CN116304751B - Operation data processing method for overhauling motor train unit components - Google Patents
Operation data processing method for overhauling motor train unit components Download PDFInfo
- Publication number
- CN116304751B CN116304751B CN202310581734.2A CN202310581734A CN116304751B CN 116304751 B CN116304751 B CN 116304751B CN 202310581734 A CN202310581734 A CN 202310581734A CN 116304751 B CN116304751 B CN 116304751B
- Authority
- CN
- China
- Prior art keywords
- data
- point
- stage
- degree
- difference
- 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
Abstract
The invention relates to the technical field of data processing, in particular to an operation data processing method for overhauling a motor train unit component, which comprises the following steps: obtaining multi-dimensional operation data at different stages and the influence degree of noise according to the speed data in the multi-dimensional operation data and the speed slope value difference between adjacent time points; obtaining a first degree of variability according to the data points of the matching point pairs in the first class component and the second class component; obtaining a time point with larger difference according to the first difference degree; obtaining a plurality of superposition results according to the time points; obtaining the degree of abnormality according to the DTW distance between the first component and the superposition result as well as between the first component and the second component; obtaining a real data trend according to the abnormality degree; obtaining the block size according to the real data trend and the influence degree; and denoising the multidimensional operation data according to the block size. The invention avoids the defect of false acquisition of real data trend, and ensures the noise removal effect and the maximum information retention in the denoising process.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to an operation data processing method for overhauling components of a motor train unit.
Background
The number of motor train units in China is numerous, and because the problems of running faults, operation maintenance quality and the like frequently occur, the maintenance of the motor train units is indispensable, and in order to solve the problems better and improve the efficiency, a set of fault prediction and health management system, namely a PHMP system, of the motor train units is established in China at present. The PHMP system monitors various state data of the motor train unit in the running process in real time by installing an advanced sensor, evaluates the health state of the motor train unit according to the established model and predicts faults. However, the motor train unit is affected by various environmental factors, such as road conditions, bad weather, etc., during operation, and these environmental factors may cause noise to the operation data of the motor train unit. In addition, the structure and the component parts of the motor train unit can generate phenomena such as vibration and friction during operation, and noise of operation data is further increased. Noise may cause abnormal values and other problems in running data of the motor train unit, so that monitoring and predicting effects of the PHMP system are seriously affected. In order to reduce the influence of noise, the operation data is required to be subjected to denoising processing, and the noise is removed by adopting discrete cosine transform conventionally, in the discrete cosine transform processing process, the data is required to be divided into a plurality of blocks with different sizes for analysis, the smaller blocks contain less information and noise information, and the denoising effect is not ideal due to the processing of the smaller blocks; the large blocks contain more information and noise information, but the high-frequency noise information cannot be extracted, and the noise removal effect is not ideal due to the fact that the high-frequency noise information cannot be processed, so that the invention provides a processing method for acquiring real fluctuation trend of data by adopting distribution characteristics of running data of a motor train unit to determine the size of the blocks, and then acquiring running data suitable for overhauling of motor train unit components, which can acquire wrong trend items when fitting the acquisition trend items, due to different data distribution characteristics in the process of acquiring the trend items.
Disclosure of Invention
The invention provides an operation data processing method for overhauling a motor train unit component, which aims to solve the existing problems.
The invention relates to an operation data processing method for overhauling a motor train unit component, which adopts the following technical scheme:
one embodiment of the present invention provides an operation data processing method for component overhaul of a motor train unit, the method comprising the steps of: collecting operation data of a motor train unit;
segmenting the operation data to obtain operation data of different stages and the influence degree of noise possibly suffered by the different stages; the operation data before different scale conversion at different stages is recorded as original data; the data after different scale transformation at different stages is recorded as result data; obtaining a first degree of variability between each time point and each scale in the original data of each stage according to the original data and the result data; the IMF component of the original data after EMD decomposition is marked as a first component, and the component of the first component after DTW matching is marked as a first class component; obtaining a time point with larger difference in the first class component of each stage according to the first difference degree; performing superposition processing according to time points with larger differences to obtain a plurality of superposition results; obtaining a second degree of variability of each time point according to the first component in the original data of each stage and the superposition results; obtaining the abnormal degree of each time point and each scale in the original data of each stage according to the first component, the superposition results and the second difference degree of each stage; obtaining the real data trend of the original data of each stage according to the degree of abnormality; obtaining the block size of each stage according to the real data trend and the influence degree of noise;
And denoising the operation data according to the acquired block size of each stage.
Further, the method for obtaining the operation data of different stages by segmenting the operation data comprises the following steps:
the method comprises the steps of marking a speed slope value of each time point in each stage as a first speed slope value, marking a speed slope value of a time point before each time point as a second speed slope value, marking an absolute value of a difference value between the second speed slope value and the first speed slope value as a segmentation point degree of each time point of each stage, carrying out linear normalization processing on the segmentation point degree of each time point of each stage, presetting a segmentation point degree threshold value T1, and if the segmentation point degree after normalization of the current time point is larger than T1, obtaining operation data of different stages by taking the current time point as a segmentation point of each stage.
Further, the acquisition expression of the degree of influence of noise which may be received at the different stages is as follows:
in the method, in the process of the invention,representing the velocity mean in stage c;representing the maximum value of the velocity in all time points of all phases;a variance value representing the velocity slope values at all points in time in the c-th phase;indicating the extent to which the c-th stage may be affected by noise.
Further, the expression of data acquisition after different scaling at different stages is as follows:
in the method, in the process of the invention,representing the range of transformed scales for the original data start point 1 in the current stage 1,]the number of corresponding data points within;representing adjacent data differences of a corresponding u-th data point of the original data starting point 1 in the current stage in the converted scale parameters;representing adjacent data differences of the v data points corresponding to the original data starting point 1 in the current stage in the converted scale parameters;representing the range of transformed scales for the original data start point 1 in the current stage 1,]a corresponding nth data point;indicating that the current stage raw data is in scale range 1,]and (3) the 1 st data is used as the data after the scale transformation of the starting point.
Further, the method for acquiring the adjacent data difference of the corresponding u-th data point in the converted scale parameter of the original data starting point 1 in the current stage is as follows:
the original data start point 1 in the current stage is in the transformed scale range 1,]the absolute value of the difference between the (u) th data point and the (u-1) th data point in the data processing unit is marked as a first absolute value A1, the absolute value of the difference between the (u) th data point and the (u+1) th data point is marked as a second absolute value A2, the A1 and the A2 are summed, and the sum result is marked as the adjacent data difference of the (u) th data point;
If the u-th data point is the first data point and the former data point is not existed, calculating the absolute value A2 of the difference between the u-th data point and the u+1th data point as the absolute value A1 of the difference between the u-th data point and the u-1th data point; if the u-th data point is the last data point, and when the u+1th data point of the following data point does not exist, calculating the absolute value A1 of the difference between the u-th data point and the u-1 th data point as the absolute value A2 of the difference between the u-th data point and the u+1th data point, and marking the sum result of the A1 and the A2 as the adjacent data difference of the u-th data point.
Further, the method for obtaining the first degree of variability between each time point and each scale in the raw data of each stage is as follows:
the first of the raw dataThe first of the scaled result data of the component and h scale downPerforming DTW matching on the components;
in the method, in the process of the invention,representing the number of corresponding matching point pairs of the j-th time point in the original data subjected to DTW matching;representing the average value of the number of the matching point pairs corresponding to all time points in the original data subjected to DTW matching;representing the average value of the absolute value of the data point difference value of the matching point pair of the original data and the data after the conversion of the h scale in the original data after the DTW matching, wherein the j-th time point corresponds to all the matching point pairs; A first degree of variability of the jth time point from the h scale in the raw data is represented.
Further, the method for obtaining the time point with larger variability in the first class component of each stage according to the first variability degree is as follows:
raw data for each stagePerforming linear normalization processing on the first degree of variability of all time points, presetting a first degree of variability threshold, and if the original data areThe normalized first degree of variability of the current time point is larger than a first degree of variability threshold, the time point is a time point with larger variability, and the first class of each stage is obtainedTime points where the differences in amounts are large.
Further, the obtained expression of the degree of abnormality of each time point and each scale in the raw data of each stage is as follows:
in the method, in the process of the invention,representing the first of the current stage raw dataFirst of component and h scaleDTW distance between components;representing the first of the current stage raw dataComponent and superposition resultDTW distance between;representing a second degree of variability according to a j-th time point in the current-stage raw data;and the abnormal degree value of the jth time point and the h scale in the original data of the current stage is represented.
Further, the method for obtaining the real data trend of the original data of each stage according to the degree of abnormality comprises the following steps:
according to the obtained abnormality degree of each time point in the current stage, carrying out STL time sequence segmentation to obtain a trend item; in the process of obtaining trend item, taking the abnormality degree of each time point in the current stage as a weight, obtaining trend item data in a mode of m-order weighted average in an STL time sequence segmentation algorithm, recording the obtained trend item data as a real data trend, and further obtaining a real data trend of original data of each stage, wherein m is a preset value.
Further, the block size obtaining expression of each stage is as follows:
wherein Z represents the initial block size of the corresponding stage;is the degree of influence of noise possibly received at the corresponding stage;representing a linear normalization function; r represents the chunk size of the corresponding stage.
The technical scheme of the invention has the beneficial effects that: according to the speed data of the running data of the motor train unit, the running data of the motor train unit is staged, the degree of influence possibly affected by noise in each stage is obtained, the abnormal degree value of each time point in the original data is obtained by calculating the frequency information distribution characteristics between the original data and the data under different scale changes, further, the real data trend is obtained in the m-order weighted average process, and the block size in the discrete cosine transform process is determined according to the real data trend. The frequency information distribution characteristics between the original data and the data after different scale changes prevent the defect that the actual data trend is obtained incorrectly due to the fact that the running data of the motor train unit is affected by various noises in the running process of the motor train unit, so that the actual data trend is more accurate, the block size is accurately set in the discrete cosine change processing process, and the noise removing effect and the information maximum reservation in the denoising process are ensured.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of steps of a method of processing operational data for component service of a motor train unit in accordance with the present invention.
Detailed Description
In order to further describe the technical means and effects adopted by the invention to achieve the preset aim, the following detailed description refers to the specific implementation, structure, characteristics and effects of the operation data processing method for overhauling the motor train unit component according to the invention by combining the accompanying drawings and the preferred embodiment. In the following description, different "one embodiment" or "another embodiment" means that the embodiments are not necessarily the same. Furthermore, the particular features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.
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.
The following specifically describes a specific scheme of the operation data processing method for overhauling the motor train unit component provided by the invention with reference to the accompanying drawings.
Referring to fig. 1, a flowchart of steps of an operation data processing method for overhauling a component of a motor train unit according to an embodiment of the present invention is shown, and the method includes the following steps:
step S001: the multi-dimensional operation data of the motor train unit are acquired by arranging various sensors.
The number of motor train units in China is numerous, and because the problems of running faults, operation maintenance quality and the like frequently occur, the maintenance of the motor train units is indispensable, and in order to solve the problems better and improve the efficiency, a set of fault prediction and health management system, namely a PHMP system, of the motor train units is established in China at present. The PHMP system monitors various state data of the motor train unit in the running process in real time by installing an advanced sensor, evaluates the health state of the motor train unit according to the established model and predicts faults. However, the motor train unit is affected by various environmental factors, such as road conditions, bad weather, etc., during operation, and these environmental factors may cause noise to the operation data of the motor train unit. In addition, the structure and the component parts of the motor train unit can generate phenomena such as vibration and friction during operation, and noise of operation data is further increased. Noise may cause abnormal values and other problems in running data of the motor train unit, so that monitoring and predicting effects of the PHMP system are seriously affected. In order to reduce the influence of noise, the operation data needs to be subjected to denoising processing, and the discrete cosine transform is adopted to remove the noise in the embodiment. In the discrete cosine transform processing process, the data is required to be divided into a plurality of blocks with different sizes for analysis, the smaller blocks contain less information and noise information, and the processing of the smaller blocks can lead to unsatisfactory denoising effect; the larger block contains more information and noise information, but cannot extract high-frequency noise information therein, and the processing of the block also causes unsatisfactory denoising effect. In the embodiment, the real fluctuation trend of the data is obtained by adopting the distribution characteristics of the running data of the motor train unit, and the block size is determined.
In the PHMP system, the multi-dimensional operation data of the motor train unit is collected by installing speed, temperature, bearing vibration and gear vibration sensors on the motor train unit, wherein the model of each sensor is not limited in the embodiment, the data collected by each sensor is recorded as the data of each dimension, and the specific collection process is as follows: the motor train unit performs driving movement on a path with the length of 100 km at the average speed of 150km/h, all deployed sensors acquire data once every 5 seconds, and the acquired multidimensional operation data are transmitted to data processing software.
So far, a large amount of multidimensional operation data can be obtained through the acquisition flow.
It should be noted that, the following needs to obtain the real data trend in a self-adaptive manner according to the distribution characteristics of the multidimensional operation data to obtain the block size. And denoising the multidimensional operation data by using discrete cosine transform according to the acquired block size. And finally, transmitting the denoised multidimensional operation data to a PHMP system to evaluate the health state of the motor train unit and predict faults.
It should be further noted that, in the multidimensional operation data collected by the sensor disposed in this embodiment, the sensor outputs one data every 0.25 seconds for one time point, and the data points are recorded, and the number of data points of the multidimensional operation data collected each time is 5/0.25=20.
In addition, it should be noted that, since most of noise generated by the motor train unit belongs to high-frequency noise, compared with other conventional denoising technologies, the discrete cosine transform denoising technology has the following advantages: 1. the precision is high; 2. the pertinence to high-frequency noise is strong; 3. is suitable for processing discrete signals with limited length, and the like. The acquired multidimensional operating data is denoised by discrete cosine transform.
Step S002: according to the distribution characteristics of the multidimensional operation data, the real data trend is acquired in a self-adaptive mode, and the block size is acquired.
In this embodiment, trend item data of the operation data is obtained by the STL time series segmentation algorithm to represent a real data trend. In the process of determining trend item data, an m-order weighted average mode is adopted to analyze trend change of running data, however, as data points of each time point are affected by various noise degrees, weight values of the data points of each time point cannot be determined, so that erroneous trend item data is obtained, erroneous real data trends are shown, and the denoising effect of multidimensional running data is affected. Because different types of noise influence exist, the embodiment divides multidimensional operation data into a plurality of local oscillation mode functions, namely IMF components through an EMD decomposition algorithm, obtains the abnormal degree of each data point through comparing the difference between the IMF components under different scales, obtains the weight value of STL time sequence segmentation according to the abnormal degree, and further obtains the real data trend, and performs denoising treatment.
It should be further noted that, since various operation phases exist in the running process of the motor train unit, the running data of the motor train unit has a stepwise characteristic, and the embodiment divides the running data into three phases, namely an acceleration phase, a deceleration phase and a uniform speed phase. And carrying out phase division on the collected multidimensional operation data, and acquiring the influence degree of each phase possibly affected by noise interference according to the data distribution characteristics of each phase.
In the running data of the motor train unit in all dimensions, since the speed data representing the speed dimension in the multi-dimensional running data has the most obvious change in the running process of the motor train unit, the multi-dimensional running data is divided into stages according to the speed data in the embodiment. Since the distinction between the respective phases is mainly represented by the difference in the change of the speed data at different time points, the present embodiment represents the change of the speed data at different times by the change of the slope value of the speed data at the change of the time points, and the degree of the segment point in which the b-th time point is the segment point of the phaseThe calculation method of (1) is as follows:
in the method, in the process of the invention,the value of the speed slope at time b is indicated, The speed slope value at time b-1 is indicated,；representing the degree of the segmentation point of the b-th time point as the stage segmentation point;the larger the absolute value representing the difference in velocity slope between the b-1 th time point and the b-th time point, the larger the velocity difference between the corresponding two time points, and the more intense the velocity change.
It should be noted that, the speed slope value at the b-th time point is obtained by the following specific method: and connecting the data point of the speed represented by the b time point with the data point of the speed represented by the b-1 time point, and recording the slope value of a line segment formed by the connection as the speed slope value of the b time point. Where if b=1, the speed slope value at this point in time is noted as 0.
Thus far, by the above-mentioned method, the degree of the stage segmentation points can be obtained by taking all the time points as the stage segmentation points, then the linear normalization process is performed on the stage segmentation point degrees of all the time points, and a segmentation point degree threshold T1 is preset, where the embodiment is described by taking t1=0.58 as an example, and the embodiment is not particularly limited, where T1 may be determined according to the specific implementation situation. If the degree of the segmentation point normalized by the current time point is greater than a set threshold value, the time point is indicated to be a stage segmentation point.
So far, according to the segmentation point degree and the segmentation point threshold value of all time points, all stage segmentation points are obtained, and then the collected multidimensional operation data is subjected to stage division according to the stage segmentation points, so that all stages after division are obtained.
The degree of influence of noise in the current stage is related to the average speed of the stage and the fluctuation of the corresponding speed data, wherein if the average speed of the stage is higher, the noise suffered by the motor train unit is more obvious, and if the fluctuation of the speed in the stage is more frequent, the interaction of all parts of the motor train unit is more frequent and is more easily influenced by the noise. Thus the c-th stage may be affected by noise to a degree in all stagesThe calculation method of (1) is as follows:
in the method, in the process of the invention,representing the velocity mean in stage c;representing the maximum value of the velocity in all time points of all phases;indicating the degree of possible noise exposure in stage c;the variance value representing the velocity slope values at all time points in the c-th stage, that is, the fluctuation variation of the velocity data of the c-th stage, the larger the fluctuation variation, the more susceptible the stage to noise; wherein the method comprises the steps of Represent the firstThe average speed in each phase affects the weight value for use as an impact weight for subsequent fluctuation changes.
So far, according to all stages after the obtained multidimensional operation data are divided, the influence degree of possible noise of each stage is obtained.
After the operation of processing in stages, because the noise has different representation capability under different data scale changes and the EMD decomposition algorithm can acquire a plurality of IMF components to represent the distribution characteristics of different frequency information, the embodiment quantifies the degree of abnormality of each time point in each stage according to the acquired difference characteristics by carrying out data scale transformation on the data of each stage and analyzing the differences among the IMF components after the EMD decomposition under different data scales. And then, according to the abnormality degree of each time point in each stage, obtaining the weight value of each time point in the STL time sequence segmentation process, and further obtaining an accurate real data trend.
In all stages, taking data of any dimension in any stage as an example for analysis, firstly, carrying out data conversion under different data scales on the data of the current dimension in the current stage, and dividing the interval [2, M ]]The integer in the matrix is recorded as the scale transformation parameter In this embodiment, m=6 is taken as an example, and the present embodiment is not limited specifically, where M may be determined according to the specific implementation. The specific process of the scale conversion is as follows:
to be used forTo describe an example of a value of (a) for a scale-conversion parameter ofWhen the scale is 1,]performing scale transformation from the first data point in the current stage as the starting point 1, and selecting the subsequent data pointsThe data points are scaled together and then this is doneThe data points are subjected to influence weight analysis to obtain the intrinsic scale range [1 ] of the current stage,]the data points after the inner scale transformation, the current stage is in the scale range [1,]inside 1 st data point as starting point and 1 st scale converted dataThe calculation method of (1) is as follows:
in the method, in the process of the invention,representing the range of transformed scales for the original data start point 1 in the current stage 1,]the number of corresponding data points within;the adjacent data difference of the corresponding u data point of the original data starting point 1 in the current stage in the converted scale parameter is represented, and the calculation method is as follows: the original data start point 1 in the current stage is in the transformed scale range 1,]in this case, the absolute value of the difference between the (u) th data point and the (u-1) th data point is denoted as a first absolute value A1, the absolute value of the difference between the (u) th data point and the (u+1) th data point is denoted as a second absolute value A2, the A1 and the A2 are summed, and the sum result is denoted as the adjacent data difference of the (u) th data point.
In the method, in the process of the invention,representing adjacent data differences of the v data points corresponding to the original data starting point 1 in the current stage in the converted scale parameters;representing the accumulated sum of adjacent data differences of all corresponding points in the scale parameters after the original data start points in the current stage are changed;representing the range of transformed scales for the original data start point 1 in the current stage 1,]a corresponding nth data point;indicating that the current stage raw data is in scale range 1,]and (3) the 1 st data is used as the data after the scale transformation of the starting point.
It should be further noted that, if the u-th data point is the first data point, and there is no u-1 data point that is the previous data point, the absolute value A2 of the difference between the u-th data point and the u+1th data point is calculated as the absolute value A1 of the difference between the u-th data point and the u-1 data point; if the u-th data point is the last data point, and when the u+1th data point of the following data point does not exist, calculating the absolute value A1 of the difference between the u-th data point and the u-1 th data point as the absolute value A2 of the difference between the u-th data point and the u+1th data point, and marking the sum result of the A1 and the A2 as the adjacent data difference of the u-th data point.
Wherein the method comprises the steps ofRepresenting the range of scales 1 after conversion for start point 1 in the current phase,]in, the adjacent data difference of the corresponding u data point, namely the scale range [1 ] of the starting point 1 after conversion in the current stage,]the difference degree between the corresponding u data point and the adjacent points before and after the data point is larger, and if the difference degree is larger, the useful information and the noise information contained in the u data point are more;indicating that the u-th data point is in the transformed scale range 1,]the influence weight value of all the corresponding points in the data point is referred to in the subsequent scale transformation processAnd the loss of information is avoided.
To this end, given a scaling parameterObtaining scale conversion data after 1 st data point in the original data of the current stage is 1 scale converted by taking the 1 st data point as a starting pointThe method comprises the steps of carrying out a first treatment on the surface of the Then from the original data of the current stage, the first stage after the initial point 1The data points are marked as a second starting point 2, and similarly, the scaled data can be calculatedThe method comprises the steps of carrying out a first treatment on the surface of the The first after the starting point 2The data points are marked as a third starting point 3, and similarly, the scaled data can be calculatedThe method comprises the steps of carrying out a first treatment on the surface of the Sequentially performing the steps to obtain the scale conversion data after all the scale conversion in the current stage; thereby obtaining the scale conversion data after all the scales are converted in the current stage of all the scale conversion parameters; and further obtaining the scale conversion data after all the scale conversion at all the stages of all the scale conversion parameters.
Raw data at the current stageConversion data after being converted into scale under the h scalePerforming EMD decomposition to obtain corresponding IMF component, wherein the p-th component of the original data is recorded asThe q-th component in the data at the h-th scale is noted asA component. In this embodiment, the difference between each IMF component of the original data and each IMF component at a different scale needs to be calculated. The local extremum of the converted data obtained after the scale conversion is changed after the original data is subjected to the scale conversion, so that the results of the IMF components obtained by the EMD algorithm are not in one-to-one correspondence. Since different IMF components are represented as information features at different frequencies, the first IMF component of the original data of the current stage and the first IMF component of the scaled data of the current stage are noted as the same bit order IMF component. Thus the IMF components of the same bit sequence may contain information of different frequencies, e.g. in the first of the original dataThe information in the component, which may include the first after downscaling at the h-th scaleAll information of the component and the secondPartial information of the components, therefore, an analysis of the IMF components is required before comparing the differences between the individual IMF components of the raw data and the individual IMF components at different scales.
Raw data of the current stageData after down-conversion of the h scaleThe IMF component of (a) is processed as follows: first with the first of the current stage raw dataThe components are analyzed as initial components, and calculatedComponents andthe components are analyzed for similarity, the DTW algorithm is adopted to calculate the similarity between the time sequences of the two components, the matching of the point pairs between the two components is performed according to the similarity between the time sequences of the two components, the matching result indicates the region with the largest difference between the two components, namely the part with the largest difference of the corresponding matching point pairs, and the matching process needs to be analyzed according to the difference of the different components because the part with the largest difference corresponds to the difference of the different components. Wherein the difference is related to the data difference between the matching point pairs and the number of the matching points corresponding to the same point, and corresponds to the original data of the current stageA first degree of variability of a jth time point from a jth scaleThe calculation method of (1) is as follows:
in the method, in the process of the invention,representing the raw data after DTW matchingThe number of corresponding matching point pairs of the jth time point, namely the number of times of occurrence of the jth time point of the original data in all the matching point pairs; Representing the raw data after DTW matchingIn the method, all time points correspond to the average value of the number of the matching point pairs;representing the raw data after DTW matchingIn the j-th time point corresponds to all matching point pairs, the original dataAn average value of absolute values of data point difference values of matching point pairs of the data after the conversion of the scale at the h scale;representing raw dataA first degree of variability of the jth time point from the h scale.
Wherein the method comprises the steps ofThe difference between the matching point pair representing the jth time point and the matching point pair of other time points in the original data in the matching point pair indicates that the change of the data point under different scale transformation is mainly distributed at the jth time point if the difference of the jth time point is large, and the probability of the occurrence of the situation at the jth time point is large if the frequency information difference of different components occurs. Will beAs the influence weight value when the average value of the matching point to the difference size is calculated later, the larger the value is, the higher the reliability of the average value of the matching point to the difference size is. Wherein the method comprises the steps ofThe mean value of the difference of the data between the matching point pairs is shown, namely the similarity obtained in the DTW algorithm process is mainly subjected to the difference, and the larger the difference is, the larger the influence of the j-th time point in the original data on the difference of the frequency information of different components is shown.
Thus, the first degree of variability between all time points and all scales in the raw data of all stages can be obtained in the above manner.
For the original data of the current stageThe linear normalization process is performed on the first degree of variability of all the time points in the above, and a first degree of variability threshold T2 is preset, where the present embodiment is described by taking t2=0.68 as an example, and the present embodiment is not limited specifically, where T2 may be determined according to the specific implementation situation. If the original dataThe first degree of difference after the normalization of the current time point is greater than T2, and the time point is the time point with greater difference.
Obtaining the original data of the current stage according to the stepsThe data point differences corresponding to the time points represent the information loss between the scale transformed data and the original data under different scale transformation, namely the difference of different frequency information, namely the difference of the IMF component of the original data of the current stage and the IMF component of the data after different scale transformation, namely the first one of the original dataThe information in the component, possibly including the first at the transformed scale at the h scaleAll information of the component and the secondPartial information of the component. This embodiment is achieved by converting the h scale to the second one at the converted scale Covering the data points which do not belong to the time points with larger differences in the components, wherein the second data point is subjected to covering treatmentComponent, with the firstThe components are overlapped to obtain a plurality of overlapped resultsWherein superposition is reconstruction in EMD decomposition algorithm, and two data values of two components at the same time point are directly added, taking superposition of a time point with larger variability as an example, and further generating superposition results corresponding to multiple time points with larger variability. Then by combining the raw dataThe first of (a)Components andthe difference features are analyzed to quantify the degree of abnormality at each time point.
For the original data of the current stageThe first of (a)Component and multiple superposition resultsPerforming DTW calculation, and beforeAnd (3) withThe DTW results are compared and analyzed, and after the components subjected to the analysis are overlapped, the first one of the original data in the current stageThe components are distributed with a plurality of superimposed results after superpositionIf the difference of the DTW results is larger in this case, the value of the degree of abnormality at the time point is indicated to be larger, and the corresponding calculation process combined with the formula (3) introduces a DTW result difference weight to represent the degree of abnormality at each time point. Wherein the degree of abnormality of the jth time point and the h scale in the current stage original data The calculation method of (1) is as follows:
in the method, in the process of the invention,representing the first of the current stage raw dataFirst of component and h scaleDTW distance between components;representing the first of the current stage raw dataComponent and superposition resultDTW distance between;representing the first of the raw data according to the current stageComponent and superposition resultA second degree of variability of the j-th time point obtained, (calculation means such as formula (3));and the abnormal degree value of the jth time point and the h scale in the original data of the current stage is represented. Wherein the method comprises the steps ofRepresenting the DTW resultant difference weight, the smaller the value, the moreAndthe more similar the time point is, the smaller the abnormality degree value of the time point is, and the smaller the reference degree of the corresponding subsequent second degree of variability is.
In addition, the corresponding jth time point is performed when the jth time point is a time point with large differenceIs calculated; if the jth time point is not a time point with large variability, thenLet T3 be a super parameter, wherein the present embodiment is described by taking t3=0.001 as an example, and the present embodiment is not particularly limited, and T3 may be determined according to the specific implementation.
By a similar operation to that described above, by a second one of the current-stage raw dataSecond of the component and the h scaleComponent and thirdAnd calculating among the components to obtain an abnormal degree value of a jth time point under the component combination, and calculating an abnormal degree value mean value of the jth time point of all the component combinations as the abnormal degree of the jth time point under the jth scale after the calculation of all the components except the residual components of the original data is completed.
By the method, the degree of abnormality of the current original data and the jth time point under all scales can be obtained, and the degree of abnormality of the final jth time point with the average value being the corresponding value is obtained.
So far, the variation of different scales is carried out on each stage, and the abnormality degree of all time points in all stages is obtained through the difference of each IMF component after EMD decomposition under different scales.
According to the obtained abnormality degree of each time point in the current stage, carrying out STL time sequence segmentation to obtain a trend item; in the process of obtaining the trend term, the degree of abnormality of each time point in the current stage is used as a weight, and the trend term data is obtained by means of m-order weighted average in the STL time series segmentation algorithm, which is a known technique in the STL time series segmentation algorithm, and is not described in the embodiment. In this embodiment, m=5 is taken as an example, and the embodiment is not limited specifically, where m may be determined according to the specific implementation, and the obtained trend item data is recorded as a true data trend.
According to the steps, the real data trend of the original data in each stage is obtained, the calculation mode of stage segmentation in the formula (1) is combined, the trend segmentation of the real data trend in each stage is obtained, the lengths of two corresponding trend segmentation points are the initial block sizes, the block sizes in each stage are obtained by combining the influence degree of noise possibly suffered by each stage in the formula (2), one block size super parameter Z is preset, the embodiment is described by taking Z=8 as an example, the embodiment is not particularly limited, and Z can be determined according to specific implementation conditions. ThenThe block sizes in the corresponding stages are:
wherein Z represents the initial block size of the corresponding stage;is the degree of influence of noise possibly received at the corresponding stage;representing a linear normalization function; r represents the chunk size of the corresponding stage.
So far, according to the obtained abnormality degree of each time point of the original data in each stage, performing self-adaptive STL time sequence segmentation to obtain the real data trend of the original data in each stage, and obtaining the block sizes in all stages according to the real data trend of the original data in each stage.
Step S003: and denoising the multi-dimensional operation data of the motor train unit according to the acquired block size.
The multi-dimensional operation data of the motor train unit is subjected to discrete cosine change denoising according to the acquired block size, wherein the multi-dimensional operation data is segmented according to the block size, each block is subjected to discrete cosine change processing, and a coefficient threshold T4 is preset, wherein the embodiment is described by taking t4=0.2 as an example, the embodiment is not particularly limited, T4 can be determined according to the specific implementation, and if the coefficient is larger than the threshold T4, the coefficient is marked as a coefficient close to 0.
So far, all coefficients close to 0 can be obtained by the above method. And then removing coefficients close to 0, wherein the retained coefficients are larger coefficients, performing inverse discrete cosine transform on each block, and connecting each block together to obtain denoised data.
This embodiment is completed.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.
Claims (9)
1. An operation data processing method for overhauling a motor train unit component is characterized by comprising the following steps of:
Collecting operation data of a motor train unit;
segmenting the operation data to obtain operation data of different stages and the influence degree of noise possibly suffered by the different stages; the operation data before different scale conversion at different stages is recorded as original data; the data after different scale transformation at different stages is recorded as result data; obtaining a first degree of variability between each time point and each scale in the original data of each stage according to the original data and the result data; the IMF component of the original data after EMD decomposition is marked as a first component, and the component of the first component after DTW matching is marked as a first class component; obtaining a time point with larger difference in the first class component of each stage according to the first difference degree; performing superposition processing according to time points with larger differences to obtain a plurality of superposition results; obtaining a second degree of variability of each time point according to the first component in the original data of each stage and the superposition results; obtaining the abnormal degree of each time point and each scale in the original data of each stage according to the first component, the superposition results and the second difference degree of each stage; obtaining the real data trend of the original data of each stage according to the degree of abnormality; obtaining the block size of each stage according to the real data trend and the influence degree of noise;
Denoising the operation data according to the acquired block size of each stage;
the block size of each stage is obtained as follows:
wherein Z represents the initial block size of the corresponding stage;
beta is the degree of influence of noise possibly in the corresponding stage;representing a linear normalization function; r represents the block size of the corresponding stage;
the method for acquiring the initial block size comprises the following steps:
calculating stage segmentation points corresponding to the real data trend of the original data of each stage, and taking the length between the stage segmentation points corresponding to the real data trend of the original data of each stage as the initial block size;
the method for obtaining the operation data of different stages by segmenting the operation data comprises the following steps:
the method comprises the steps of marking a speed slope value of each time point in each stage as a first speed slope value, marking a speed slope value of a time point before each time point as a second speed slope value, marking an absolute value of a difference value between the second speed slope value and the first speed slope value as a segmentation point degree of each time point of each stage, carrying out linear normalization processing on the segmentation point degree of each time point of each stage, presetting a segmentation point degree threshold value T1, and if the segmentation point degree after normalization of the current time point is larger than T1, obtaining operation data of different stages by taking the current time point as a segmentation point of each stage.
2. The method for processing operation data for maintenance of components of a motor train unit according to claim 1, wherein the degree of influence of noise which may be received at the different stages is obtained as follows:
in the method, in the process of the invention,representing the velocity mean in stage c; />Representing all stagesA maximum value of the velocity in all time points of (a); />A variance value representing the velocity slope values at all points in time in the c-th phase; />Indicating the extent to which the c-th stage may be affected by noise.
3. The method for processing operation data for maintenance of a motor train unit component according to claim 1, wherein the expression of acquisition of data after different scaling at the different stages is as follows:
in the method, in the process of the invention,representing the scale range [1, ] of the original data starting point 1 after conversion in the current phase>]The number of corresponding data points within; />Representing adjacent data differences of a corresponding u-th data point of the original data starting point 1 in the current stage in the converted scale parameters; />Representing adjacent data differences of the v data points corresponding to the original data starting point 1 in the current stage in the converted scale parameters; />Representing the scale range [1, ] of the original data starting point 1 after conversion in the current phase >]A corresponding nth data point; />Representing that the original data of the current stage is in the scale range [1, ]>]And (3) the 1 st data is used as the data after the scale transformation of the starting point.
4. A method for processing operation data for overhauling components of a motor train unit according to claim 3, wherein the method for acquiring the adjacent data difference of the u-th data point corresponding to the original data starting point 1 in the converted scale parameter in the current stage is as follows:
the original data start point 1 in the current stage is in the transformed scale range 1,]the absolute value of the difference between the (u) th data point and the (u-1) th data point in the data processing unit is marked as a first absolute value A1, the absolute value of the difference between the (u) th data point and the (u+1) th data point is marked as a second absolute value A2, the A1 and the A2 are summed, and the sum result is marked as the adjacent data difference of the (u) th data point;
if the u-th data point is the first data point and the former data point is not existed, calculating the absolute value A2 of the difference between the u-th data point and the u+1th data point as the absolute value A1 of the difference between the u-th data point and the u-1th data point; if the u-th data point is the last data point, and when the u+1th data point of the following data point does not exist, calculating the absolute value A1 of the difference between the u-th data point and the u-1 th data point as the absolute value A2 of the difference between the u-th data point and the u+1th data point, and marking the sum result of the A1 and the A2 as the adjacent data difference of the u-th data point.
5. The method for processing the operation data for overhauling the components of the motor train unit according to claim 1, wherein the method for acquiring the first degree of variability between each time point and each scale in the raw data of each stage is as follows:
the first of the raw dataThe first of the scaled result data of the component and h scale downPerforming DTW matching on the components;
in the method, in the process of the invention,representing the number of corresponding matching point pairs of the j-th time point in the original data subjected to DTW matching; />Representing the average value of the number of the matching point pairs corresponding to all time points in the original data subjected to DTW matching; />Representing the average value of the absolute value of the data point difference value of the matching point pair of the original data and the data after the conversion of the h scale in the original data after the DTW matching, wherein the j-th time point corresponds to all the matching point pairs; />A first degree of variability of the jth time point from the h scale in the raw data is represented.
6. The method for processing operation data for maintenance of motor train unit components according to claim 1, wherein the method for obtaining a time point with a larger difference in the first class component of each stage according to the first degree of difference is as follows:
Raw data for each stagePerforming linear normalization on the first degree of variability of all time points, presetting a first degree of variability threshold, and if the original data is +.>The normalized first degree of difference of the current time point is larger than a first degree of difference threshold, and the time point is a time point with larger difference, so that the time point with larger difference in the first type of components of each stage is obtained.
7. The method for processing operation data for overhauling components of motor train unit according to claim 1, wherein the obtained expression of the degree of abnormality of each time point and each scale in the raw data of each stage is as follows:
in the method, in the process of the invention,first +.>First +.of component and h-th scale>DTW distance between components; />First +.>Multiple superposition results of component and superposition ∈>DTW distance between; />Representing a second degree of variability according to a j-th time point in the current-stage raw data; />And the abnormal degree value of the jth time point and the h scale in the original data of the current stage is represented.
8. The method for processing operation data for maintenance of motor train unit components according to claim 1, wherein the method for obtaining the actual data trend of the raw data of each stage according to the degree of abnormality is as follows:
According to the obtained abnormality degree of each time point in the current stage, carrying out STL time sequence segmentation to obtain a trend item; in the process of obtaining trend item, taking the abnormality degree of each time point in the current stage as a weight, obtaining trend item data in a mode of m-order weighted average in an STL time sequence segmentation algorithm, recording the obtained trend item data as a real data trend, and further obtaining a real data trend of original data of each stage, wherein m is a preset value.
9. The method for processing operation data for maintenance of components of a motor train unit according to claim 1, wherein the block size obtaining expression of each stage is as follows:
wherein Z represents the initial block size of the corresponding stage; />Is the degree of influence of noise possibly received at the corresponding stage; />Representing a linear normalization function; r represents the chunk size of the corresponding stage.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310581734.2A CN116304751B (en) | 2023-05-23 | 2023-05-23 | Operation data processing method for overhauling motor train unit components |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310581734.2A CN116304751B (en) | 2023-05-23 | 2023-05-23 | Operation data processing method for overhauling motor train unit components |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116304751A CN116304751A (en) | 2023-06-23 |
| CN116304751B true CN116304751B (en) | 2023-08-01 |
Family
ID=86830915
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310581734.2A Active CN116304751B (en) | 2023-05-23 | 2023-05-23 | Operation data processing method for overhauling motor train unit components |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116304751B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116991199B (en) * | 2023-09-25 | 2024-01-02 | 深圳扬兴科技有限公司 | Clock signal intelligent noise reduction method for clock chip |
| CN117195007B (en) * | 2023-11-08 | 2024-02-06 | 张家港长寿工业设备制造有限公司 | Heat exchanger performance prediction method and system |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9832351B1 (en) * | 2016-09-09 | 2017-11-28 | Cisco Technology, Inc. | Reduced complexity video filtering using stepped overlapped transforms |
| CN110197211A (en) * | 2019-05-17 | 2019-09-03 | 河海大学 | A kind of similarity data clustering method towards dam safety monitoring data |
| CN113128781B (en) * | 2021-04-30 | 2021-12-10 | 大连理工大学 | Distributed industrial energy operation optimization platform for automatically constructing intelligent model and algorithm |
| CN115659160B (en) * | 2022-12-28 | 2023-06-16 | 北京中航路通科技有限公司 | Data quality measurement method for digital twin model optimization |
-
2023
- 2023-05-23 CN CN202310581734.2A patent/CN116304751B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN116304751A (en) | 2023-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116304751B (en) | Operation data processing method for overhauling motor train unit components | |
| Li et al. | The entropy algorithm and its variants in the fault diagnosis of rotating machinery: A review | |
| CN107832525B (en) | Method for extracting bearing fault characteristic frequency by optimizing VMD (variable minimum vector machine) through information entropy and application of method | |
| CN105275833A (en) | CEEMD (Complementary Empirical Mode Decomposition)-STFT (Short-Time Fourier Transform) time-frequency information entropy and multi-SVM (Support Vector Machine) based fault diagnosis method for centrifugal pump | |
| CN113158814B (en) | Bearing health state monitoring method based on convolution self-encoder | |
| CN111413089A (en) | Gear fault diagnosis method based on combination of VMD entropy method and VPMCD | |
| CN116320043B (en) | Method and system for improving transmission efficiency of multi-carrier communication system | |
| CN109918417B (en) | Time sequence data self-adaptive segmentation, dimension reduction and characterization method based on wavelet transformation and application | |
| Ma et al. | Early fault diagnosis of rotating machinery based on composite zoom permutation entropy | |
| CN113537102B (en) | Feature extraction method of microseismic signals | |
| CN113283028A (en) | Fault diagnosis method for gear of gear box | |
| CN117076955A (en) | Fault detection method and system for high-voltage frequency converter | |
| CN110222390B (en) | Gear crack identification method based on wavelet neural network | |
| CN112733612A (en) | Cross-domain rotating machinery fault diagnosis model establishing method and application thereof | |
| Han et al. | Fault diagnosis of rolling bearings using dual-tree complex wavelet packet transform and time-shifted multiscale range entropy | |
| Bulla et al. | Feature extraction analysis using filter banks for faults classification in induction motors | |
| Matania et al. | Transfer across different machines by transfer function estimation | |
| CN115577237A (en) | Elevator fault diagnosis method based on one-dimensional convolutional neural network and meta-learning | |
| CN113639985B (en) | Mechanical fault diagnosis and state monitoring method based on optimized fault characteristic frequency spectrum | |
| Li et al. | Fault Feature Extraction Method of Reciprocating Compressor Valve Based on SPA-MF | |
| CN110530630B (en) | Planetary gear box gear local fault diagnosis method based on improved dynamic time warping | |
| Wang et al. | Degradation pattern identification and remaining useful life prediction for mechanical equipment using SKF-EN | |
| CN113188651A (en) | Vibration signal feature extraction and tool wear value prediction method based on IMF-PSD spectral line | |
| Liu et al. | A Method for Identifying Weak Faults of Rolling Bearings Around Railway Based on PCA&SVD-LMD | |
| CN112948752B (en) | Typical working condition construction method for long movement segment of train |
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 |