CN104679982B - A kind of recognition methods of the THDS system result of detection validity of feature based value detection - Google Patents
A kind of recognition methods of the THDS system result of detection validity of feature based value detection Download PDFInfo
- Publication number
- CN104679982B CN104679982B CN201410837264.2A CN201410837264A CN104679982B CN 104679982 B CN104679982 B CN 104679982B CN 201410837264 A CN201410837264 A CN 201410837264A CN 104679982 B CN104679982 B CN 104679982B
- Authority
- CN
- China
- Prior art keywords
- value
- line segment
- detection
- grader
- threshold value
- 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
Landscapes
- Image Analysis (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention discloses a kind of recognition methods of the THDS system result of detection validity of feature based value detection, belong to railway freight-car operation troubles detection field;Comprise the following steps:Step 1: choosing the invalid training sample data of measuring temperature of bearing value, 32 all voltage value datas are normalized;Step 2: extracting characteristic value to the training sample after each normalized, characteristic data set is formed;Step 3: being generated using the characteristic data set of training sample and training 32 waveform validity graders;Step 4: utilizing validity grader progress Effective judgement to 32 waveforms collected.The present invention can increase feature Value Types and corresponding grader to provide more accurate recognition result, can fully automatic operation, without manual intervention, recognition speed is fast, while exporting recognition result and judging reason, is easy to user to check recognition result.
Description
Technical field
The invention belongs to railway freight-car operation troubles detection field, and in particular to a kind of THDS systems of feature based value detection
The recognition methods for result of detection validity of uniting.
Background technology
THDS (THDS), also known as rolling stock axle temperature intelligent detecting system;Utilize rail side infrared ray
Probe and intelligent tracking device detect the bearing temperature of operation vehicle in real time at a high speed, are tracked forecast, are to find vehicle in time
Hot axle, prevention overheating of axle bearing, it is ensured that the visual plant of safety of railway traffic.
What The Ministry of Railway of the People's Republic of China, MOR issued《Vehicle axle temperature intelligent detecting system (THDS) equipment repair and maintenance are managed
Code》Middle regulation, THDS equipment encashment ratio must reach more than 70%.The computational methods of encashment ratio are as follows:
When the train of operation passes through THDS equipment, infrared probe understands continuous acquisition vehicle bearing position or so totally 50 centimetres of models
Interior temperature is enclosed, actual acquisition value is magnitude of voltage;The continuous voltage value collected is averagely divided into 32 in 50 cm ranges
Individual region, takes out the maximum in each region, i.e., 32 magnitudes of voltage;32 magnitudes of voltage of data are normalized, obtained
32 0~100 interval numerical value, form oscillogram, represent 32 waveforms of axle temperature, be referred to as " 32 waveforms ";Afterwards by program
Take out " 32 waveforms " in voltage max and by formula scales be temperature value, as the maximum temperature of the bearing portion,
Forecast according to hot axle Forecast Standard.
During the complete temperature computation of THDS systems, three groups of data outputs are had, are bearing portion continuous 50 respectively
The magnitude of voltage of cm range, 32 magnitudes of voltage, 32 voltage max.
32 voltage max are maximum magnitudes of voltage in " 32 waveforms ", are one or more identical value.
If carrying out validity identification to all magnitudes of voltage near bearing in continuous 50 cm range, amount of calculation is too big, holds
Easily influence the real-time of system;If validity identification is carried out to 32 final voltage max, due to the unobvious nothing of feature
Method is accurately identified.
Therefore it is general to carry out validity identification using to " the 32 waveforms " data generated in calculating process, optimize amount of calculation
And improve identification accuracy.
Foundation is forecast because infrared probe is easily influenceed in detection process by extraneous factor, therefore as hot axle
The maximum of " 32 waveforms " be possible to be not true bearing temperature, if not carrying out abnormality detection to it, easily miss
The phenomenon of hot axle is reported, encashment ratio is reduced.
The content of the invention
It is an object of the invention to reduce the quantity of THDS system hot box misreportings, the encashment ratio of THDS equipment is improved, therefore
Need to judge the validity of temperature-measuring results, when finding have abnormal interference during thermometric, do not forecast hot axle or
Degrade the hot axle of forecast.
A kind of recognition methods of the use THDS system result of detection validity of feature based value detection, specific steps are such as
Under:
Step 1: choosing the invalid training sample data of measuring temperature of bearing value, 32 all voltage value datas are returned
One change is handled;
Step 2: extracting characteristic value to the training sample after each normalized, characteristic data set is formed;
Each training sample has multiple characteristic values, and characteristic value includes:Flat-top points characteristic value;Waveform widths characteristic value;
Saturation width characteristics value;Saturation detection characteristic value;Line segment detection characteristic value;Energy eigenvalue;Sharp thermoelectricity forces down mark;Trailing edge
Points maximum;The difference of voltage maximin;The count feature value of line segment folding;Ascending tendency line segment quantity;Tank car, boxcar
Sunlight interference mark and specific characteristic value;All eigenvalue clusters are into characteristic data set.
1st, flat-top points characteristic value:
After 32 magnitudes of voltage in each training sample are normalized, first obtained greatest measure
Point and its left and right sides continuously exceed threshold value a points, are used as flat-top points characteristic value;Threshold value a scope is 80-100.
2nd, waveform widths characteristic value:
After 32 magnitudes of voltage in each training sample are normalized, first obtained greatest measure
Point and its left and right sides continue to exceed threshold value a points, are used as waveform widths characteristic value;
3rd, saturation width characteristics value:
When 32 voltage max in training sample are more than 9990 millivolts but not less than 10000 millivolts, by 32 points of electricity
After the normalization of pressure value, corresponding first greatest measure point and its left and right sides continue to exceed threshold value a points, wide as saturation
Spend characteristic value;
4th, saturation detection characteristic value;
Saturation detection characteristic value:Refer to the points for being less than threshold value b during saturation after normalization;
When 32 voltage max in training sample are more than 9990 millivolts but not less than 10000 millivolts, by 32 points of electricity
After the normalization of pressure value, statistic is less than threshold value b points, should as the points characteristic value for being less than 20 after being normalized during saturation
Characteristic value result is numeral;Threshold value b scopes are:20-35;
5th, Line segment detection characteristic value:
Line segment detection characteristic value includes:Long line segment quantative attribute value, rising edge characteristics value and flat-top position indicate rearward;
51st, long line segment quantative attribute value:
In line segment array using the formation of line segment tendency determination methods, statistics line segment sum ignores line segment two ends label poor
Be worth for 1 and amplitude be less than threshold value c line segment, be used as long line segment quantative attribute value;Amplitude refers to the magnitude of voltage normalizing at line segment two ends
The absolute value of the difference of numerical value after change;Threshold value c scopes are:8-10.
52nd, rising edge characteristics value;
In line segment array using the formation of line segment tendency determination methods, after beginning label is threshold value d point, if starting
There is the rising edge that amplitude is more than threshold value e in point and terminating point, and characteristic value is set into 1,0 is otherwise set to.
Threshold value d scopes are:6-10, threshold value e scopes are:40-50.
53rd, flat-top position indicates rearward:
In line segment array using the formation of line segment tendency determination methods, if the beginning label of first horizontal tendency line segment is big
Exceed threshold value a in numerical value after threshold value f and normalization;The training sample mark is set to 1,0 is otherwise set to;
Threshold value f scopes are:8-12, preferably 12;Threshold value a preferably 90;
The determination methods of line segment tendency:
The first step:After 32 magnitudes of voltage are normalized, 32 oscillograms are judged:
1), since first label, continuous two labels are chosen successively, and corresponding numerical value after normalization is made poor,
If the difference of latter numerical value and previous numerical value exceedes threshold value g, it is ascending tendency to set at this 2 points;If instead previous
The difference of individual numerical value and latter numerical value exceedes threshold value g, is set to decline tendency;Threshold value g range is:4-6.
After treatment, 32 magnitudes of voltage are divided into line segment between any two, 31 altogether, respectively marked as 1-31, i.e.,
The corresponding start-stop of line segment 1 marked as 0 and 1, the like.
If 2), the difference of latter numerical value and previous numerical value is no more than after threshold value g, but addition thirdly, 3 points pairs
The numerical value answered continuously increases, and thirdly and the difference of first point of corresponding numerical value is more than or equal to threshold value c, then sets at this 3 points to be upper
Rise tendency;Conversely, after adding thirdly, 3 points of corresponding numerical value are continuously reduced, and first point and thirdly the difference of numerical value is more than
Equal to threshold value c, then set at this 3 points to decline tendency.
If 3), be unsatisfactory for above-mentioned two condition, it is horizontal tendency to set at this 2 points;
Second step:After the first step, if continuous 3 line segments meet following condition:Line segment 1 is identical with the tendency of line segment 3, line
Section 2 is the horizontal tendency line segment that line segment two ends label difference is 1, then is set to walk with line segment 1 and the identical of line segment 3 by line segment 2
Three line segments are merged into a line segment by gesture;Length is that 1 to refer to that the termination label of line segment 2 subtracts the value of beginning label be 1;
3rd step:If certain the start-stop label for rising or falling line segment, the voltage difference of corresponding points is no more than 50 millivolts, then
It is horizontal tendency to set the line segment tendency, that is, has filtered the fluctuating that industry disturbance information is likely to result in;
4th step:Merge continuous and tendency identical line segment;
The final result that is obtained by the judgement of line segment tendency is:Line segment array;
6th, energy eigenvalue:Use 32 point value sums after normalization;The result of energy is located between 0-3200.
7th, swash thermoelectricity and force down mark:When thermal level is swashs heat, when 32 voltage max are not less than -3 volt, training sample is set
In sharp thermoelectricity force down flag sign value for 1, be otherwise provided as 0;
8th, trailing edge points maximum;
According to the judgement of line segment tendency, the line segment beginning label of trailing edge and termination label are made into poor, absolute value as a result
For trailing edge points maximum;
9th, the difference of voltage maximin;
32 voltage max of data in each training sample make poor with 32 voltage minimums, are used as this feature
Value.
10th, the count feature value of line segment folding:
It is raising and lowering or decline and rising by continuous two lines section tendency, folding, which is counted, Jia one;Continuous three line segments
In, line segment 1,3 tendencies are raising and lowering or decline and rising, and the tendency of line segment 2 is horizontal tendency, and folding, which is counted, Jia one;
11st, ascending tendency line segment quantity;
According to the tendency of line segment, statistics provides the line segment quantity of the line segment ascending tendency between label.
12nd, tank car, boxcar sunlight interference mark:
First three point value is all higher than threshold value a after vehicle is G or X, and normalization;And there is day null value high standard will, and axle
When position is 1 axle or 3 axle, it is 1 to put this flag bit, is otherwise set to 0;
13rd, specific characteristic value:Specifically 2,4 axial extent characteristic values;
When acquisition station is tries to win the champion or tea room, and axle position calculates eigen value when being 2 or 4 axle, calculate greatest measure point and its
The quantity of the point of left and right continuously more than or equal to threshold value a.
Step 3: being generated using the characteristic data set of training sample and training 32 waveform validity graders.
Whole training samples are extracted after characteristic value, by the analytic induction to training sample characteristic value, finding out can be accurate
Really distinguish training sample and the validity classifier parameters of normal sample.
Validity grader includes:Flat-top detects grader, waveform widths detection grader, the classification of saturation width detection
Device, polarization detection grader during saturation, long line segment quantity detects grader, and rising edge detection grader, flat-top is examined rearward
Grader, energy measuring grader are surveyed, sharp thermoelectricity forces down detection grader, trailing edge points maximum detection grader, voltage
Amplitude detection grader, line segment folding number of times detection grader, rising edge tendency detection grader, sunlight interference detection grader and
2nd, 4 axial extents detection grader;
In the present invention, with reference to the various features value in step 2, the grader of establishment is as follows:
Described threshold value a' scopes are:2-4;Threshold value b' scopes are:2-5;Threshold value c' is:5-7;Threshold value d' scopes are:4-
7;Threshold value f' scopes are:More than 1 and less than or equal to 7;Threshold value g' scopes are:More than or equal to 500 and less than 2500;Threshold value g " scopes
For:More than or equal to 500 and less than 2200;Threshold value h' scopes are:Less than 15;Threshold value i' scopes are 0 to 2;Threshold value i " scopes are:2
To 4.
Step 4: carrying out Effective judgement to 32 waveforms collected.
Detailed process is as follows:
Step 401,32 point datas collected are normalized;
By collect 32 magnitudes of voltage, it is normalized according to step one.
Step 402, to after normalized numerical value extract characteristic value;
To the numerical value after normalization, the data of combined training sample;Respectively according to the characteristic value in step 2 to collecting
Data calculated, obtain 13 category feature value results.
Step 403, characteristic value is input in effective grader and detected.
13 category feature values were added to be input in validity grader and detected;Validity recognition result is exported, as a result
Display:This feature value is effective or invalid, and exports detailed basis for estimation.
The advantage of the invention is that:
1) a kind of, recognition methods of the use THDS system result of detection validity of feature based value detection of the invention, can
The validity of automatic identification THDS equipment result of detections, without manual intervention;
2) a kind of, recognition methods of the use THDS system result of detection validity of feature based value detection of the invention, can
The validity of Real time identification THDS equipment result of detections, amount of calculation is small, and THDS equipment normal works are not influenceed;
3), a kind of recognition methods of the use THDS system result of detection validity of feature based value detection of the invention, warp
The test of long-time mass data is crossed, discrimination is high, no wrong report.
Brief description of the drawings
Fig. 1 is the recognition methods flow chart for the THDS system result of detection validity that feature based value of the present invention is detected.
Fig. 2 carries out the flow chart of Effective judgement for the present invention to 32 voltage value datas collected.
Embodiment
The present invention relates to the recognition methods that railway truck bearing temperature sensing result validity is carried out using THDS systems, carry
The identification to carrying out railway truck bearing temperature sensing result validity using THDS systems detected for a kind of feature based value
Method.Detailed process has including carrying out feature extraction to training sample and generating validity grader and application class device
Effect property deterministic process.
Comprise the following steps that:
Step 1: choosing the invalid training sample data of measuring temperature of bearing value, 32 all voltage value datas are returned
One change is handled;
In order to improve the accuracy of validity grader, the sample of selection is on October 8,1 day to 2014 January in 2013
The bearing data of national all forecast thermal levels of goods train, whole sample datas totally 9952.Wherein, the strong bearing for swashing heat is forecast
Data are 299, and the actual data sample for making final thermometric value invalid due to the influence of the factors such as interference has 114, by this 114
Bar is used as training sample.Remaining 185 sample datas are referred to as normal sample.
Each training sample data include:Thermal level, THDS device types, a sequence, axle position, inside and outside spy mark, vehicle, 32 points
Voltage max, 32 voltage minimums, 32 magnitudes of voltage and the too high mark of day null value.
32 magnitudes of voltage of data are normalized, the interval numerical value in 32 obtained 0~100 forms oscillogram,
Represent 32 waveforms of axle temperature.
Wherein 32 magnitudes of voltage are marked as 0-31, and 0 is beginning label, and 31 be termination label;Each label corresponds to one
Numerical value after individual normalized.
Step 2: extracting characteristic value to the training sample after each normalized, characteristic data set is formed;
Each training sample has multiple characteristic values, and characteristic value includes following several:Flat-top points characteristic value;Waveform widths are special
Value indicative;Saturation width characteristics value;Saturation detection characteristic value;Line segment detection characteristic value;Energy eigenvalue;Sharp thermoelectricity forces down mark;
Trailing edge points maximum;The difference of voltage maximin;The count feature value of line segment folding;Ascending tendency line segment quantity;Tank
Car, boxcar sunlight interference mark and specific characteristic value.
1st, flat-top points characteristic value:
After 32 magnitudes of voltage in each training sample are normalized, first obtained greatest measure
The continuous points more than 95 of point and its left and right sides, are used as flat-top points characteristic value;
For example:First greatest measure point is 100, and numerical value has 2 more than 95 corresponding continuous labels after its left side is normalized
Individual, its right scale has 1 point more than 95 corresponding continuous labels, then the characteristic value of flat-top points is 4.
2nd, waveform widths characteristic value:
After 32 magnitudes of voltage in each training sample are normalized, first obtained greatest measure
Point and its left and right sides continue to exceed 80 points, are used as waveform widths characteristic value;
For example:First greatest measure point is 100, and numerical value continuously has 2 more than 80 corresponding labels after its left side is normalized
Individual, there is 0 point on its right side more than 80 corresponding continuous labels, then the characteristic value of the waveform widths is 3.
3rd, saturation width characteristics value:
When 32 voltage max in training sample are more than 9990 millivolts but not less than 10000 millivolts, by 32 points of electricity
After the normalization of pressure value, corresponding first greatest measure point and its left and right sides continue to exceed 90 points, are used as saturation width
Characteristic value;
The saturation width characteristics value result is numeral.
4th, saturation detection characteristic value;
It is less than 20 points after being normalized when what this feature value was specifically detected is saturation:
When 32 voltage max in training sample are more than 9990 millivolts but not less than 10000 millivolts, by 32 points of electricity
After the normalization of pressure value, statistic is less than 20 points, is used as the points characteristic value for being less than 20 after being normalized during saturation, this feature
It is numeral to be worth result;
5th, Line segment detection characteristic value:
This feature value includes:Long line segment quantative attribute value, rising edge characteristics value and flat-top position indicate rearward;
, it is necessary to which the judgement of advanced line section tendency, obtains line segment array before extraction Line segment detection characteristic value, spy is being carried out
The extraction of value indicative.
Specifically line segment tendency determination methods are:
The first step:After 32 magnitudes of voltage are normalized, 32 oscillograms are judged:
1), since first label, continuous two labels are chosen successively, and corresponding numerical value after normalization is made poor,
If the difference of latter numerical value and previous numerical value is more than 6, it is ascending tendency to set at this 2 points;If instead previous number
Value and the difference of latter numerical value are set to decline tendency more than 6;
After treatment, 32 magnitudes of voltage are divided into line segment between any two, 31 altogether, respectively marked as 1-31, i.e.,
The corresponding start-stop of line segment 1 marked as 0 and 1, the like.
2), if the difference of latter numerical value and previous numerical value is no more than behind 6, but addition thirdly, 3 points corresponding
Numerical value continuously increases, and thirdly and first point of corresponding numerical value difference be more than or equal to 10, then set this 3 points be ascending tendency;
Conversely, after adding thirdly, 3 points of corresponding numerical value are continuously reduced, and first point with the thirdly difference of numerical value more than or equal to 10,
Then set at this 3 points to decline tendency.
If 3), be unsatisfactory for above-mentioned two condition, it is horizontal tendency to set at this 2 points;
Second step:After the first step, if continuous 3 line segments meet following condition:Line segment 1 is identical with the tendency of line segment 3, line
Section 2 is horizontal tendency line segment that length is 1, then by line segment 2 be set to line segment 1 and the identical tendency of line segment 3, by three line segments
Merge into a line segment;Length is that 1 to refer to that the termination label of line segment 2 subtracts the value of beginning label be 1;
3rd step:If certain the start-stop label for rising or falling line segment, the voltage difference of corresponding points is no more than 50 millivolts, then
It is horizontal tendency to set the line segment tendency, that is, has filtered the fluctuating that industry disturbance information is likely to result in;
4th step:Merge continuous and tendency identical line segment;
51st, long line segment quantative attribute value:
Line segment sum is counted, it is 1 and line segment of the amplitude less than 8 to ignore length, is used as long line segment quantative attribute value;Amplitude is
Refer to the absolute value of the difference of numerical value after the magnitude of voltage normalization at line segment two ends;The characteristic value result of long line segment quantity is numeral.
52nd, rising edge characteristics value:
Significantly rising edge indication after this feature value refers at 10 points:
After beginning label is 10 point, if the rising edge that amplitude is more than 50 occurs in starting point and ending point, this is instructed
Significantly rising edge indication is set to 1 after 10 points of white silk sample, is otherwise set to 0.
53rd, flat-top position indicates rearward:
In line segment, if the beginning label of first horizontal tendency line segment is more than after 12 and normalization, numerical value is more than 90, by this
Training sample mark is set to 1, is otherwise set to 0;
6th, energy eigenvalue:Use 32 point value sums after normalization;The result of energy is located between 0-3200.
7th, swash thermoelectricity and force down mark:When thermal level is swashs heat, when 32 voltage max are not less than -3 volt, training sample is set
In sharp thermoelectricity force down flag sign value for 1, be otherwise provided as 0.
8th, trailing edge points maximum;
According to the judgement of line segment tendency, the line segment beginning label of trailing edge and termination label are made into poor, absolute value as a result
For trailing edge points maximum;
9th, the difference of voltage maximin;
32 voltage max of data in each training sample make poor with 32 voltage minimums, are used as this feature
Value.
10th, the count feature value of line segment folding:
It is raising and lowering or decline and rising by continuous two lines section tendency, folding, which is counted, Jia one;Continuous three line segments
In, line segment 1,3 tendencies are raising and lowering or decline and rising, and the tendency of line segment 2 is horizontal tendency, and folding, which is counted, Jia one;
For example:Four line segments are respectively rising, transverse direction, decline, transverse direction and rising, then the count feature value of line segment folding is 2.
11st, ascending tendency line segment quantity;
This feature value is the ascending tendency line segment quantity of first 5 points of statistics;
According to the tendency of line segment, by the line segment between first 5 points marked as 0-4, the line segment quantity of ascending tendency is counted.Example
Such as:If label is rising edge always between 0 to 4, this feature value is 1;If line segment for laterally, rise, laterally and on
The line segment risen, then this feature value is 2.
12nd, tank car, boxcar sunlight interference mark:
First three point value is all higher than 80 after vehicle is G or X, and normalization, and there is day null value high standard will, and axle position is
When 1 axle or 3 axle, it is 1 to put this flag bit, is otherwise set to 0;
13rd, specific characteristic value:
This feature value refers to:Special circuit, special acquisition station, the processing of extraordinary vehicle and other external interferences;Specifically
2nd, 4 axial extent characteristic value:
When acquisition station is tries to win the champion or tea room, and axle position calculates eigen value when being 2 or 4 axle, calculate greatest measure point 100 and
The quantity of the continuous point for being more than or equal to 90 in its left and right.
13rd is expansible item, it is expansible it is more for special circuit, special acquisition station, extraordinary vehicle and other outside
The characteristic value of boundary's interference is calculated.
Characteristic value selected by the present invention be not limited only to it is above-mentioned, if being found that features described above in actual use
It is invalid characteristic value that beyond value, can explicitly indicate that " 32 waveforms ", then these characteristic values are included in the present invention's
In the range of.
Step 3: being generated using the characteristic data set of training sample and training 32 waveform validity graders.
Whole training samples are extracted after characteristic value, by the analytic induction to training sample characteristic value, finding out can be accurate
Really distinguish training sample and the validity classifier parameters of normal sample.
According to the characteristic value extracted in step 2, multiple graders are created, each grader uses one in step 2
Or multiple characteristic values are used as input.
In the present invention, with reference to the various features value in step 2, the grader of establishment is as follows:
The quantity of grader is not limited only to upper table, and because the 13rd is expansible item in characteristic value, therefore grader can be with
The increase of characteristic value project and increase;And may be due to increasing new grader, and bar is judged to the grader in upper table
Part makes corresponding modification.
Step 4: carrying out Effective judgement to 32 waveforms collected.
Detailed process is as follows:
Step 401,32 point datas collected are normalized;
By collect 32 magnitudes of voltage, it is normalized according to step one.
Step 402, to after normalized numerical value extract characteristic value;
To the numerical value after normalization, the data of combined training sample:It is thermal level, THDS device types, a sequence, axle position, inside and outside
Visit mark, vehicle, 32 voltage max, 32 voltage minimums, 32 magnitudes of voltage and the too high mark of day null value;Respectively according to
Characteristic value in step 2 is calculated the data collected, obtains 13 category feature value results.
Step 403, characteristic value is input in effective grader and detected.
13 category feature values were added to be input in validity grader and detected;Validity recognition result is exported, as a result
Including:Effective judgement result and basis for estimation, wherein Effective judgement result output information include effectively and invalid two kinds, sentence
Disconnected is the Rule of judgment of the grader and the value of input parameter according to output information.
As result is:Flat-top detection, the characteristic value serial number used:1. flat-top is counted;It is invalid to be judged to, reason:Plat topping point
Number is less than 3.
When grader carries out Effective judgement to some characteristic value, during output null result, validity can be directly exported
Judged result is invalid, and exports basis for estimation, or proceeds the judgement of other graders, exports complete basis for estimation,
So that user checks.
Integrated use of the present invention big data digging technology, realizes the auto judge to THDS result of detection validity,
Avoid because external interference causes the phenomenon of reporting hot axle by mistake, to improve the encashment ratio of THDS systems, ensure train safety, it is normal,
Operation plays an important role on time.
Claims (7)
1. a kind of recognition methods of the THDS system result of detection validity of feature based value detection, it is characterised in that:Including with
Lower step:
Step 1: choosing the invalid training sample data of measuring temperature of bearing value, 32 all voltage value datas are normalized
Processing;
Step 2: extracting characteristic value to the training sample after each normalized, characteristic data set is formed;
Step 3: being generated using the characteristic data set of training sample and training 32 waveform validity graders;
Step 4: utilizing validity grader progress Effective judgement to 32 waveforms newly collected;
Detailed process is as follows:
Step 401,32 voltage value datas newly collected are normalized;
Step 402, to after normalized numerical value extract characteristic value;
Step 403, characteristic value is input in effective grader and detected.
2. a kind of recognition methods of the THDS system result of detection validity of feature based value detection as claimed in claim 1,
It is characterized in that:Characteristic value includes in described step two:Flat-top points characteristic value, waveform widths characteristic value, saturation width are special
It is maximum that value indicative, saturation detection characteristic value, Line segment detection characteristic value, energy eigenvalue, sharp thermoelectricity force down mark, trailing edge points
Value, the difference of voltage maximin, line segment folding count feature value, ascending tendency line segment quantity, specific characteristic value and tank car,
Boxcar sunlight interference mark;
The method for extracting flat-top points characteristic value is as follows:After 32 magnitudes of voltage are normalized, first obtained maximum number
Value point and its left and right sides continuously exceed threshold value a points, are used as flat-top points characteristic value;
The method for extracting waveform widths characteristic value is as follows:After 32 magnitudes of voltage are normalized, first obtained maximum number
Value point and its left and right sides continue to exceed threshold value a points, are used as waveform widths characteristic value;
The method for extracting saturation width characteristics value is as follows:When 32 voltage max are more than 9990 millivolts but not less than 10000 millis
Fu Shi, corresponding first greatest measure point and its left and right sides continue to exceed threshold value a points after normalization, wide as saturation
Spend characteristic value;
The method for extracting saturation detection characteristic value is as follows:When 32 voltage max in training sample more than 9990 millivolts but not
During more than 10000 millivolts, after 32 magnitude of voltage normalization, statistic is less than threshold value b points, is used as saturation detection feature
Value;
The method for extracting Line segment detection characteristic value is as follows:
Line segment detection characteristic value includes:Long line segment quantative attribute value, rising edge characteristics value and flat-top position indicate rearward;
The method for extracting long line segment quantative attribute value is as follows:
In line segment array using the formation of line segment tendency determination methods, statistics line segment sum, it is 1 to ignore line segment two ends label difference
And amplitude is less than threshold value c line segment, is used as long line segment quantative attribute value;Amplitude is counted after referring to the magnitude of voltage normalization at line segment two ends
The absolute value of the difference of value;
The method for extracting rising edge characteristics value is as follows:
Using line segment tendency determination methods formation line segment array in, beginning label for threshold value d point after, if starting point and
There is the rising edge that amplitude is more than threshold value e in terminating point, and characteristic value is set into 1,0 is otherwise set to;
Extracting flat-top position, denotation approach is as follows rearward:
In line segment array using the formation of line segment tendency determination methods, if the beginning label of first horizontal tendency line segment is more than threshold
Value f and normalization after numerical value exceed threshold value a;The training sample mark is set to 1,0 is otherwise set to;
Extract energy eigenvalue method as follows:Use 32 point value sums after normalization;
Extract sharp thermoelectricity and force down mark:When thermal level is swashs heat, when 32 voltage max are not less than -3 volt, this is set to be masked as
1, it is otherwise provided as 0;
Extract the count feature value of line segment folding:
It is raising and lowering or decline and rising by continuous two lines section tendency, folding, which is counted, Jia one;In continuous three line segments, the
One line segment, Article 3 line segment tendency are raising and lowering or decline and rising, and Article 2 line segment tendency is horizontal tendency, folding
Counting Jia one;
Extract tank car, boxcar sunlight interference denotation approach as follows:
First three point value is all higher than threshold value a after vehicle is G or X, and normalization, and there is day null value high standard will, and axle position is 1
When axle or 3 axle, it is 1 to put this flag bit, is otherwise set to 0;
Extract specific characteristic value method:Specific characteristic value is 2,4 axial extent characteristic values;
Eigen value is calculated when acquisition station axle position is 2 or 4 axle, greatest measure point and its left and right is calculated and is continuously more than or equal to threshold value
The quantity of a point.
3. a kind of recognition methods of the THDS system result of detection validity of feature based value detection as claimed in claim 1,
It is characterized in that:
Validity grader includes in the step 3:Flat-top detects grader, waveform widths detection grader, the inspection of saturation width
Grader is surveyed, polarization detection grader during saturation, long line segment quantity detects grader, rising edge detection grader, flat-top
Grader, energy measuring grader are detected rearward, and sharp thermoelectricity forces down detection grader, trailing edge points maximum detection classification
Device, voltage magnitude detection grader, line segment folding number of times detection grader, rising edge tendency detection grader, sunlight interference detection
Grader and 2,4 axial extents detection grader;
Described flat-top detection grader, is judged using flat-top points characteristic value, is more than when flat-top points characteristic value is met
During equal to threshold value a', output is effective, otherwise invalid;
Described waveform widths detection grader, is judged using waveform widths characteristic value, when axle position is 1 axle or 3 axle,
Waveform widths characteristic value is more than threshold value b', and output is effective, otherwise invalid;When axle position is 2 axles or 4 axle, waveform widths feature
Value is more than threshold value c', and output is effective, otherwise invalid;
Described saturation width detection grader, is judged using saturation width characteristics value;Saturation width be more than 0 and for 1 axle,
During 3 axle, saturation width characteristics value is more than or equal to threshold value d', and output is effective, otherwise invalid;Saturation width is more than 0 and is 2 axles, 4 axles
When, saturation width characteristics value is more than or equal to threshold value c', and output is effective, otherwise invalid;There is output when saturation width characteristics value is equal to 0
Effect;
Polarization detection grader, is carried out jointly using saturation width characteristics value and saturation detection characteristic value during described saturation
Judge, when 60% × (32- saturation width characteristics value) is less than saturation detection characteristic value, grader output is effective, otherwise output
It is invalid;
Described long line segment quantity detection grader, is judged using long line segment quantative attribute value, when this feature value meets threshold
During value f', it is judged as effectively;Otherwise to be invalid;
Described rising edge detection grader, is judged using rising edge characteristics value, is judged as having when this feature value is 0
Effect is invalid when being 1;
Described flat-top detects grader rearward, is indicated rearward using flat-top position and is judged, judges when this feature value is 0
It is invalid when being 1 to be effective;
Described energy measuring grader, is judged using energy eigenvalue, works as satisfaction:Exist when different in line segment rise and
Decline line segment, energy eigenvalue is effective;Or there is raising and lowering line segment simultaneously in line segment, and vehicle is when being C70E, energy
Meet threshold value g';Energy eigenvalue is effective;Or there is raising and lowering line segment simultaneously in line segment, and during the non-C70E of vehicle, energy
Amount meets threshold value g ", and grader output is effective;Otherwise it is invalid;
Described sharp thermoelectricity forces down detection grader, and forcing down mark using sharp thermoelectricity is judged, the mark when swashing thermal level
For 0 or thermal level be non-sharp hot when, grader output is effective, conversely, when thermal level is swashs heat, sharp thermoelectricity, which is forced down, is masked as 1,
Grader output is invalid;
Described trailing edge points maximum detection grader, is judged using trailing edge points maximum, if this feature
Value meets threshold value h', and output is effective, otherwise invalid;
Described voltage magnitude detection grader, the difference using voltage maximin is judged, if this feature value is big
In 200 millivolts, grader output is effective, and it is invalid otherwise to export;
Described line segment folding number of times detection grader, the count feature value rolled over using line segment is judged, if being non-C80E cars
Type, it is effective that this feature value is less than or equal to threshold value i' outputs;If C80E vehicles, this feature value is less than or equal to threshold value i ", output
Effectively;Otherwise it is invalid;
Described rising edge tendency detection grader, is judged using ascending tendency line segment quantity, if this feature value is less than
Equal to threshold value i', grader output is effective, and it is invalid otherwise to export;
Described sunlight interference detection grader, is judged that this is masked as 0 output using tank car, boxcar sunlight interference mark
Effectively, this to be masked as 1 output invalid;
Described 2,4 axial extents detection grader, is judged using 2 in specific characteristic value, 4 axial extent characteristic values, when this
Characteristic value is more than threshold value c' outputs effectively, and it is invalid otherwise to export.
4. a kind of recognition methods of the THDS system result of detection validity of feature based value detection as claimed in claim 2,
It is characterized in that:The determination methods of described line segment tendency are as follows:
The first step:After 32 magnitudes of voltage are normalized, 32 oscillograms are judged:
1), since first label, continuous two labels are chosen successively, and numerical value after normalization is made poor, if latter
The difference of numerical value and previous numerical value exceedes threshold value g, then it is ascending tendency to set at this 2 points;If instead previous numerical value is with after
The difference of one numerical value exceedes threshold value g, is set to decline tendency;
2), if the difference of latter numerical value and previous numerical value is no more than after threshold value g, but addition thirdly, 3 points corresponding
Numerical value continuously increases, and thirdly and first point of corresponding numerical value difference be more than or equal to threshold value c, then set walked for rising at this 3 points
Gesture;Conversely, being then set to decline tendency;
If 3), be unsatisfactory for above-mentioned two condition, it is horizontal tendency to set at this 2 points;
Second step:After the first step, if continuous 3 line segments meet following condition:First line segment and Article 3 line segment tendency
Identical, Article 2 line segment is the horizontal tendency line segment that line segment two ends label difference is 1, then is set to Article 2 line segment and first
Three line segments are merged into a line segment by bar line segment and Article 3 line segment identical tendency;
3rd step:If certain the start-stop label for rising or falling line segment, the voltage difference of corresponding points is no more than 50 millivolts, then sets
The line segment tendency is horizontal tendency;
4th step:Merge continuous and tendency identical line segment.
5. a kind of recognition methods of the THDS system result of detection validity of feature based value detection as claimed in claim 2,
It is characterized in that:Described threshold value a scope is 80-100;Threshold value b scopes are:20-35;Threshold value c scopes are:8-10;Threshold value d
Scope is:6-10;Threshold value e scopes are:40-50;Threshold value f scopes are:8-12.
6. a kind of recognition methods of the THDS system result of detection validity of feature based value detection as claimed in claim 4,
It is characterized in that:Described threshold value g range is:4-6.
7. a kind of recognition methods of the THDS system result of detection validity of feature based value detection as claimed in claim 3,
It is characterized in that:Described threshold value a' scopes are:2-4;Threshold value b' scopes are:2-5;Threshold value c' scopes are:5-7;Threshold value d' is excellent
Scope is:4-7;Threshold value f' scopes are:More than 1 and less than or equal to 7;Threshold value g' scopes are:More than or equal to 500 and less than 2500;
Threshold value g " scopes are:More than or equal to 500 and less than 2200;Threshold value h' scopes are:Less than 15;Threshold value i' scopes are 0 to 2;Threshold value
I " scopes are:2 to 4.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410837264.2A CN104679982B (en) | 2014-12-29 | 2014-12-29 | A kind of recognition methods of the THDS system result of detection validity of feature based value detection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410837264.2A CN104679982B (en) | 2014-12-29 | 2014-12-29 | A kind of recognition methods of the THDS system result of detection validity of feature based value detection |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104679982A CN104679982A (en) | 2015-06-03 |
CN104679982B true CN104679982B (en) | 2017-08-08 |
Family
ID=53315015
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410837264.2A Active CN104679982B (en) | 2014-12-29 | 2014-12-29 | A kind of recognition methods of the THDS system result of detection validity of feature based value detection |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104679982B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107179455B (en) * | 2017-04-27 | 2019-04-23 | 华中科技大学 | A kind of real-time electrical appliance recognition and system based on edge features vector model |
CN117558140B (en) * | 2024-01-11 | 2024-04-12 | 四川九通智路科技有限公司 | Traffic flow detection method for double-lane tunnel |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0979915A (en) * | 1995-09-14 | 1997-03-28 | East Japan Railway Co | System for detecting abnormal temperature of bearing of railcar |
CN102837711A (en) * | 2011-06-21 | 2012-12-26 | 中国铁道科学研究院机车车辆研究所 | Infrared waveform based intelligent identification method for railway bearing |
CN102982684A (en) * | 2012-11-28 | 2013-03-20 | 深圳市迈科龙影像技术有限公司 | Vehicle model recognition method and system |
-
2014
- 2014-12-29 CN CN201410837264.2A patent/CN104679982B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0979915A (en) * | 1995-09-14 | 1997-03-28 | East Japan Railway Co | System for detecting abnormal temperature of bearing of railcar |
CN102837711A (en) * | 2011-06-21 | 2012-12-26 | 中国铁道科学研究院机车车辆研究所 | Infrared waveform based intelligent identification method for railway bearing |
CN102982684A (en) * | 2012-11-28 | 2013-03-20 | 深圳市迈科龙影像技术有限公司 | Vehicle model recognition method and system |
Also Published As
Publication number | Publication date |
---|---|
CN104679982A (en) | 2015-06-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104049244B (en) | The radar velocity measurement machine carbon brush abrasion recognition methods analyzed based on temporal signatures value | |
CN106960285B (en) | Subway train operation service quality detection device and method | |
CN103809163B (en) | A kind of Radar for vehicle object detection method based on local maximum | |
CN104731083B (en) | A kind of industrial method for diagnosing faults and application based on self-adaptive feature extraction | |
CN104677997B (en) | A kind of transformer oil chromatographic on-line monitoring differentiation method for early warning | |
CN103103570B (en) | Based on the aluminium cell condition diagnostic method of pivot similarity measure | |
CN103455820A (en) | Method and system for detecting and tracking vehicle based on machine vision technology | |
CN102721397A (en) | Method for extracting road surface characteristic parameters based on modern time series of vertical dynamic load | |
CN108710637A (en) | Taxi exception track real-time detection method based on time-space relationship | |
CN104679982B (en) | A kind of recognition methods of the THDS system result of detection validity of feature based value detection | |
CN102658298A (en) | Plate-shape quality online judgment method applicable to hot-rolled thin strip steel | |
CN105675274A (en) | Time-domain parameter and D-S evidence theory-based rotor running state monitoring method | |
CN106772656A (en) | A kind of indoor human body detection method based on infrared sensor array | |
CN106407555A (en) | Accelerated degradation data analysis method based on principle of invariance of accelerating factor | |
CN110223522A (en) | A kind of vehicle location recognition methods based on three axis geomagnetic sensors | |
CN115600067A (en) | Building structure deformation monitoring system and method based on Internet of things | |
CN105572143A (en) | Method for detecting periodic defect of calendered material surface in calendering process | |
CN103063674B (en) | Detection method for copper grade of copper block, and detection system thereof | |
Califano et al. | Heavy rainfall temporal characterization in the peri-urban Solofrana river basin, Southern Italy | |
CN117518982B (en) | Method and system for improving machining precision of machine tool | |
CN103217590A (en) | Method automatically acquiring atmospheric electric field thunder and lightning early-warning characteristic parameter threshold values | |
CN105260814A (en) | Power transmission and transformation equipment evaluation model and processing method based on big data | |
CN107330264A (en) | A kind of verification method of bridge monitoring data reliability | |
CN102004076A (en) | Method and system for detecting foreign fiber in ginned cotton | |
CN105447511B (en) | A kind of SVM object detection method based on Adaboost Haar-Like feature |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CP03 | Change of name, title or address |
Address after: 9th floor, No. 61 Zhichun Road, Haidian District, Beijing, 100190 Patentee after: Beijing Aerospace Shenzhou Intelligent Equipment Technology Co.,Ltd. Address before: 100080 No. 61, Haidian District, Beijing, Zhichun Road Patentee before: BEIJING CTROWELL INFRARED TECHNOLOGY Co.,Ltd. |
|
CP03 | Change of name, title or address |