CN115290368B - Fault diagnosis method in unloading process of powder tanker - Google Patents
Fault diagnosis method in unloading process of powder tanker Download PDFInfo
- Publication number
- CN115290368B CN115290368B CN202211205048.7A CN202211205048A CN115290368B CN 115290368 B CN115290368 B CN 115290368B CN 202211205048 A CN202211205048 A CN 202211205048A CN 115290368 B CN115290368 B CN 115290368B
- Authority
- CN
- China
- Prior art keywords
- time period
- vector
- abnormal
- sequence
- power takeoff
- 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
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M99/00—Subject matter not provided for in other groups of this subclass
- G01M99/005—Testing of complete machines, e.g. washing-machines or mobile phones
Abstract
The invention relates to the technical field of containers for material transportation, in particular to a fault diagnosis method in the unloading process of a powder tank truck, which comprises the steps of collecting the pressure, the temperature and the air flow of the powder tank truck at each detection moment in a plurality of time periods; collecting sound waveform signals corresponding to a plurality of time periods of the power takeoff; acquiring power takeoff vectors corresponding to all time periods according to the sound waveform signals and the custom convolution kernels; obtaining state vectors of all time periods according to the power takeoff vector, the pressure, the temperature and the air flow, and training an One Class SVM vector machine by using the state vectors to obtain a trained One Class SVM vector machine; and inputting the state vector of the current time period into a trained One Class SVM vector machine, and judging whether the unloading process corresponding to the current time period fails according to the output value. The invention can accurately diagnose the fault in the unloading process and improve the detection efficiency.
Description
Technical Field
The invention relates to the technical field of containers for material transportation, in particular to a fault diagnosis method for a powder tank truck in the unloading process.
Background
The powder tanker is suitable for bulk transportation of dry powder materials with the particle diameter of not more than 0.1mm, such as fly ash, cement, lime powder, ore powder, granular alkali and the like, is mainly used for cement factories, cement warehouses and large-scale construction sites, and can save a large amount of packaging materials and loading and unloading labor.
However, the powder tanker is easy to cause the problems of personal safety and economic loss due to the over-high pressure and tank body explosion during the unloading operation; therefore, it is extremely important to accurately detect whether the powder tanker malfunctions during the unloading operation. At present, abnormal monitoring of the powder tanker in the unloading process is carried out, on one hand, an operator directly refers to the pressure in the tank, and if an overpressure condition occurs, pressure relief operation is carried out, so that accidents are avoided; in the mode, an operator is required to monitor the pressure in the tank and the discharging condition at any time, namely, the operator must pay attention to the states of the pressure gauge and the pressure relief valve at any time, and the operator must be in a state of concentrated attention at any time; on the other hand, whether the powder tank truck is abnormal or not is judged according to the state of the air compressor, namely, when a power takeoff related to a power source of the air compressor is abnormal, abnormal sound is generated, and the powder tank truck is judged to be abnormal according to the abnormal sound.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a fault diagnosis method in the unloading process of a powder tank truck, which adopts the following technical scheme:
collecting pressure, temperature and air flow of the powder tanker corresponding to each detection moment in a plurality of time periods; obtaining each pressure sequence, temperature sequence and air flow sequence; collecting sound waveform signals corresponding to a plurality of time periods of the power takeoff;
convolving the sound waveform signal with a user-defined convolution kernel to obtain a first characteristic signal, and multiplying the first characteristic signal by the sound waveform signal to obtain a second characteristic signal; acquiring a power takeoff vector corresponding to each time period according to the second characteristic signal;
obtaining state vectors corresponding to all time periods according to the power takeoff vector, the pressure sequence, the temperature sequence and the gas flow sequence, and training a One Class SVM vector machine by using the state vectors to obtain a trained One Class SVM vector machine;
inputting the state vector corresponding to the current time period into a trained One Class SVM vector machine, and judging whether the unloading process corresponding to the current time period fails according to the output value;
when training the One Class SVM vector machine, adjusting a penalty coefficient and determining a normal time period and an abnormal time period;
the method for acquiring the normal time period and the abnormal time period comprises the following steps: calculating internal abnormal indexes corresponding to all time periods according to the pressure sequence, the temperature sequence and the standard temperature; determining a normal time period and an abnormal time period according to the internal abnormal index and the power takeoff vector corresponding to each time period;
the method for adjusting the penalty coefficient comprises the following steps: calculating the Euclidean distance between any two internal abnormal indexes, and grouping a plurality of time periods according to the Euclidean distance to obtain a plurality of groups; calculating the similarity degree of any two time segments in each group according to the power takeoff vector and the air flow sequence; further calculating the difference between groups corresponding to each group; the penalty factor is adjusted based on the inter-group difference.
Preferably, the custom convolution kernel is
。
Preferably, the method for acquiring the power takeoff vector comprises the following steps: acquiring each inflection point value in the second characteristic signal, comparing each inflection point value with an inflection point threshold value, and counting the number of inflection point values which are larger than the inflection point threshold value and correspond to the inflection point threshold value; recording the ratio of the number of the inflection values to the total number of the inflection values as an abnormal proportion;
arranging the inflection point values in a descending order, and sequentially selecting n inflection point values from front to back and from back to front respectively to obtain 2n inflection point values; then 2n inflection values and the abnormal ratio together form a power takeoff vector; wherein n is greater than 1.
Preferably, the method for acquiring the state vector includes: acquiring a maximum value corresponding to the pressure sequence, calculating an average value corresponding to the temperature sequence, and calculating a variance corresponding to the air flow sequence; and obtaining elements of all dimensions in the power takeoff vector, wherein the maximum value, the average value, the variance and the elements of all dimensions jointly form a state vector.
Preferably, the internal abnormality index is:
wherein, the first and the second end of the pipe are connected with each other,
is an internal abnormality index>
In a pressure sequence>
In a temperature sequence, in>
Is a standard temperature->
Is an exponential function with e as a base>
In function of the maximum value>
As a function of the average.
Preferably, the method for determining the normal time period and the abnormal time period according to the internal abnormal index and the power takeoff vector corresponding to each time period specifically comprises: when the internal abnormal index corresponding to the time period is a negative value and the abnormal proportion in the power takeoff vector is greater than a normal threshold value, the time period is a normal time period; otherwise, the time period is an abnormal time period.
Preferably, the similarity degrees are:
wherein the content of the first and second substances,
to a similar extent for time period A and time period B>
Is the power take-off vector corresponding to time period A, < >>
Is the power take-off vector corresponding to time period B>
For a corresponding sequence of mass flow rates for time period A, <' >>
The air flow sequence corresponding to the time period B; />
Is a function of the similarity of the cosine; />
Is->
And &>
DTW distance of (d).
Preferably, the method for obtaining the difference between the groups is as follows: randomly selecting a time period in any one group, and calculating the cumulative sum of the time period and the similarity degree of other time periods in the group to obtain the membership degree corresponding to the time period; further obtaining the membership degree corresponding to each time period in the group, and taking the time period corresponding to the maximum membership degree as the representative time period of the group; further obtaining the representative time periods of the groups;
and calculating the cumulative sum of the similarity degrees of the representative time periods of the group and the representative time periods of the rest other groups to obtain the characteristic difference of the group, and calculating the ratio of the characteristic difference to the membership corresponding to the representative time periods of the group to obtain the difference between the groups corresponding to the group.
Preferably, the method for adjusting the penalty coefficient based on the difference between groups comprises the following steps:
wherein the content of the first and second substances,
an adjustment value representing the t-th penalty coefficient; />
An adjustment value representing the (t-1) th penalty coefficient; r denotes a penalty factor of->
Then, the verification is concentrated OnThe average distance from all samples to the hypersphere when the output result of the e Class SVM vector machine is inconsistent with the actual result; w represents the error rate of the validation set; />
Representing a constraint parameter; />
Represents a penalty factor of>
And verifying the normalized value of the output result of the concentrated One Class SVM vector machine and the average value of the interclass difference corresponding to the group to which all samples belong when the output result is actually inconsistent.
The embodiment of the invention at least has the following beneficial effects:
according to the method, a power takeoff vector corresponding to each time period is obtained through a second characteristic signal, a state vector corresponding to each time period is obtained according to the power takeoff vector, a pressure sequence, a temperature sequence and a gas flow sequence, a One Class SVM vector machine is trained by using the state vector, the trained One Class SVM vector machine is obtained, the state vector corresponding to the current time period is input into the trained One Class SVM vector machine, and whether a fault occurs in the unloading process corresponding to the current time period is judged according to an output value. The state vector comprises pressure information, airflow information, temperature information and sound information, namely, a plurality of kinds of information are fused to obtain the state vector, and the state of the corresponding unloading process can be more specifically and accurately reflected; and then more accurate analysis results can be obtained in the subsequent process. Meanwhile, when the One Class SVM vector machine is trained, the penalty coefficient is adjusted based on the difference between groups, and the penalty coefficient is corrected through the difference between groups, so that the corrected penalty coefficient can judge the training sample more flexibly, the training time is shortened, and the trained One Class SVM vector machine can be obtained more quickly; namely, the invention has the advantages of high speed and high efficiency.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
Fig. 1 is a flow chart showing steps of an embodiment of a fault diagnosis method in a discharging process of a powder tanker according to the present invention.
Detailed Description
To further explain the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description of the proposed solution, its specific implementation, structure, features and effects will be made with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
Referring to fig. 1, a flow chart of steps of a fault diagnosis method in a discharging process of a powder tanker according to an embodiment of the present invention is shown, where the method includes the following steps:
step 1, collecting pressure, temperature and air flow of a powder tanker corresponding to each detection moment in a plurality of time periods; obtaining each pressure sequence, temperature sequence and air flow sequence; collecting sound waveform signals corresponding to a plurality of time periods of the power takeoff in the historical unloading process.
In the unloading process of the powder tanker, pressure is generally added into a tank body of the powder tanker through an air compressor, namely compressed air (or externally connected compressed air) generated by the air compressor enters the tank, powder and particle materials are fluidized through a fluidizing device, and the powder and particle materials are conveyed to a specified position outside the tank along an unloading pipeline along with the air by means of the pressure difference between the inside and the outside of the tank body. If the pressure in the tank body is abnormal, dangerous conditions may occur, and damage may be caused to the air compressor.
Therefore, in the embodiment, the pressure corresponding to the powder tanker, that is, the pressure in the tank body of the powder tanker, is acquired by using the shock-resistant pressure gauge, so as to obtain the pressure sequence corresponding to each time period
，/>
In which>
Is the pressure corresponding to the 1 st detection moment of the powder tank truck in any time period, and is combined with the pressure regulation device>
The pressure of the powder tanker corresponding to the 2 nd detection moment in the time period,
the pressure corresponding to the mth detection time of the powder tanker in the time period is m, and m is the total number of the detection times in the time period.
The temperature in the tank body can affect the pressure in the tank body, and if the temperature rises, the pressure in the tank body rises; the safety of the tank body is influenced to a certain extent, so that an industrial temperature sensor is arranged on the inner wall of the tank body and used for collecting the temperature in the tank body, and then the temperature sequences corresponding to all time periods are obtained
，/>
Wherein is present>
For the powder tank truck in any time periodThe temperature corresponding to 1 detection instant->
Is the temperature corresponding to the 2 nd detection moment of the powder tanker in the time period>
The temperature corresponding to the mth detection time in the time period of the powder tanker, wherein m is the total number of the detection times in the time period.
In the unloading process of the powder tanker, when the opening angle of the ball valve is large, the material-gas ratio (the ratio of the mass of the material in the pipeline to the mass of the gas) can be reduced, the unloading distance is increased, the conveying height is increased, but the unloading efficiency is reduced, and the unloading time is increased; on the contrary, the material-gas ratio is increased, the conveying distance is shortened, the conveying height is reduced, the unloading efficiency is high, and the unloading time is short; however, when the opening angle of the ball valve is too small, the discharge pipeline is easily blocked and explosion occurs, so for safety, the opening angle of the ball valve needs to be measured, and since the opening angle of the ball valve cannot be directly measured, but the opening angle of the ball valve and the gas flow in the discharge pipeline present a positive correlation, the opening angle of the ball valve is reflected from the side by measuring the gas flow of the discharge pipeline in the embodiment; specifically, an air flow meter is arranged at the front end of the ball valve, and the air flow meter is used for collecting the size of the air flow so as to obtain an air flow sequence
，/>
Wherein is present>
Is the air flow quantity corresponding to the 1 st detection moment of the powder tanker in any time period>
For the powder tank truck at the 2 nd detection moment in the time periodThe desired air flow quantity>
The flow rate is the flow rate corresponding to the mth detection time in the time period of the powder tanker, and m is the total number of the detection times in the time period.
Therefore, in the unloading process of the powder tanker, pressure is generally added into the tank body of the powder tanker through an air compressor, and the power of an engine of the powder tanker transmits a power takeoff, a transmission shaft and other action devices to the air compressor, so that if the power takeoff fails, the air compressor can have insufficient power, and the unloading progress is affected; therefore, characteristic parameters characterizing the power takeoff device need to be acquired, that is, in the embodiment, the MEMS acoustic sensor is disposed on the power takeoff device, and the MEMS acoustic sensor is used to acquire an acoustic waveform signal when the power takeoff device is in operation, and the acoustic waveform signal is used as the characteristic parameters characterizing the power takeoff device.
Specifically, firstly, a sound waveform signal corresponding to the force taking device in a historical unloading process is collected through the MEMS sound sensor, and then the sound waveform signal is cut into a plurality of sections of sound waveform signals, namely, each time section corresponds to one section of sound waveform signal.
It should be noted that the information is collected in a plurality of historical unloading processes corresponding to different powder tank trucks, the plurality of historical unloading processes comprise a normal unloading process and an abnormal unloading process, and the information is collected from the beginning of unloading; meanwhile, one historical unloading process comprises a plurality of time periods, the time length of each time period is 10 seconds, and the time interval between two adjacent detection moments in each time period is 0.5 second; the implementer can adjust the time length corresponding to the time period and the time interval between two adjacent detections according to the actual situation.
Step 2, convolving the sound waveform signal with a custom convolution kernel to obtain a first characteristic signal and multiplying the first characteristic signal by the sound waveform signal to obtain a second characteristic signal; and acquiring a power takeoff vector corresponding to each time segment according to the second characteristic signal.
Custom convolution kernel of
The first characteristic signal obtained by convolving the sound waveform signal with the custom convolution kernel can highlight the local variation trend of the sound waveform signal; then multiplying the first characteristic signal by the sound waveform signal to obtain a second characteristic signal; and acquiring a power takeoff vector corresponding to each time segment according to the second characteristic signal.
The method for acquiring the vector of the power takeoff comprises the following steps: acquiring each inflection point value in the second characteristic signal, comparing each inflection point value with an inflection point threshold value, and counting the number of inflection point values which are larger than the inflection point threshold value and correspond to the inflection point threshold value; recording the ratio of the number of the inflection values to the total number of the inflection values as an abnormal proportion; that is, the formula of the anomaly ratio is expressed as:
(ii) a Wherein +>
Is abnormally accounted for, based on>
For a number of inflection values greater than the inflection threshold>
The total number of knee values. The value of the inflection point threshold in this embodiment is 10 5 . The size of the inflection point threshold value can have different values due to different power takeoff devices, different types of MEMS sound sensors and different placement positions of the MEMS sound sensors; therefore, in the actual operation process, the implementer needs to set the value of the inflection point threshold according to specific situations.
Then arranging the inflection point values in a descending order, and sequentially selecting n inflection point values from front to back and from back to front respectively to obtain 2n inflection point values; then 2n inflection values and the abnormal ratio together form a power takeoff vector; wherein n is greater than 1; namely dimension of power takeoff vector is 2n +1 dimension.
In this embodiment, the value of n is 10, and an implementer can adjust the value of n according to specific situations; selecting the larger 10 inflection points and the smaller 10 inflection points for representing the sound characteristics of the sound waveform signal; wherein the larger 10 inflection points characterize the increasing trend characteristic of the sound waveform signal, and the smaller 10 inflection points characterize the decreasing trend characteristic of the sound waveform signal.
It should be noted that, if the power takeoff is abnormal, the loudness of the sound emitted by the power takeoff and the corresponding frequency of the sound are both relatively high, which indicates that the change of the inflection point value in the second characteristic signal is high, that is, if the power takeoff is abnormal, the corresponding inflection point value is abnormally high, so that the inflection point value can be directly used as the determination of the abnormal operation of the power takeoff. The inflection point value not only has the sound amplitude characteristic, but also contains the change rate characteristic of the inflection point corresponding to the inflection point value in a certain length, and is represented as the frequency point characteristic on a frequency spectrum; therefore, the inflection point in the second characteristic signal is utilized to form a power takeoff vector, the data processing amount can be reduced, the abnormal characteristic of the power takeoff is visually displayed, and the subsequent data processing is facilitated.
And 3, obtaining a state vector corresponding to each time period according to the power takeoff vector, the pressure sequence, the temperature sequence and the gas flow sequence, and training the One Class SVM vector machine by using the state vector to obtain the trained One Class SVM vector machine.
The method for acquiring the state vector comprises the following steps: acquiring a maximum value corresponding to the pressure sequence, calculating an average value corresponding to the temperature sequence, and calculating a variance corresponding to the air flow sequence; obtaining elements of all dimensions in the power takeoff vector, wherein the maximum value, the average value, the variance and the elements of all dimensions form a state vector together; the dimension of the state vector is 2n + 4.
Specifically, the state vector is:
wherein the content of the first and second substances,
is a status vector>
Is a pressure sequence, is>
In a temperature sequence, in>
Is a sequence of the air flow quantity>
For an element in the 1 st dimension in the power take-off vector, is>
For an element in the 2 nd dimension of the power take-off vector, be->
Is the fifth ^ in the power takeoff vector>
Element in dimension->
For taking a function of the maximum value>
In order to find the function of the mean value,
as a function of the variance.
The state vector comprises pressure information, airflow information, temperature information and sound information, namely, a plurality of information are fused to obtain the state vector, and the state of the corresponding unloading process can be more specifically and accurately reflected; and then more accurate analysis results can be obtained in the subsequent process.
Then training an One Class SVM vector machine by using the state vector to obtain a trained One Class SVM vector machine; when training the One Class SVM vector machine, the method comprises the steps of adjusting a penalty coefficient and determining a normal time period and an abnormal time period.
When the One Class SVM vector machine is trained, taking the state vectors corresponding to all time periods as a sample set, taking 80% of the sample set as a training set of the One Class SVM vector machine, taking the rest 20% of the state vectors as a verification set of the One Class SVM vector machine, and training the One Class SVM vector machine by utilizing the training set; and verifying whether the trained One Class SVM vector machine is a trained One Class SVM vector machine or not by using a verification set.
The specific training process is as follows: since One Class SVM is unsupervised learning, no labeling is required for the training set, but by answering: the method of yes or no supports vector data description (support vector domain description SVDD), and trains sample data into a minimum hypersphere (larger than three-dimensional features), wherein a curve is formed in two dimensions, and the data is completely wrapped, that is, abnormal points are eliminated. Therefore, before training, the number of abnormal samples in a training set needs to be determined, and therefore the One Class SVM vector machine is trained to determine a hypersphere. Setting parameters of an One Class SVM vector machine, using a Gaussian kernel function, and setting the training error size as the ratio of the number of unqualified samples in the current sample to the total number of the samples participating in training, namely the ratio of the current abnormal points; inputting the obtained state vectors corresponding to all the time periods into a current One Class SVM vector machine, automatically establishing a current hypersphere by using the One Class SVM, determining whether the unloading process of the time period corresponding to the state vectors has faults or not according to the state vectors corresponding to all the time periods, and outputting the state vectors to be-1 if the unloading process of the time periods corresponding to the state vectors has faults; if no fault occurs, the output of the One Class SVM vector machine is 1.
And after the training is finished, inputting the verification set into an One Class SVM vector machine, determining whether the output value corresponding to each sample is consistent with the actual state (whether the unloading process has a fault) corresponding to the sample, and finishing the training of the One Class SVM vector machine if the output value is consistent with the actual state (whether the unloading process has a fault), namely finishing the training of the One Class SVM vector machine which is finished by the current training, namely finishing the training of the One Class SVM vector machine which is the trained One Class SVM vector machine. And if the parameters are inconsistent, adjusting the parameters of the One Class SVM vector machine, improving the training accuracy, and training the One Class SVM vector machine again until the trained One Class SVM vector machine is obtained.
In the above, it is mentioned that the number of abnormal samples in the training set needs to be determined before training, so the embodiment determines the abnormal samples by determining the normal time period and the abnormal time period; namely, the state vector corresponding to the normal time period is a normal sample, and the state vector corresponding to the abnormal time period is an abnormal sample.
Specifically, the method for acquiring the normal time period and the abnormal time period includes:
(1) And calculating the internal abnormal indexes corresponding to each time period according to the pressure sequence, the temperature sequence and the standard temperature.
The internal anomaly indicators are:
wherein, the first and the second end of the pipe are connected with each other,
is an internal abnormality index>
Is a pressure sequence, is>
In a temperature sequence, in>
Is a standard temperature->
Is an exponential function with e as base>
For taking a function of the maximum value>
As a function of the average.
In the discharging process, the pressure in the tank body is required to be kept not more than 0.2Mpa, meanwhile, the temperature in the tank body is proper during discharging, otherwise, an air compressor or the tank body is damaged; therefore, the internal abnormal index is calculated according to the temperature and the pressure in the tank body and the standard temperature, and the internal abnormal index can preliminarily divide a plurality of time periods into a normal time period and an abnormal time period; when the internal abnormal index is larger than 0 and larger, the unloading process corresponding to the time period is unsafe, and needs to be adjusted, namely, the time period is preliminarily divided into abnormal time periods; if the internal abnormal index is less than 0, the unloading process corresponding to the time period is safe, adjustment is not needed, and the time period is preliminarily divided into normal time periods;
characterizing the difference between the maximum pressure and 0.2 Mpa; as can be seen from the description in step 1, the temperature in the tank influences the pressure in the tank, and if the temperature rises, the pressure in the tank rises, so that a temperature-based ^ based on the pressure is used>
And correcting the maximum pressure, and amplifying the difference between the maximum pressure and 0.2Mpa so as to obtain a more accurate internal abnormal index.
In this example, the standard temperature
The value of (A) is 25 ℃, and the implementer can set the value according to specific conditions.
It should be noted that, if the internal abnormal index is greater than zero, the pressure relief operation is immediately performed on the powder tank, so as to ensure that the air pressure in the powder tank is lower than 0.2Mpa, and avoid the occurrence of danger.
(2) And determining a normal time period and an abnormal time period according to the internal abnormal index and the power takeoff vector corresponding to each time period.
Specifically, when the internal abnormal index corresponding to the time period is a negative value and the abnormal proportion in the power takeoff vector is greater than the normal threshold value, the time period is a normal time period; otherwise, the time period is an abnormal time period. In this embodiment, the value of the normal threshold is 34%, the size of the normal threshold is restricted by the powder tanker and the environment, and in the specific operation process, an implementer can determine the value of the normal threshold according to the actual situation.
The method for adjusting the penalty coefficient comprises the following steps:
the method comprises the steps of firstly, calculating the Euclidean distance between any two internal abnormal indexes, and grouping a plurality of time periods according to the Euclidean distance to obtain a plurality of groups.
Grouping a plurality of time periods by adopting a DBSCAN clustering algorithm to obtain a plurality of groups; when grouping is performed, the present embodiment sets the search radius eps to 0.05, and sets the minimum value minpts in a cluster to 5, that is, at least 5 set time periods are included in one group; in the actual operation process, an implementer can set the values of the search radius and the minimum value in the cluster according to the actual situation. The DBSCAN clustering algorithm is a known technology and will not be described in detail.
Then calculating the similarity degree of any two time periods in each group according to the power takeoff vector and the airflow sequence; further calculating the difference between groups corresponding to each group; the penalty factor is adjusted based on the inter-group difference.
The degree of similarity is:
wherein the content of the first and second substances,
to a similar extent for time period A and time period B>
For the power take-off vector corresponding to time period A>
Is the power take-off vector corresponding to time period B>
For a corresponding sequence of mass flow rates for time period A, <' >>
The air flow sequence corresponding to the time period B; />
Is a function of the similarity of the cosine; />
Is->
And &>
The DTW distance is obtained through a DTW algorithm, which is a known technology and is not described in detail.
It should be noted that, in the following description,
the larger, the more similar the characterization time period A and the time period B are>
Is the vector of the power takeoff>
And power takeoff vector->
Cosine similarity therebetween, in the absence of a base value>
The larger, the more the representative power takeoff vector
And power takeoff vector->
The more similar, and thus the more similar the characterized time period A and the time period B are, the more likely>
Is->
And/or>
The smaller the DTW distance, characterizes &>
And/or>
The more similar, the more similar the characterization time period a and the characterization time period B, that is, the more similar the DTW distance between the airflow volume sequences corresponding to the two time periods and the similarity degree of the two time periods are in a negative correlation relationship, but not a linear relationship, and the more similar the cosine similarity between the power takeoff vectors corresponding to the two time periods and the similarity degree of the two time periods are in a positive correlation relationship, but not a linear relationship; therefore, the calculation formula of the similarity degree is obtained by the mathematical modeling method, and the relationship between each factor and the similarity degree is satisfied.
The reason for calculating the similarity degree by combining the power takeoff vector with the air flow is that the fault of the power takeoff can cause insufficient power of the air compressor, so that the size of the air flow in the discharge pipeline is influenced, and therefore the similarity degree between the two time periods is further determined by the characteristic of the air flow.
The method for acquiring the difference among the groups comprises the following steps: randomly selecting a time period in any one group, and calculating the cumulative sum of the time period and the similarity degree of other time periods in the group to obtain the membership degree corresponding to the time period; further obtaining the membership degree corresponding to each time period in the group, and taking the time period corresponding to the maximum membership degree as the representative time period of the group; further obtaining the representative time periods of the groups; and calculating the cumulative sum of the similarity degrees of the representative time periods of the group and the representative time periods of the rest other groups to obtain the characteristic difference of the group, and calculating the ratio of the characteristic difference to the membership corresponding to the representative time periods of the group to obtain the difference between the groups corresponding to the group.
In this embodiment, the penalty coefficient C is adjusted according to the difference between groups, and the specific adjustment process is as follows:
the penalty factor C is typically selected as
And t has a value of [ -4,4]That is, the value range of C is [0.0001,10000 ]]. The larger the value of the penalty factor C is selected, the greater the penalty for errors, which may result in model overfitting. The generalization capability is deteriorated if the penalty coefficient C is too large or too small, and the initial value of the penalty coefficient C is set to 1 in this embodiment. Then, according to the average distance r between all samples and the hypersphere when the output result of the vector machine in the verification set is inconsistent with the reality, the error rate w in the verification set and the difference between groups, the penalty coefficient C is adjusted, namely ^ er or greater than or equal to>
Wherein, in the step (A),
t and t-1 denote the number of adjustments>
An adjustment value representing a tth penalty factor>
An adjustment value representing a t-1 th penalty factor, r representing a penalty factor of ≥>
And then, the average distance from all samples to the hypersphere when the output result of the One Class SVM vector machine in the verification set is inconsistent with the actual value, w represents the error rate of the verification set, and/or the value is greater than or equal to the threshold value>
Represents a constraint parameter, in this embodiment->
Is 0.1; />
Indicates a penalty factor of->
And verifying the normalized value of the output result of the concentrated One Class SVM vector machine and the average value of the interclass differences corresponding to the groups to which all samples belong when the output result is actually inconsistent.
This example introduces
The effect of (a) is that for a part of samples, which show a certain difference between the current detection factor and the final result, the sample state is indefinite when being classified, and then>
The correction of (2) enables the punishment coefficient to judge the training sample more flexibly, and enables the training of the vector machine to be better. And obtaining the adjusted punishment coefficient through training, if the punishment coefficient is adjusted, the error rate is still low, the state of the fixed sample is uncertain, the sample is considered to be in a judgment boundary, and the current training is ended to obtain the trained One Class SVM vector machine. The training process of One Class SVM vector machine is the prior art, and the embodiment is only based on ^ based on/based on ^ based on>
The penalty factor is adjusted so that it is no longer exceededFor the sake of brevity.
And 4, inputting the state vector corresponding to the current time period into a trained One Class SVM vector machine, and judging whether the unloading process corresponding to the current time period fails according to the output value.
The method for acquiring the state vector corresponding to the current time period in this embodiment is the same as the method for acquiring the state vector corresponding to each time period, and is not described herein again.
Further, the state vector corresponding to the current time period is input into the trained One Class SVM vector machine, whether the unloading process corresponding to the current time period is in fault or not is judged according to the output value, namely when the output value of the trained One Class SVM vector machine is 1, the unloading process corresponding to the current time period is not in fault, and when the output value of the trained One Class SVM vector machine is-1, the unloading process corresponding to the current time period is in fault.
If the unloading process corresponding to the current time period is in fault, the worker makes corresponding adjustment on the powder tank truck according to the state vector corresponding to the current time period, and if only the maximum air pressure in the state vector exceeds 0.2Mp, the worker performs pressure relief operation on the powder tank truck and then continues to operate; if the temperature or the power takeoff is abnormal, the machine needs to be stopped for maintenance.
The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; the modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present application, and are included in the protection scope of the present application.
Claims (7)
1. A fault diagnosis method in the unloading process of a powder tanker is characterized by comprising the following steps:
collecting pressure, temperature and air flow of the powder tanker corresponding to each detection time in a plurality of time periods; obtaining each pressure sequence, temperature sequence and air flow sequence; collecting sound waveform signals corresponding to a plurality of time periods of the power takeoff;
convolving the sound waveform signal with a user-defined convolution kernel to obtain a first characteristic signal, and multiplying the first characteristic signal by the sound waveform signal to obtain a second characteristic signal; acquiring a power takeoff vector corresponding to each time period according to the second characteristic signal;
obtaining state vectors corresponding to all time periods according to the power takeoff vector, the pressure sequence, the temperature sequence and the gas flow sequence, and training a One Class SVM vector machine by using the state vectors to obtain a trained One Class SVM vector machine;
inputting the state vector corresponding to the current time period into a trained One Class SVM vector machine, and judging whether the unloading process corresponding to the current time period fails according to the output value;
when training the One Class SVM vector machine, adjusting a penalty coefficient and determining a normal time period and an abnormal time period;
the method for acquiring the normal time period and the abnormal time period comprises the following steps: calculating internal abnormal indexes corresponding to all time periods according to the pressure sequence, the temperature sequence and the standard temperature; determining a normal time period and an abnormal time period according to the internal abnormal index and the power takeoff vector corresponding to each time period;
the method for adjusting the penalty coefficient comprises the following steps: calculating the Euclidean distance between any two internal abnormal indexes, and grouping a plurality of time periods according to the Euclidean distance to obtain a plurality of groups; calculating the similarity degree of any two time segments in each group according to the power takeoff vector and the air flow sequence; further calculating the difference between groups corresponding to each group; adjusting a penalty factor based on the inter-group differences;
the method for acquiring the vector of the power takeoff comprises the following steps: acquiring each inflection point value in the second characteristic signal, comparing each inflection point value with an inflection point threshold value, and counting the number of the inflection point values corresponding to the inflection point threshold values; recording the ratio of the number of the inflection values to the total number of the inflection values as an abnormal proportion;
arranging the inflection point values in a descending order, and sequentially selecting n inflection point values from front to back and from back to front respectively to obtain 2n inflection point values; then 2n inflection values and the abnormal proportion jointly form a power takeoff vector; wherein n is greater than 1;
the internal abnormality index is:
wherein, the first and the second end of the pipe are connected with each other,
is an internal abnormality index>
Is a pressure sequence, is>
Is a temperature sequence, is->
Is a standard temperature->
Is an exponential function with e as a base>
In function of the maximum value>
As a function of the average.
2. The method for diagnosing the fault in the unloading process of the powder tanker according to claim 1, wherein the custom convolution kernel is
。
3. The method for diagnosing the fault in the unloading process of the powder tanker according to claim 1, wherein the method for obtaining the state vector comprises the following steps: acquiring a maximum value corresponding to the pressure sequence, calculating an average value corresponding to the temperature sequence, and calculating a variance corresponding to the air flow sequence; and obtaining elements of all dimensions in the power takeoff vector, wherein the maximum value, the average value, the variance and the elements of all dimensions jointly form a state vector.
4. The method for diagnosing the fault during the unloading process of the powder tanker according to claim 1, wherein the method for determining the normal time period and the abnormal time period according to the internal abnormal index and the power takeoff vector corresponding to each time period specifically comprises the following steps: when the internal abnormal index corresponding to the time period is a negative value and the abnormal proportion in the power takeoff vector is greater than a normal threshold value, the time period is a normal time period; otherwise, the time period is an abnormal time period.
5. The method for diagnosing the fault in the unloading process of the powder tanker according to claim 1, wherein the similarity degrees are as follows:
wherein the content of the first and second substances,
to a similar extent for time period A and time period B, <' >>
Is the power take-off vector corresponding to time period A, < >>
For the power take-off vector corresponding to time period B>
For a corresponding sequence of mass flow rates for time period A, <' >>
A sequence of airflow rates corresponding to time period B; />
Is a function of the similarity of the cosine; />
Is->
And/or>
DTW distance of (d).
6. The method for diagnosing the fault in the unloading process of the powder tanker according to claim 1, wherein the method for acquiring the difference between the groups comprises the following steps: randomly selecting a time period in any one group, and calculating the cumulative sum of the time period and the similarity degree of other time periods in the group to obtain the membership degree corresponding to the time period; further obtaining the membership degree corresponding to each time period in the group, and taking the time period corresponding to the maximum membership degree as the representative time period of the group; further obtaining the representative time periods of the groups;
and calculating the accumulated sum of the similarity degrees of the representative time periods of the group and the representative time periods of the rest other groups to obtain the characteristic difference of the group, and calculating the ratio of the characteristic difference to the membership degree corresponding to the representative time periods of the group to obtain the group-to-group difference corresponding to the group.
7. The method for diagnosing the fault in the unloading process of the powder tanker according to claim 1, wherein the method for adjusting the penalty coefficient based on the difference between groups comprises the following steps:
wherein the content of the first and second substances,
an adjustment value representing the t-th penalty coefficient; />
An adjustment value representing the (t-1) th penalty coefficient; r denotes a penalty factor of->
Verifying the average distance from all samples to the hypersphere when the output result of the concentrated One Class SVM vector machine is inconsistent with the actual result; w represents the error rate of the validation set; />
Representing a constraint parameter; />
Indicates a penalty factor of->
And verifying the normalized value of the output result of the concentrated One Class SVM vector machine and the average value of the interclass difference corresponding to the group to which all samples belong when the output result is actually inconsistent. />
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211205048.7A CN115290368B (en) | 2022-09-30 | 2022-09-30 | Fault diagnosis method in unloading process of powder tanker |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211205048.7A CN115290368B (en) | 2022-09-30 | 2022-09-30 | Fault diagnosis method in unloading process of powder tanker |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115290368A CN115290368A (en) | 2022-11-04 |
| CN115290368B true CN115290368B (en) | 2023-04-18 |
Family
ID=83834860
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211205048.7A Active CN115290368B (en) | 2022-09-30 | 2022-09-30 | Fault diagnosis method in unloading process of powder tanker |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115290368B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116579663B (en) * | 2023-07-05 | 2023-10-20 | 江苏慧远智能科技有限公司 | Abnormal early warning method in unloading process of powder tank truck |
| CN117434906B (en) * | 2023-12-18 | 2024-02-20 | 山东长兴塑料助剂股份有限公司 | Automatic speed control adding method and system for filter aid powder |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN205708245U (en) * | 2016-04-06 | 2016-11-23 | 荆门宏图特种飞行器制造有限公司 | A kind of overvoltage protective system and install the tank car of this system |
| CN106951197A (en) * | 2017-01-19 | 2017-07-14 | 北京印刷学院 | A kind of printing equipment remote fault diagnosis method and system based on signal testing |
| CN207078049U (en) * | 2017-04-27 | 2018-03-09 | 丘东成 | A kind of powder tank car discharging induction installation |
| CN111703897A (en) * | 2020-05-22 | 2020-09-25 | 山东科洋工控自动化有限公司 | Automatic control device, system and control method for unloading of powder tank truck |
| CN113962095A (en) * | 2021-10-26 | 2022-01-21 | 中车大同电力机车有限公司 | Dumper fault prediction method, system, equipment and storage medium |
| CN114676781A (en) * | 2022-03-28 | 2022-06-28 | 三一电动车科技有限公司 | Vehicle fault diagnosis and prediction method, device and equipment and vehicle |
| CN114645844B (en) * | 2022-05-23 | 2022-07-19 | 蘑菇物联技术（深圳）有限公司 | Method, computing device and computer medium for determining flow state of air compression station |
| CN115009885B (en) * | 2022-07-02 | 2023-03-24 | 中国矿业大学 | Fault monitoring method and device for ore tank discharging and dust removing system |
-
2022
- 2022-09-30 CN CN202211205048.7A patent/CN115290368B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN115290368A (en) | 2022-11-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115290368B (en) | Fault diagnosis method in unloading process of powder tanker | |
| CN116579663B (en) | Abnormal early warning method in unloading process of powder tank truck | |
| US7212953B1 (en) | Portable diagnostic analysis of gas meter and electronic corrector | |
| CN101424948B (en) | Flow rate verification failure diagnosis apparatus | |
| CN110056530B (en) | Anti-surge control method and system | |
| CN116881745B (en) | Pressure transmitter abnormality monitoring method based on big data | |
| CN104303122B (en) | By automatically determining the method that decision threshold monitors aircraft airborne ageing equipment | |
| CN105651325A (en) | Degrading sensor detection implemented within a transmitter | |
| US20140012791A1 (en) | Systems and methods for sensor error detection and compensation | |
| JP5689947B2 (en) | Automatic operation method of measuring device for particle measurement in gas | |
| CN116702082B (en) | Dust monitoring method of micron-sized powder dust-free charging system | |
| CN105424105B (en) | Dust pelletizing system is responsible for air volume detecting method and on-line measuring device | |
| CN105424147B (en) | Silo gravimetric analysis sensing method and device based on grain bulk height Yu bottom surface pressure relation | |
| CN115290316B (en) | Fault diagnosis method for eccentric rotary valve | |
| CN115420437B (en) | Ultrasonic leak detection device based on deep learning | |
| CN116735804A (en) | Intelligent sensor precision monitoring system based on Internet of things | |
| US20060122747A1 (en) | Method for detection of low leak rates in a tire | |
| EP3783320A1 (en) | Detecting and identifying a change in a vibratory meter | |
| EP3406908B1 (en) | Method of anti-surge protection for a dynamic compressor using a surge parameter | |
| US10048238B2 (en) | Method of monitoring production of a chemical product and a chromatograph used therewith | |
| JP2017186939A (en) | Abnormality detection device, abnormality detection method and abnormality detection system | |
| CN108981906A (en) | A kind of track wave mill synthetic fault diagnosis method | |
| EP2922713A1 (en) | System for monitoring the tyre pressure of a motor vehicle | |
| CN114880887B (en) | Health degree early warning method for new energy liquefied gas tank in transportation process | |
| CN116222870B (en) | Dry powder tank car tank body and monitoring system thereof |
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 |