CN106976468A - A kind of switch breakdown diagnostic method based on DWT and C SVM - Google Patents

Abstract

The invention discloses a kind of switch breakdown diagnostic method based on DWT and C SVM.This method comprises the following steps：First, it is determined that the track switch type to be diagnosed, selects the track switch of a same type and install current transducer, common fault mode is simulated on the track switch, and gather some groups of electric current CRANK PULSES under every kind of fault mode respectively as simulated failure signal；Secondly, current transducer is installed on the track switch being diagnosed, its electric current CRANK PULSES is gathered as measured signal；Then DWT is carried out to simulated failure signal and measured signal, and be normalized, and the simulated failure characteristic signal after processing is input in C svm classifier models be trained and parameter optimization；Finally, the actual measurement characteristic signal after processing is input to the C SVM models that training is completed, carries out last Fault Identification.The present invention has the advantages that cost is low, engineering construction is simple and convenient and rate of correct diagnosis is high.

Description

A kind of switch breakdown diagnostic method based on DWT and C-SVM

Technical field

The present invention relates to railway switch state-detection field, particularly a kind of switch breakdown diagnosis based on DWT and C-SVM Method.

Background technology

Track switch connects different tracks and is typically mounted between two strands or multiply track..Due to by itself complicated machine Tool structure and the restriction of executing agency, the mechanical strength of track switch each several part are generally below the equipment on circuit, therefore mechanical structure Easily occur fatigue change, trigger the excessive situation in gap between point tongue and stock rail, so as to cause train to occur to squeeze trouble Or even derailment accident.At the same time, track switch is generally mounted to outdoor, and its working environment is larger by inside even from weather, such as strong wind Weather can cause to block debris between point tongue and stock rail so as to jam, and sleety weather causes slide plate by abnormal resistance So as to influence track switch to change, these are likely to the potential risk as train traffic safety.

At present, major railways still rely on traditional periodic preventative detection and artificial scheduled overhaul to complete track switch equipment The detection of running status, therefore field maintenance workman can not have found the track switch of generation failure the very first time, artificial detection pair in addition The skill requirement of maintainer is higher, often occurs new employee because of the situation for wrong diagnosis of lacking experience.In order to solve this problem, mesh Track switch equipment microcomputer monitoring equipment is installed on some preceding rail tracks, electric current that the equipment can be acted according to collection track switch and The signals such as power are changed, and are integrated with the fault diagnosis software based on threshold decision in a device to realize fault alarm.But, road The complicated and severe operation working environment in trouble scene causes in microcomputer monitoring equipment default threshold value in turnout work for a period of time Reference value is just lost afterwards, in addition, these threshold values are typically also to be set by maintenance expert, the ambiguity of expertise also causes it The threshold rule of summary can not for a long time work and effectively be promoted.

The content of the invention

It is an object of the invention to provide the track switch that a kind of cost is low, engineering construction is simply and easily based on DWT and C-SVM Method for diagnosing faults, realizes the Fault Identification to track switch.

The technical solution for realizing the object of the invention is：A kind of switch breakdown diagnostic method based on DWT and C-SVM, Comprise the following steps：

Step 1, it is determined that the track switch type to be diagnosed, selects the track switch of a same type, and installed on selected track switch Current transducer；

Step 2, multiple fault modes are simulated on the track switch, and gather the operating of the electric current under every kind of fault mode respectively 20~30 groups of signal；

Step 3, current transducer is installed on the track switch being diagnosed, collection is diagnosed the electric current CRANK PULSES 5-10 of track switch Group；

Step 4, it is DWT to carry out wavelet transform to simulated failure electric current CRANK PULSES and measured current CRANK PULSES, And be normalized；

Step 5, the simulated failure electric current CRANK PULSES after processing is input to non-linear soft margin support vector machine i.e. C- It is trained in svm classifier model, and carries out parameter optimization；

Step 6, the measured signal after processing is input to the C-SVM disaggregated models that training is completed, carries out last failure Identification.

Further, fault mode described in step 2 includes difficult unblock, conversion step resistance, conversion sawtooth resistance and lock Close difficulty.

Further, the wavelet basis function selected in the wavelet transform described in step 4 is Haar small echos, Haar small echos Expression be：

In formula, ψ is Haar small echos, and t represents the time.

Further, the normalized described in step 4, using linear normalizing, the specific formula of linear normalizing is as follows：

In formula, x_normFor the signal value after normalization, x_maxFor maximum signal level, x_minFor minimum signal value, x is normalizing Signal value before change.

Further, the C-SVM disaggregated models selection RBF kernel functions K described in step 5, specific formula is as follows：

In formula, x_iInput vector is represented, x represents map vector, σ²Represent input vector variance.

Compared with prior art, its remarkable advantage is the present invention：(1) cost is low, it is to avoid the operation of microcomputer detecting system, Maintenance cost；(2) current transducer installs simple and convenient, and adapts to the severe working environment in scene；(3) diagnostic result is credible Degree is higher, and effective guidance can be provided for the maintenance of track switch.

Brief description of the drawings

Fig. 1 is the flow chart of the switch breakdown diagnostic method of the invention based on DWT and C-SVM.

Fig. 2 is sensor scheme of installation in the present invention, wherein (a) is front view, (b), to overlook threading figure, (c) is to bow View.

Fig. 3 is the electric current CRANK PULSES figure under 5 kinds of states of track switch in embodiment, wherein being the electricity under (a) unfaulty conditions CRANK PULSES figure is flowed, (b) is the electric current CRANK PULSES figure of failure 1, and (c) is the electric current CRANK PULSES figure of failure 2, and (d) is failure 3 Electric current CRANK PULSES figure, (e) be failure 4 electric current CRANK PULSES figure.

Fig. 4 is the characteristic signal figure under 5 kinds of states of track switch in embodiment, wherein being the feature letter under (a) unfaulty conditions Number figure, (b) be failure 1 characteristic signal figure, (c) be failure 2 characteristic signal figure, (d) be failure 3 characteristic signal figure, (e) For the characteristic signal figure of failure 4.

Fig. 5 is the normalizing characteristic signal figure under 5 kinds of states of track switch in embodiment, wherein being returning under (a) unfaulty conditions One characteristic signal figure, (b) is the normalizing characteristic signal figure of failure 1, and (c) is the normalizing characteristic signal figure of failure 2, and (d) is failure 3 Normalizing characteristic signal figure, (e) be failure 4 normalizing characteristic signal figure.

Fig. 6 is switch breakdown diagnostic result figure in embodiment.

Embodiment

Below in conjunction with the accompanying drawings and specific embodiment is described in further detail to the present invention.

With reference to Fig. 1, the switch breakdown diagnostic method of the invention based on DWT and C-SVM is comprised the steps of：

Step 1, it is determined that the track switch type to be diagnosed, selects the track switch of a same type, and installed on selected track switch Current transducer；

Step 2, multiple fault modes are simulated on the track switch, and gather the operating of the electric current under every kind of fault mode respectively 20~30 groups of signal；It is difficult that the fault mode includes difficult unblock, conversion step resistance, conversion sawtooth resistance and locking；

Step 3, current transducer is installed on the track switch being diagnosed, collection is diagnosed the electric current CRANK PULSES 5-10 of track switch Group；

Step 4, simulated failure electric current CRANK PULSES and measured current CRANK PULSES are carried out wavelet transform (DWT, Discrete Wavelet Transform), and be normalized；

The wavelet basis function selected in described wavelet transform is Haar small echos, the expression of Haar small echos For：

In formula, ψ is Haar small echos, and t represents the time.

Described normalized, using linear normalizing, the specific formula of linear normalizing is as follows：

In formula, x_normFor the signal value after normalization, x_maxFor maximum signal level, x_minFor minimum signal value, x is normalizing Signal value before change.

Step 5, the simulated failure electric current CRANK PULSES after processing is input to non-linear soft margin support vector machine (C- SVM, C-Support Vector Machine) it is trained in disaggregated model, and carry out parameter optimization；

Described C-SVM disaggregated models selection RBF kernel functions K, specific formula is as follows：

In formula, x_iInput vector is represented, x represents map vector, σ²Represent input vector variance.

Step 6, the measured signal after processing is input to the C-SVM disaggregated models that training is completed, carries out last failure Identification.

With reference to specific embodiment, the present invention is described in further detail.

Embodiment 1

The present embodiment emphasis carries out fault diagnosis to certain ZD6-A types track switch of certain MTR's track laying, and in statistics After the historical failure data of the type track switch, the fault type of diagnosis is set as four, respectively (unblock is tired for failure 1 It is difficult), failure 2 (conversion step resistance), failure 3 (conversion sawtooth resistance) and failure 4 (locking difficulty).

In order to gather the failure CRANK PULSES that can train C-SVM models, one is randomly choosed on MTR's circuit Individual ZD6-A types track switch and the HK-D4I type current transducers for installing the industrial production of China's control, as shown in Fig. 2 wherein (a) is to face Figure, (b), to overlook threading figure, (c) is top view；Next allows maintenance engineering teacher to simulate above-mentioned four kinds of failures on the track switch, Each 30 groups of electric current CRANK PULSES under every kind of malfunction is gathered respectively, and frequency acquisition is set to 5K, and the sampling time is 4S.Consider It is possible to do not occur any failure to track switch is diagnosed, therefore equally gathers electric current CRANK PULSES of the track switch under unfaulty conditions 30 groups.Electric current CRANK PULSES under the 5 kinds of states collected is as shown in figure 3, be wherein the electric current operating under (a) unfaulty conditions Signal graph, (b) is the electric current CRANK PULSES figure of failure 1, and (c) is the electric current CRANK PULSES figure of failure 2, and (d) is the electric current of failure 3 CRANK PULSES figure, (e) is the electric current CRANK PULSES figure of failure 4.

After simulated failure signal acquisition terminates, current transducer is equally installed in the ZD6-A type track switches being diagnosed, gathered 10 groups of its electric current CRANK PULSES.The 9 layer scattering small echos based on Haar small echos as shown in Figure 4 are carried out to all signals collected Conversion, wherein being the characteristic signal figure under (a) unfaulty conditions, (b) is the characteristic signal figure of failure 1, and (c) is the spy of failure 2 Signal graph is levied, (d) is the characteristic signal figure of failure 3, and (e) is the characteristic signal figure of failure 4；Then these characteristic signals are carried out Normalized as shown in Figure 5, wherein being the normalizing characteristic signal figure under (a) unfaulty conditions, (b) is the normalizing of failure 1 Characteristic signal figure, (c) is the normalizing characteristic signal figure of failure 2, and (d) is the normalizing characteristic signal figure of failure 3, and (e) is failure 4 Normalizing characteristic signal figure；Finally 150 groups of simulated failure normalizing characteristic signals are input in many disaggregated models of C-SVM and instructed Practice, Selection of kernel function parameter takes 2, and penalty coefficient C takes 0.25, gamma to take 4.

10 groups of characteristic signals of actual measurement are input to the model after training and carry out final fault diagnosis, the knot of fault diagnosis Fruit is as shown in Figure 6.It can be seen that except the 3rd group and the 8th group, the diagnostic result of remaining group characteristic signal is failure 2 (conversion step resistance), the Artificial Diagnosis result of the fault diagnosis result and field engineer are consistent.

To sum up, the present invention can recognize the failure occurred during turnout work in time, be that field apparatus maintenance worker carries For maintenance foundation, improve tracing trouble accuracy rate and solve failure efficiency so that ensure urban rail circuit traffic safety and Efficiency of operation.

Claims (5)

1. a kind of switch breakdown diagnostic method based on DWT and C-SVM, it is characterised in that comprise the following steps：

Step 1, it is determined that the track switch type to be diagnosed, selects the track switch of a same type, and install electric current on selected track switch Transmitter；

Step 2, multiple fault modes are simulated on the track switch, and gather the electric current CRANK PULSES under every kind of fault mode respectively 20~30 groups；

Step 3, current transducer is installed on the track switch being diagnosed, collection is diagnosed the electric current CRANK PULSES 5-10 groups of track switch；

Step 4, it is DWT to carry out wavelet transform to simulated failure electric current CRANK PULSES and measured current CRANK PULSES, is gone forward side by side Row normalized；

Step 5, the simulated failure electric current CRANK PULSES after processing is input to non-linear soft margin support vector machine i.e. C-SVM points It is trained in class model, and carries out parameter optimization；

Step 6, the measured signal after processing is input to the C-SVM disaggregated models that training is completed, carries out last Fault Identification.

2. the switch breakdown diagnostic method according to claim 1 based on DWT and C-SVM, it is characterised in that step 2 institute Stating fault mode includes unblock difficulty, conversion step resistance, conversion sawtooth resistance and locking difficulty.

3. the switch breakdown diagnostic method according to claim 1 based on DWT and C-SVM, it is characterised in that step 4 institute The wavelet basis function selected in the wavelet transform stated is Haar small echos, and the expression of Haar small echos is：

$\mathrm{\&psi;}=\left\{\begin{array}{cc}1& 0\&le;t\&le;0.5\\ -1& 0.5\&le;t<1\\ 0& else\end{array}\right.---\left(1\right)$

In formula, ψ is Haar small echos, and t represents the time.

4. the switch breakdown diagnostic method according to claim 1 based on DWT and C-SVM, it is characterised in that step 4 institute The normalized stated, using linear normalizing, the specific formula of linear normalizing is as follows：

$x_{norm}=\frac{x-x_{min}}{x_{max}-x_{min}}---\left(2\right)$

In formula, x_normFor the signal value after normalization, x_maxFor maximum signal level, x_minFor minimum signal value, x is normalization Preceding signal value.

5. the switch breakdown diagnostic method according to claim 1 based on DWT and C-SVM, it is characterised in that step 5 institute The C-SVM disaggregated models selection RBF kernel functions K stated, specific formula is as follows：

$K(x_i,x)=\exp \{-\frac{\left|\right|x_i-x|{|}^2}{{\mathrm{\&sigma;}}^2}\}---\left(3\right)$

In formula, x_iInput vector is represented, x represents map vector, σ²Represent input vector variance.