CN100464151C - Method for inspecting article surface vein and its sensor - Google Patents

Abstract

A method for detecting texture of object surface includes placing kynar piezoelectric film on pressure transducer, using said film to receive electric charge signal generated by pressing detected object on said film and using said transducer to receive pressure signal generated by pressing detected object on said film, picking up character value of electric charge after electric charge signal and pressure signal are amplified and applying trained GA-VIBP network to classify vector of said character value to obtain texture information of surface on detected object. The transducer used for realizing said method is also disclosed.

Description

Method for inspecting article surface vein and sensor thereof

Technical field

The invention belongs to sense of touch telepresenc technical field, is a kind of method for inspecting article surface vein and sensor thereof that is used for inspected object superficial makings situation.

Background technology

In general, touch is the highest can experience the concavo-convex of 1 to 3 micron on surface.Our mankind's haptic system is the set of the stereognosis ability, kinesthesia ability and the kinetism that combine with cognitive process, shows the bidirectional information passage of the uniqueness that connects brain.If can come out the tactile feedback information true reappearance and offer other parts of hand or health, will emerge a large amount of breathtaking application so.For example, studies show that and in virtual environment, provide real tactile sensation to help to increase work efficiency.The operator can therefrom extract useful information by touching virtual environment, and they are applied to wherein the knowledge of real world sensation.In general, the tactile data that obtains approaches true environment more, and the operator just can rely on the consciousness of real world more, is engaged in the task in the virtual environment.If in virtual environment, carry out the task training of true environment, the ability to work in true environment can be improved, and the serious consequence that causes owing to misoperation can not be produced, be similar to present virtual driving training.Therefore, the sense of touch telepresenc is not only significant in theory research, also has wide application value in the virtual reality of practical application and distant operation.

So-called sense of touch telepresenc, be meant the operator by the sense of touch detecting instrument touch, perception and manipulation is virtual or far object carry out the tactile data perception that a series of interactions obtain to characterize atural object bulk properties virtual or far away.Tactilely-perceptible mainly can be divided into two classes, and a class is the slip tactilely-perceptible, and another kind of is flexible (or rigidity) tactilely-perceptible.Wherein, slip tactilely-perceptible is meant the perception of people to the article surface vein characteristic.The present invention realizes one of gordian technique of slip sense of touch telepresenc, is the method that is used for inspected object superficial makings feature.

Be subjected to human limitation about the detection of texture information, can't set up the accurate model of texture so far, so that the progress of this respect gets is very slow texture essence understanding.Institute of robot of CMU proposes to set up the article surface vein model with probe in the body surface motion, and as shown in Figure 1,1 is probe, and 2 is direction of motion, and 3 is the amplitude of pinpoint movement track, and 4 is the pinpoint movement track, and 5 is the object under test surface.Probe 1 is perpendicular to object under test surface 5, and the power that applies a certain size makes probe 1 contact with object under test surface 5, and with constant motion, the degree of depth and the frequency of the superfine micro groove that the pinpoint movement track 4 of probe 1 can the reflection surface like this, indirect reflection the texture information of body surface, this is to detect the more common method of texture at present, and is simple and be convenient to operation.But, also there is very big deficiency, at first probe and body surface are to contact, movement locus is wire, can only reflect the concavo-convex information on the trajectory, even the dense degree of increase probe motion can not be included the 2 d texture information on whole surface; Secondly the size of probe tip is directly connected to the Oscillation Amplitude of the track of surveying; Once more, the detection method that probe drags at body surface may cause damage to soft object surfaces.The another kind of method that detects texture is according to the image processing theory, a series of image processing means such as utilization color image gray processing, medium filtering, image binaryzation, the texture information of the material that abstract image reflected.What Fig. 2 showed is the texture of ashtree plank.This method has only been measured the trend of texture, not too is suitably for the superficial makings modeling, and at first, the light source condition directly influences the shooting of picture, and the texture information that picture shot is extracted under Different Light can be different; Secondly, the texture information that extracts is a black and white binary image, depth information that can not gully, reflection surface.

Summary of the invention

The invention provides and a kind ofly can measure simple method for inspecting article surface vein of plane information and sensor thereof.

The present invention adopts following technical scheme:

The technical scheme of method for inspecting article surface vein of the present invention:

A kind of method for inspecting article surface vein, the polyvinylidene fluoride piezoelectric membrane is placed on the pressure transducer, and receive that object to be detected presses it and the charge signal correspondingly that produces by the polyvinylidene fluoride piezoelectric membrane, receive object to be detected pressure signal by pressure transducer simultaneously to the polyvinylidene fluoride piezoelectric membrane, described charge signal and pressure signal are gathered after amplifying, and then to charge signal respectively from the statistics, width of cloth threshold and frequency domain angle extraction eigenwert, comprise the average of absolute value, mean square value, mean square deviation, approximate length, signal passes the number of times of mean value, signal slope changes number of times and log power spectrum, use the good GA-VLBP network of training in advance that above-mentioned feature value vector is classified, obtain the texture information of body surface to be detected.

The technical scheme of sensor of the present invention:

A kind of sensor that is used for described method for inspecting article surface vein comprises the superficial makings sensor base, is provided with pressure transducer on pedestal, is covered with the polyvinylidene fluoride piezoelectric membrane on pressure transducer.

Compared with prior art, the present invention has following advantage:

The present invention has utilized polyvinylidene fluoride (PVDF) to have the characteristic of suppressing electrical effect, and making can detect the sensor of texture.The isotropic texture of body surface can be regarded fine particle as, when the PVDF film when body surface slides, these fine particles produce induced charge to the extruding of PVDF film, the electric signal that collects by analysis, judgment object superficial makings characteristic indirectly.This method belongs to Non-Destructive Testing, not only can measure plane information, and is not subjected to environmental interference.That the present invention has is simple in structure, precise control and the characteristics that are easy to realize.

Description of drawings

Fig. 1 is existing superficial makings probe in detecting model.

Fig. 2 is the ashtree plank texture that image method detects.

Fig. 3 is the transducing model of polyvinylidene fluoride (PVDF).

Fig. 4 is a superficial makings sensor sectional view.

Fig. 5 is the test macro block diagram.

Fig. 6 is the design philosophy block diagram of software.

Fig. 7 is that long-range texture detection system is from mechanical autograph letter structural representation.

Embodiment

Embodiment 1

A kind of method for inspecting article surface vein: the polyvinylidene fluoride piezoelectric membrane is placed on the pressure transducer, and receive that object to be detected presses it and the charge signal correspondingly that produces by the polyvinylidene fluoride piezoelectric membrane, receive object to be detected pressure signal by pressure transducer simultaneously to the polyvinylidene fluoride piezoelectric membrane, described charge signal and pressure signal are gathered after amplifying, and then to charge signal respectively from the statistics, width of cloth threshold and frequency domain angle extraction eigenwert, comprise the average of absolute value, mean square value, mean square deviation, approximate length, signal passes the number of times of mean value, signal slope changes number of times and log power spectrum, use the good GA-VLBP network of training in advance that above-mentioned feature value vector is classified, obtain the texture information of body surface to be detected.

Above-mentionedly from the method for statistics, width of cloth threshold and frequency domain angle extraction eigenwert be respectively to charge signal:

Putting down of absolute value $A_1=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\left|x_i\right|,$ X in the formula _iRepresent the numerical values recited of i charge signal, n is the sampled point number;

Mean square value $A_{}$ ${}_2=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\left({x_i}^2\right),$ X in the formula _iRepresent the numerical values recited of i charge signal, n is the sampled point number;

Mean square deviation, ${\mathrm{<figure-callout\; id="3"\; label="A"\; filenames="C200710022961E00111.png"\; state="\{\{state\}\}">A</figure-callout>}}_3=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(x_i-x_{\mathrm{avg}})}^2,$ X in the formula _iRepresent the numerical values recited of i charge signal, n is a sampling number, x _AvgBe signal averaging;

Approximate length $L=\underset{i=1}{\overset{n-1}{\mathrm{\&Sigma;}}}\left|\mathrm{\&Delta;}{x}_i\right|=\underset{i=1}{\overset{n-1}{\mathrm{\&Sigma;}}}|x_{i+1}-x_i|,$ X in the formula _iRepresent the numerical values recited of i charge signal, n is the sampled point number;

Signal passes the number of times of mean value $C=\underset{i=1}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i,$ T in the formula _iExpression goes the line of i o'clock to i+1 point of average back charge signal curve whether to pass X-axis, and two promptly adjacent point value contrary signs can be shown by simple table

(i=1,2...n-1);

Signal slope changes number of times $D=\underset{i=1}{\overset{n-2}{\mathrm{\&Sigma;}}}{k}_i,$ K in the formula _iThe i-1 of expression charge signal curve, i, whether i+1 some the change of slope takes place, and three promptly adjacent signaling point numerical value change amount contrary signs can be shown by simple table

(i=2...n-1);

Adopting Bartlett method average period is to ask power spectrum to be averaged finite length sequence x (n) segmentation again, carries out the log power spectrum feature according to the following steps and asks for:

(1) original signal sample sequence x (n) is divided into 8 continuous frames, 1024 sampled points of every frame.

L _m(n)＝x(1024×(m-1)+n) (m＝1，2...8，n＝1，2...1024)

(2) to every frame signal sample L _m(n) make normalization and centralization and handle,

Normalized:

$L_m\&prime;\left(n\right)=\frac{L_m\left(n\right)}{\underset{1\&le;i\&le;1024}{\max }\left[L_m\left(i\right)\right]}$

Centralization is handled:

$y_m\left(n\right)=L_m\&prime;\left(n\right)-\frac{1}{1024}\underset{i=1}{\overset{1024}{\mathrm{\&Sigma;}}}{L}_m\&prime;\left(n\right)$

(3) to y _m(n) make Fourier transform, obtain Y _m(k):

Y _m(k)＝FFT[y _m(n)]

(4) ask log power spectrum G _m(k)

$G_m\left(k\right)=\frac{10\&times;{\log }_{10}{\left|Y_m\left(k\right)\right|}^2}{1024}$

(5) log power spectrum of all frames is averaged obtains G (K)

$\stackrel{\&OverBar;}{G\left(K\right)}=\frac{1}{8}\underset{m=1}{\overset{8}{\mathrm{\&Sigma;}}}{G}_m\left(K\right)$

(6) dimension that obtains the log power spectrum of signal is 1024 dimensions, the space more complicated of frequency domain character value is carried out dimension-reduction treatment to it here, and continuous 4 mean value of the power spectrum of taking the logarithm is as power spectrum characteristic, and, only get preceding half section log power spectrum because power spectrum has symmetry

$\begin{array}{cc}{G\&prime;}_n\left(K\right)=\frac{\underset{m=4\&times;(n-1)+1}{\overset{4\&times;(n-1)+4}{\mathrm{\&Sigma;}}}\stackrel{\&OverBar;}{G_m\left(K\right)}}{4}& (n=\mathrm{1,2}...128)\end{array}$

The GA-VLBP network that above-mentioned use training in advance is good is categorized as above-mentioned feature value vector:

The dimension of the statistics of extracting from charge signal, width of cloth threshold, frequency threshold eigenwert is:

K＝S+M+F

＝3+3+128

＝134

Total dimension of K representation feature vector wherein, S represents the dimension of statistical characteristics vector, and M represents the dimension of width of cloth threshold proper vector, and F represents the dimension of threshold proper vector frequently.

The use characteristic vector is discerned texture information, just need classify to the space vector of 134 dimensions.Here adopt neural network to be classified in this complex features value space.Because learning rate is very sensitive to the local shape and the curvature of error curved surface in the basic BP algorithm, but learning rate remains unchanged again in iterative process, therefore can not adapt to complicated error curved surface.Here adopt variable learning speed VLBP algorithm that network is trained, in training process, constantly adjust the learning rate and the inertia factor, this constantly cumulative interaction that in good time suppresses again can make iterative process comparatively fast flee from local minimum and walk out " platform ", accelerated speed of convergence, and less to the susceptibility of parameter.But because the architectural feature of multilayer neural network, also there are some defectives in this algorithm, promptly has the local minimum problem inevitably for a nonlinear optimization.In view of the major defect of multilayer neural network algorithm is to avoid local minimum, and genetic algorithm (being called for short GA) has global optimizing ability, here we utilize the global optimizing improved properties multilayer neural network algorithm of GA, promptly utilize GA to train the VLBP network, can on sizable degree, avoid the appearance of local minimum like this, thereby the quickening training speed, and frequency of training is also relatively stable with final weights.

The texture information of many groups different measuring sample that process measures is trained as training set them to the GA-VLBP network, optimize network parameter, the GA-VLBP network that obtains having the texture recognition ability.

Embodiment 2

A kind of sensor that is used to implement described method for inspecting article surface vein comprises superficial makings sensor base 7, is provided with pressure transducer 9 on pedestal 7, is covered with polyvinylidene fluoride piezoelectric membrane 10 on pressure transducer 9.

The sensor is provided with elastic body 8 between polyvinylidene fluoride piezoelectric membrane 10 and pressure transducer 9.

With reference to the accompanying drawings, specific embodiments of the present invention being done one describes in detail:

Principle of work of the present invention is as follows:

PVDF is the abbreviation of Polyvinylidene floride, and its chemical name is a polyvinylidene fluoride, (also available PVF ₂Expression) being a kind of organic polymer functional material, is the highest macromolecular material of finding up to now of piezoelectricity, is a kind of novel flexible organic sensitive material.Compare with other piezoelectric, PVDF has that piezoelectric modulus is big, frequency response is wide, acoustic impedance is easily mated, physical strength height, pliability are good, light weight, shock-resistant, be easy to large tracts of land film forming and advantage such as cheap.And therefore its numerous characteristics is very suitable for making the touch sensor on doing evil through another person near the characteristic of human skin.

Use the PVDF piezoelectric membrane to be among the present invention: when the PVDF piezoelectric membrane produces distortion in the external force of bearing certain orientation as the ultimate principle of touching sensor, its material crystal face or plane of polarization can produce certain electric charge, assemble the positive and negative charge of equivalent contrary sign on the electrode of film both sides.

Under external force, the electric charge of PVDF sensitive element release is the single-valued function of suffered stress.After piezoelectric membrane is stressed, have between output charge and the external force:

$\left\{\begin{array}{c}{q}_i=d_{\mathrm{ij}}{\mathrm{\&sigma;}}_j\\ Q_i=d_{\mathrm{ij}}{F}_j\end{array}\right\}$

In the formula: q _iElectric charge for the output of film unit area; Q _iBe the total output charge of film; σ _jThe stress that bears for film; F _jThe external force of bearing for film; d _IjPiezoelectric strain constant for film.

Because two interpolars have high insulation resistance, therefore, the PVDF piezoelectric film sensor can be regarded a charge generators as, perhaps capacitor, and its capacitance is:

$C=\frac{{\mathrm{\&epsiv;}}_0{\mathrm{\&epsiv;}}_{r}A}{h}$

Wherein: A is the electrode area of aluminizing on the film; H is a film thickness; ε ₀Be the permittivity in the vacuum; ε _rRelative permittivity for film.

The strain stress of the draw direction of PVDF piezoelectric membrane (t) is:

$\mathrm{\&epsiv;}\left(t\right)=\frac{Q\left(t\right)-Q\left(t_0\right)}{E_p{d}_{31}\mathrm{lw}}$

In the formula: E _pElastic modulus for PVDF; Q (t) is the electric charge on the PVDF; d ₃₁Be piezoelectric constant; 1 is the length of PVDF; W is a width.

The transducing model of PVDF as shown in Figure 3,6 is amplifier:

The transducing signal that the piezoelectric effect of PVDF generates is faint charge signal.During STRESS VARIATION on acting on the PVDF micro unit, will produce electric charge at electric capacity the two poles of the earth:

$\mathrm{\&Delta;q}=\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{d}_{3j}\mathrm{\&Delta;}{\mathrm{\&sigma;}}_j$

In the formula: Δ q is the charge variation on the unit area; Δ σ _jThen be that each is to STRESS VARIATION; d _3jFor each to piezoelectric constant.

(x, initial charge surface density y) is q (x, y, t to a point on the investigation sensor ₀), then t surface density of charge constantly is:

$q(x,y,t)=q(x,y,t_0)+\underset{0}{\overset{t}{\&Integral;}}\left[\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{d}_{3j}\frac{\&PartialD;\mathrm{\&sigma;}(x,y,t)}{\&PartialD;t}\right]\mathrm{dt}$

Be electric charge on the PVDF of Ω for area so:

$Q\left(t\right)=Q\left(t_0\right)+\underset{\mathrm{\&Omega;}}{\&Integral;}\&Integral;\left[\underset{0}{\overset{t}{\&Integral;}}\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{d}_{31}\frac{\&PartialD;\mathrm{\&sigma;}(x,y,t)}{\&PartialD;t}\mathrm{dt}\right]\mathrm{dxdy}$

Equate with R voltage by PVDF two interpolar voltages:

$i\&ap;\frac{Q\left(t\right)}{R\&CenterDot;C}$

After sensor inserted amplifying circuit, consider to leak and obtain:

$Q\left(t\right)=Q\left(t_0\right)+\underset{\mathrm{\&Omega;}}{\&Integral;}\&Integral;\left[\underset{0}{\overset{t}{\&Integral;}}\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{d}_{31}\frac{\&PartialD;\mathrm{\&sigma;}(x,y,t)}{\&PartialD;t}\mathrm{dt}\right]\mathrm{dxdy}-\underset{0}{\overset{t}{\&Integral;}}\mathrm{idt}$

In the formula: Q (t) is the t generation total amount of electric charge of pvdf membrane two interpolars constantly; Q (t ₀) be the initial charge amount; I is an electric current; Ω is the sensitive area of PVDF.So:

$Q\left(t\right)=Q\left(t_0\right)+\underset{\mathrm{\&Omega;}}{\&Integral;}\&Integral;\left[\underset{0}{\overset{t}{\&Integral;}}\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{d}_{31}\frac{\&PartialD;\mathrm{\&sigma;}(x,y,t)}{\&PartialD;t}\mathrm{dt}\right]\mathrm{dxdy}-\underset{0}{\overset{t}{\&Integral;}}\frac{Q\left(t\right)}{R\&CenterDot;C}\mathrm{dt}$

In the formula: C is the PVDF equivalent capacity; R _iInput resistance for follow up amplifier.

For obtaining a kind of output intuitively, consider a kind of simple situation, promptly PVDF only acts on the power (d of vertical direction ₃₁Work) and the variation of stress on sensitive area be uniformly, make σ ₁=σ ₂=0, σ ₃Stepless action, and Q (t ₀)=0, then dynamic equation is to σ ₃Be step input separate for:

$Q\left(t\right)=\mathrm{\&Omega;}\&CenterDot;d_{33}\&CenterDot;e^{-\frac{1}{R\&CenterDot;C}}\&CenterDot;U\left(t\right)=Q_0{e}^{-\frac{t}{i}}U\left(t\right)$

In the formula, U (t) is a step function.T=RC is the time constant of the power exciter response electric charge of PVDF generation.The quantity of electric charge that PVDF produces constant T=RC is in time decayed with index law, and this is that leakage by amplifier and sensor causes.Show that PVDF does the static force test and has defective, but do not mean powerless this.Quasi-static charge amplifier can make r on higher value, thereby the leakage of circuit is reduced to minimum level, and another kind of better method compensates when being data processing.

After the piezoelectric effect of having analyzed the PVDF film, made the superficial makings sensor, Fig. 4 is its sectional view, and 7 is the superficial makings sensor base, and 8 is elastic filling material, and 9 is pressure transducer, 10 is the PVDF piezoelectric membrane.

When the superficial makings sensor is subjected to the dynamic pressure of texture particle in the body surface slip, PVDF piezoelectric membrane 4 both sides produce induced charge, and form the millivolt step voltage, and carry out active data by charge amplifier and amplify, through data acquisition unit the data that collect are imported in the microcomputer again and handled.The test macro block diagram as shown in Figure 5.

The detection core of texture signal is to utilize modern theory that the data that collect are handled and pattern-recognition, utilizes functions such as realization control such as the communication technology simultaneously.The design philosophy of software as shown in Figure 6, software platform mainly is divided into the five functional module as can be seen, i.e. integrated circuit board control, data acquisition, data storage and demonstration, software interface and algorithm are integrated.Wherein algorithm is integrated, i.e. data analysis and pattern-recognition are the cores of this testing software part.

Fig. 7 is the simplified structure of long-range texture detection system from mechanical arm, and 11 is from mechanical arm, and 12 is the superficial makings sensor.In remote control system, the communication link with position command pass in the environment at a distance from mechanical arm 1, control its motion, when the object in the environment of distant place when superficial makings sensor 2 from mechanical arm 1 has relative sliding motion, superficial makings sensor 2 is with the article surface vein signal of gathering, turn back to master station through the communication link, discern by master station.

Claims (3)

1. method for inspecting article surface vein, it is characterized in that: the polyvinylidene fluoride piezoelectric membrane is placed on the pressure transducer, and receive that object to be detected presses it and the charge signal correspondingly that produces by the polyvinylidene fluoride piezoelectric membrane, receive object to be detected pressure signal by pressure transducer simultaneously to the polyvinylidene fluoride piezoelectric membrane, described charge signal and pressure signal are gathered after amplifying, and then to charge signal respectively from the statistics, width of cloth threshold and frequency domain angle extraction eigenwert, this eigenwert comprises the average of absolute value, mean square value, mean square deviation, approximate length, charge signal passes the number of times of mean value, signal slope changes number of times and log power spectrum, the good variable learning speed multilayer neural network that utilizes the genetic algorithm training is classified to above-mentioned eigenwert to use training in advance, obtains the texture information of body surface to be detected.

2. sensor that is used to implement the described method for inspecting article surface vein of claim 1, it is characterized in that comprising superficial makings sensor base (7), on superficial makings sensor base (7), be provided with pressure transducer (9), on pressure transducer (9), be covered with polyvinylidene fluoride piezoelectric membrane (10).

3. sensor according to claim 2 is characterized in that being provided with elastic body (8) between polyvinylidene fluoride piezoelectric membrane (10) and pressure transducer (9).

Images