P and the change over time at Xcn meets the functional relationship in step S3, then device a is operating normally, continues to monitor device a,
if pn can find a matching P, but the change over time at Xcn does not satisfy the functional relationship in step S3, the feedback device a operates abnormally.
2. The probability hybrid finite state machine-based device state monitoring method of claim 1, wherein the functional relationships in step S3 include a univariate functional relationship and a multivariate functional relationship,
when the continuous variable data group Xc only contains the change data of one continuous variable, the calculated functional relationship is a unitary functional relationship;
when the continuous variable data group Xc contains the variation data of a plurality of continuous variables, the calculated functional relationship is a multivariate functional relationship, and the multivariate functional relationship represents the correlation relationship between the plurality of continuous variables.
3. The method as claimed in claim 1, wherein when the finite state machine is trained by using the discrete variable data set Xd, a transition condition set Tn between any two states of the state set Pn that can be transited between each other can be obtained, where each transition condition T includes a trigger condition for a state transition, a probability of occurrence of the state transition, and a time interval during which a last state stays before the transition.
4. The device state monitoring method based on the probability hybrid finite-state machine, according to claim 3, wherein the discrete variable data refers to variable data whose value is a natural number or an integer, and the discrete variable data group Xd contains the variation data of at least one discrete variable; the continuous variable data refers to variable data with any value in a defined interval, and the continuous variable data group Xc contains change data of at least one continuous variable.
5. The method for monitoring device status according to claim 4, wherein the training of the finite-state machine with the discrete variable data set Xd in step S2 specifically comprises:
s21: forming a signal vector V by all discrete variables in a discrete variable data set Xd, initializing a state set and a conversion condition set in a finite state machine to obtain an initial state set P ' and an initial conversion condition set T ', setting the state formed by the current signal vector V as an initial state P0, and storing the initial state P0 in the initial state set P ';
s22: monitoring whether the value in the signal vector V changes every other sampling period, wherein the change of the median value in the signal vector V is the trigger condition of state conversion;
if the value in the signal vector V changes, first checking whether the trigger condition of the state transition caused by the current change tc is already recorded in the initial transition condition set T ', and if the trigger condition of the state transition caused by the change tc is not recorded in the initial transition condition set T ', writing the trigger condition of the state transition caused by the change tc into the initial transition condition set T '; if the record exists in the initial conversion condition set T', the record is not modified;
then checking whether the next state st triggered by the change tc is recorded in the initial state set P ', and if the initial state set P ' is not recorded, writing the state st into the initial state set P '; if the record exists in the initial state set P', no modification is made;
calculating the time interval of the stay of the last state before the tc trigger state is converted according to the timestamp information;
s23: if the value in the signal vector V has not changed, repeating step S22 until all the data in the discrete variable data group Xd are used up, writing all the states st into the initial state set P' to obtain a state set Pn in which the device a normally operates; counting the total times of converting the same state into other states, and calculating the probability of converting the same state into other states, namely the probability of state conversion; and writing the triggering conditions of all state conversion, the probability of state conversion and the time interval of the stay of the last state before the conversion into an initial conversion condition set T' to obtain a conversion condition set Tn.
6. The probabilistic hybrid finite state machine-based device state monitoring method of claim 3, wherein the step S4 further comprises:
monitoring a working device A, collecting flow data of the device in real time, and assuming that discrete variable data collected under a certain time stamp are Xdn and continuous variable data are Xcn; the current discrete variable data Xdn constitutes a current state vector Pn, matching Pn with the state P in the state set Pn when device a is operating normally, and matching the transition condition Tn caused by the previous state transition to state Pn with the transition condition T in the transition condition set Tn,
if pn cannot find a perfect match P or tn cannot find a perfect match T, the feedback device a operates abnormally,
if pn can find a perfect match for P, tn can find a perfect match for T, and Xcn meets the functional relationship in step S3 over time, then device a is operating normally, continues to monitor device a,
if pn can find a perfect match P and tn can find a perfect match T, but the change over time at Xcn does not satisfy the functional relationship in step S3, then the feedback device a operates abnormally.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111071624.9A CN113515158B (en) | 2021-09-13 | 2021-09-13 | Equipment state monitoring method based on probability hybrid finite state machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111071624.9A CN113515158B (en) | 2021-09-13 | 2021-09-13 | Equipment state monitoring method based on probability hybrid finite state machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113515158A true CN113515158A (en) | 2021-10-19 |
| CN113515158B CN113515158B (en) | 2021-12-24 |
Family
ID=78063069
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111071624.9A Active CN113515158B (en) | 2021-09-13 | 2021-09-13 | Equipment state monitoring method based on probability hybrid finite state machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113515158B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114923715A (en) * | 2022-04-24 | 2022-08-19 | 扬州惠特科技有限公司 | Abnormal state detection method and system for homogenization and tackifying reaction kettle device |
| CN115129547A (en) * | 2022-06-29 | 2022-09-30 | 苏宁消费金融有限公司 | A black-box monitoring system for consumer finance and its quality analysis method |
| CN116958864A (en) * | 2023-06-26 | 2023-10-27 | 国网江西省电力有限公司南昌供电分公司 | Cable accessory installation process control system and method based on deep neural network |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6223544B1 (en) * | 1999-08-05 | 2001-05-01 | Johnson Controls Technology Co. | Integrated control and fault detection of HVAC equipment |
| CN104931263A (en) * | 2015-06-18 | 2015-09-23 | 东南大学 | Bearing fault diagnosis method based on symbolic probabilistic finite state machine |
| CN108132423A (en) * | 2017-12-14 | 2018-06-08 | 武汉大学 | A kind of method for rapidly positioning based on state transition probability power system monitoring data distortion |
| CN111066003A (en) * | 2017-10-10 | 2020-04-24 | 西门子股份公司 | Method, device and medium for monitoring state of equipment in process industry |
| CN112131069A (en) * | 2019-06-24 | 2020-12-25 | 中船重工特种设备有限责任公司 | Equipment operation monitoring method and system based on clustering |
| CN112198857A (en) * | 2020-12-08 | 2021-01-08 | 浙江中自庆安新能源技术有限公司 | Industrial equipment control optimization method and system based on monitoring data |
-
2021
- 2021-09-13 CN CN202111071624.9A patent/CN113515158B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6223544B1 (en) * | 1999-08-05 | 2001-05-01 | Johnson Controls Technology Co. | Integrated control and fault detection of HVAC equipment |
| CN104931263A (en) * | 2015-06-18 | 2015-09-23 | 东南大学 | Bearing fault diagnosis method based on symbolic probabilistic finite state machine |
| CN111066003A (en) * | 2017-10-10 | 2020-04-24 | 西门子股份公司 | Method, device and medium for monitoring state of equipment in process industry |
| CN108132423A (en) * | 2017-12-14 | 2018-06-08 | 武汉大学 | A kind of method for rapidly positioning based on state transition probability power system monitoring data distortion |
| CN112131069A (en) * | 2019-06-24 | 2020-12-25 | 中船重工特种设备有限责任公司 | Equipment operation monitoring method and system based on clustering |
| CN112198857A (en) * | 2020-12-08 | 2021-01-08 | 浙江中自庆安新能源技术有限公司 | Industrial equipment control optimization method and system based on monitoring data |
Non-Patent Citations (2)
| Title |
|---|
| ALBERTO BARIGOZZI 等: ""A Probabilistic Approach to Fault Diagnosis of Industrial Systems"", 《IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY》 * |
| 包健 等: ""基于有限状态机的电梯控制系统故障诊断方法"", 《计算机应用》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114923715A (en) * | 2022-04-24 | 2022-08-19 | 扬州惠特科技有限公司 | Abnormal state detection method and system for homogenization and tackifying reaction kettle device |
| CN115129547A (en) * | 2022-06-29 | 2022-09-30 | 苏宁消费金融有限公司 | A black-box monitoring system for consumer finance and its quality analysis method |
| CN116958864A (en) * | 2023-06-26 | 2023-10-27 | 国网江西省电力有限公司南昌供电分公司 | Cable accessory installation process control system and method based on deep neural network |
| CN116958864B (en) * | 2023-06-26 | 2026-01-06 | 国网江西省电力有限公司南昌供电分公司 | Cable Accessory Installation Process Control System and Method Based on Deep Neural Network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113515158B (en) | 2021-12-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113515158B (en) | Equipment state monitoring method based on probability hybrid finite state machine | |
| CN1940781B (en) | Method and system for controlling a batch process | |
| KR101518448B1 (en) | Baseline predictive maintenance method for target device and computer program product thereof | |
| Bonada et al. | AI for improving the overall equipment efficiency in manufacturing industry | |
| US10394973B2 (en) | Methods and apparatus for using analytical/statistical modeling for continued process verification (CPV) | |
| US11550288B2 (en) | Method for replacing legacy programmable logic controllers | |
| US9946244B2 (en) | Intelligent programmable logic controller | |
| US20240004355A1 (en) | System and Method for Building and Deploying Prescriptive Analytics to Predict and Control End Product Quality in Batch Production Monitoring and Optimization | |
| EP2587323B1 (en) | System and method for process control using a plurality of supplemental procedures | |
| JP2007188473A5 (en) | ||
| US20200074029A1 (en) | Method for building a model of a physical system | |
| CN104020667A (en) | Improved Kalman filters in process control systems | |
| US11966217B2 (en) | Faulty variable identification technique for data-driven fault detection within a process plant | |
| US20230063452A1 (en) | Systems and methods for evaluating configuration changes on a machine in a manufacturing line | |
| Napoleone et al. | Core characteristics of reconfigurability and their influencing elements | |
| Ciancio et al. | Development of a flexible data management system, to implement predictive maintenance in the Industry 4.0 context | |
| Gatica et al. | An industrial analytics approach to predictive maintenance for machinery applications | |
| US7242989B2 (en) | Apparatus and method for batch property estimation | |
| US20240411300A1 (en) | Method and System for Improving a Production Process in a Technical Installation | |
| Zeiser et al. | Requirements towards optimizing analytics in industrial processes | |
| CN119442858A (en) | A method for detecting and predicting equipment rule anomalies in digital twins | |
| Deng et al. | Fault relevant variable selection for fault diagnosis | |
| Arzac et al. | Industrial Data Science for Batch Reactor Monitoring and Fault Detection | |
| Yang et al. | Machine-learning-enabled decision support for screwdriving process | |
| CN112231912A (en) | Multi-linear regression quality prediction system based on discrete industry industrial Internet of things |
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 |