CN101303802A - Method and system for on-line automatically beforehand judgment of overloading wagon - Google Patents

Abstract

The invention discloses an overloaded truck online automatic pre-judgment method and a system thereof, which overcomes the problems of difficult classification of motor vehicle types and slow weighing speed, etc. The method includes two steps, that is, a vehicle type recognition step and a overloading pre-judgment step, and the overloading pre-judgment step includes the following processes: 1) a vibration sensor is utilized for collecting the vibration signal of a running vehicle; 2) an endpoint detection method is utilized for judging whether a running vehicle is running on the road or not, if no running vehicle, the signal does not enter the next step; 3) whether the running vehicle is a goods carrying vehicle or not is judged by the collected sound signal of the running vehicle, if not, the signal does not enter the next step; 4) when the running vehicle is judged to be the goods carrying vehicle by the vehicle recognition step, and signal preprocessing is carried out on the vibration signal of the goods carrying vehicle, that is, signal de-noising, frame interception and normalization; 5) a wavelet packet parameter calculation is carried out on the vibration signal after the signal preprocessing; and 6) BP network overloading is pre-judged. The system implementing the method consists of a signal input module, a signal processing module, and a microcomputer interface module, which are sequentially connected by electric wires.

Description

The on-line automatic pre-judgement method and system of overloading wagon

Technical field

The present invention relates to a kind of online quick identification type of vehicle and to the method for the real-time detection that whether it overloads, the overload degree is judged in advance and pre-judgement system, more particularly, it relates to the on-line automatic pre-judgement method and system of a kind of overloading wagon.

Background technology

One. automatic vehicle classification

The fundamental purpose of vehicle classification is exactly to guarantee the fairness of charging between type of vehicle, vehicle flowrate added up automatically, and to the overweight pre-automatically assistance of judging of vehicle.

Automatic vehicle classification AVC (Automatic Vehicle Classification) is the comprehensive embodiment of modern technologies, it is by detecting the intrinsic physical parameter of vehicle itself, the suitable classification and identification algorithm of utilization is carried out somatotype to vehicle on one's own initiative under certain vehicle classification standard.At present, existing automobile automatic recognition technology all is to realize classifying by the vehicle feature that wagon detector detects vehicle, and its method mainly contains profile scan, axletree counting, electromagnetic induction coil detection, dynamic weighing, discern automatically based on the car plate of Flame Image Process and pattern-recognition or vehicle etc.

1. profile scan

The purpose of profile scan is to obtain the appearance information (mainly being physical dimension) of vehicle, thereby vehicle is classified.General radiowave or infrared ray, the more advanced then use laser of using of scanning.But the method is responsive especially to environment visibility, and very poor for profile close passenger vehicle and lorry classifying quality.

2. axletree is counted

The axletree counting is exactly that utilization detects the number of axle of vehicle someway, with a kind of standard of testing result as vehicle classification.But because the axletree number is not principal element, judge that there is bigger error easily in vehicle in the standard of vehicle classification, so the recognition technology of axletree counting do not use separately generally so use this method, but with other system's cooperating.

3. electromagnetic induction coil detects

The electromagnetic induction coil detection is exactly utilize the chassis structure of different automobile types and ferromagnetic material distribution different, and it is also different that electric current changes the variation in the magnetic field that causes, dissimilar vehicles is distinguished according to the difference of induction curve by system.But need coil is imbedded underground in force, equipment also can be subjected to the vehicle extruding, therefore have shortcomings such as the road surface of destruction, poor, the easy damage of mobility, life-span weak point, influenced by Vehicle Speed and other disturbing factor etc., the vehicle characteristics curve that obtains is undesirable to the classifying quality of vehicle, so use common servicing unit seldom separately as other system.

4. dynamic weighing

Dynamic weighing belongs to the passive detection technique of contact, when vehicle when imbedding the detecting device under the road surface, the stressed generation deformation of detecting device, according to back information vehicle is detected, but the axle weight of measuring vehicle, wheelbase, gross weight, the speed of a motor vehicle etc., and, discern vehicle automatically by the vehicle classification table that pre-establishes.But because its technology is ripe not enough, equipment is installed complicated, life-span weak point in addition, and temperature, Vehicular vibration, road quality etc. all can produce a very large impact system accuracy, so also be not used widely in vehicle detection and vehicle classification.

5. based on Flame Image Process and pattern-recognition

Utilization video image technology is differentiated vehicle, the statistics magnitude of traffic flow is the focus of traffic circle research always, and has used on the minority highway.This method is all taken in the image of vehicular traffic by being arranged on expressway access or video frequency pick-up head along the line, image information is transferred to freeway surveillance and control administrative authority again and handles.Its know-why is to utilize methods such as motion segmentation and Model Matching that contour of the vehicle, color character are made judgement, perhaps finds out the relevant information of type of vehicle from the various features of car plate.If the type of vehicle that features such as the size of car plate, shape and color are corresponding different just can be classified to vehicle so thus.And, can read concrete number on the car plate to the further identification of car plate, just can in relevant data (as the registration record database of vehicle license), search for according to this number, obtain the details of type of vehicle.Its advantage is the comprehensive information that can obtain comprising the variety of event that takes place on the highway of vehicle accurately and rapidly, and detects the influence that effect is not changed by the volume of traffic.The shortcoming of this method then mainly is because cost height, condition of work harshness that video detection technology self exists cause: the construction cost of a Video Detection point is at least at 20,000 yuans, and the amount of image information that the utilization video technique obtains is very big, and the post-processed need of work expends lot of manpower and material resources; In addition, video detection technology is subjected to condition effect such as visibility, illumination bigger, under adverse weather situations such as dense fog, sleet, dust, detects effect and declines to a great extent.Certainly because have shade, image overlay, unfavorable factor such as contain much information, utilize the method for contour of the vehicle feature identification vehicle to be still waiting further to improve at aspects such as real-time, discriminations.

6. automatic vehicle identification

Electronic tag is the important component part of (claim E-payment system Electronic TollCollection System again, be called for short the ETC system) of state-of-the-art no-stop charging system in the world today.No-stop charging system is Electronic Identification Card (being electronic tag) the read-write transceiver other with being installed in the lane in which the drivers should pay fees that is mounted on the automobile, utilize automatic vehicle identification (AutomaticVehicle Identification is called for short AVI) technology to finish the wireless data communication between vehicle and the charge station, carry out the exchange of automatic vehicle identification and relevant charge data, carry out the processing of charge data by computer network, the full-electronic Fare Collection System of charge automatically of realizing not stopping.This kind method realizes the robotization of charge by the automatic vehicle identification technology, makes when vehicle passes through the expressway tol lcollection crossing, and the dispense with parking wait is paid dues, thereby has improved the capacity and level-of-service of highway greatly.The utilization no-stop charging system can obtain the vehicle relevant information that comprises vehicle, and its conveniently one side is arranged.But need carry out the mass upgrade transformation to original charge station and system owing to build no-stop charging system, and go back the unified construction criteria of neither one both at home and abroad at present, to build maintenance cost huge in addition, the maintainer is had relatively high expectations, system to the problems such as processing of the compatibility that prevents fee evasion, numerous user and system and transit vehicle also difficulty solution is preferably arranged, therefore after a very long time also be difficult to obtain at home really popularize.Even if no-stop charging system obtains popularizing on a large scale in the future, also can also there be the possibility that is damaged, forges, changes because of electronic tag itself, also need the effective automatic vehicle classification of cover system, so that as standby and verification, consider that from the angle of supervision back-up system also can reduce the generation of leaking receipts and wrong receipts situation in addition.

In sum, various automatic vehicle classification technology all have its merits and demerits separately, but from practical application effect, owing to have technology and economic dispatch aspect, and practical requirement well still at present.Therefore, be different from the principle of existing Automatic Measurement Technique, acoustical signal in the employing vehicle operating process is as information source, study a kind of new high-level efficiency, low cost, be applicable to the highway vehicle automatic distinguishing method of technological system all weather operations requirement, it all is necessary still putting into practice in theory.

Two. automatic judgement that vehicle is overweight

An important enormous expenditure of traffic system is exactly to cause the damage of road to carry out road maintenance or even reconstruction because of heavy-duty vehicle rolls.Obviously vehicle heavy more (surpassing the highway limitation standard) is just big more to the damage influence of road.The method of measuring vehicle overload at present mainly is that platform scale is weighed, and wherein the part weighing technology speed of weighing is slow and influence traffic.

By the end of 2007, national highway total kilometrage was 180.98 ten thousand kilometers, and the standard highway mileage accounts for 79.5% of highway total kilometrage for 143.87 ten thousand kilometers.Wherein 29745 kilometers of highways, 29903 kilometers of Class I highways, Class II highway 211929 highways, 324788 kilometers of Class III highways, 842373 kilometers of Class IV highwaies.

By the end of 2007, national automobile pollution was 1.598 hundred million.1-12 month truck freight volume amounted to 162.8 hundred million tons of accumulative totals in 2007, increased by 11% on a year-on-year basis than accumulative total in 2006; 1-12 month highway rotation volume of goods transport amounted to accumulative total 11257.6 hundred million ton kilometres in 2007.As seen, overload transportation management, weight metering charging all have huge current demand to the rapid weighing technology.

Dynamic Weighing Technology is the top technology in the field of weighing, and is the effective means of the fast automatic Weighing of large bulkload.Dynamic weighing is the process of measuring the dynamic tire force of driving vehicle and calculating the corresponding static vehicle weight; A Highway Weigh-in-Motion Systems is a cover sensor and supports instrument, be used for measuring at the appearance of a driving vehicle of locality special time and dynamic tire force thereof, calculate the weight, the speed of a motor vehicle, wheelbase, type of vehicle of vehicle and about other parameter of vehicle and processes and displays with store these information.Dynamic weighing system mainly comprises sensor and supports instrument, according to the difference of its Fundamentals of Sensors, mainly comprises Dynamic Weighing Technology such as resistance-strain type, piezomagnetic, condenser type, type vibration wire, inductance type, optical fiber type, gyro ceremony at present.

1. pressure resistance type LOAD CELLS

After some solid material was subjected to the external force effect, except producing distortion, its resistivity also will change.This phenomenon that resistivity of material is changed owing to stress is called " piezoresistive effect ".The LOAD CELLS of utilizing piezoresistive effect to make is called the pressure resistance type LOAD CELLS.The pressure resistance type LOAD CELLS has two types, promptly is semiconductor strain formula LOAD CELLS, diffused pressure resistance type LOAD CELLS.

The version of semiconductor strain formula LOAD CELLS is identical with resistance strain weighing transducer basically, and the sensitive grid of different is foil gauge is to make with semiconductor.Compare with metal strain plate, its advantage is that volume is little and highly sensitive.The same problem that it also exists sensitive grid to be easy to damage.

The substrate of diffused pressure resistance type LOAD CELLS is a single crystal silicon semiconductor.Therefore the monocrystalline silicon anisotropy must come processing and fabricating diffuse si sensitive resistance grid sheet according to sensor stress deformation situation.Be difficult for guaranteeing precision.

2. piezoelectric type LOAD CELLS

Some ion-type dielectrics are subjected to the external force effect and when being out of shape in a certain direction, the inner polarization phenomena that can produce, and produce electric charge from the teeth outwards, after external force is removed, get back to not electriferous state again.The phenomenon that this mechanical energy is converted to electric energy is called " along piezoelectric effect ".Otherwise, be called " inverse piezoelectric effect ".Material with piezoelectric effect is called piezoelectric.

The ultimate principle of piezoelectric type LOAD CELLS is to utilize the piezoelectric effect of piezoelectric to make.Its principle of work just is based on piezoelectric crystal under external force, can produce electric charge on two surfaces of piezoelectric element, assembles positive charge and assemble the negative charge of equivalent on a surface on another surface.

The natural frequency of piezoelectric type LOAD CELLS is very high, so its high frequency response is better than resistance strain weighing transducer and pressure resistance type LOAD CELLS.But piezoelectric transducer requires supporting equipment many, realizes that cost is higher.

3. resistance-strain type

Resistance strain weighing transducer is made up of strain ga(u)ge and two main devices of metallic elastic beam.Strain ga(u)ge is made thread resistance wire by the constantan paper tinsel at the dielectric base material, utilize the resistivity of lead and the characteristic that diameter is inversely proportional to, when thread resistance wire under external force, in elastic range during its diameter generation slight change, its resistance value also changes thereupon, thereby strain ga(u)ge can the power that LOAD CELLS is suffered become electric signal.

The characteristics of resistance strain-gauge transducer are that measurement range is wide, generally can measure tens load that restrain several kilotons, and structure is simpler, and install at the scene that is suitable for commercial unit, and its use has accounted for about 90% of all LOAD CELLS.

The shortcoming of resistance strain weighing transducer is that the diameter of thread resistance wire is superfine, is used for abominable working environment for a long time and is easy to damage, and output signal is little, and overload capacity is lower.

4. capacitance weighing sensor

It utilizes the proportional relationship work of the oscillation frequency f and the polar plate spacing d of capacitor oscillatory circuit capacitance weighing sensor.Pole plate has two, and one maintains static, and another piece is removable.When bearing platform loads measured object, the leaf spring deflection, the distance between the two-plate changes, and the oscillation frequency of circuit also changes thereupon.The variation of measuring frequency can be obtained the quality of measured object on the bearing platform.Capacitance type sensor power consumption is few, and cost is low, and accuracy is 1/200～1/500.

5. oscillatory type LOAD CELLS

After oscillatory type sensor flexible member was stressed, its natural vibration frequency was directly proportional with the square root of acting force.Measure the variation of natural frequency, can obtain measured object and act on power on the flexible member, and then obtain its quality.The oscillatory type sensor has two kinds of type vibration wire and tuning-fork types.

The flexible member of vibrating string type sensor is the string silk.When being added with measured object on the bearing platform, the intersection point of V-arrangement string silk is pulled to down, and the increase of the pulling force of left string, and the pulling force of right string reduces.Different variations takes place in the natural frequency of two strings.Obtain frequency poor of two strings, can obtain the quality of measured object.The accuracy of vibrating string type sensor is higher, can reach 1/1000～1/10000, and range of weighing is 100 to restrain to the hundreds of kilogram, but complex structure, difficulty of processing is big, the cost height.

The flexible member of tuning-fork type sensor is a tuning fork.The tuning fork end is fixed with piezoelectric element, and it vibrates with the natural frequency of tuning fork, and can measure oscillation frequency.When being added with measured object on the bearing platform, the stressed and natural frequency of tuning fork draw direction increases, and the degree of increase is directly proportional with the square root that applies power.Measure the variation of natural frequency, can obtain weight and put on power on the tuning fork, and then obtain the weight quality.Tuning-fork type sensor power consumption is little, and the accuracy of measuring is up to 1/10000～1/200000, and range of weighing is 500g～10kg.

6. fiber-optic grating sensor

The Moire fringe that the raster pattern sensor utilizes grating to form converts angular displacement to photosignal.Grating has two, and one is fixed grating, and another is the mobile grating that is contained on the dial plate axle.The measured object that is added on the bearing platform makes the rotation of dial plate axle by the force transferring lever system, drives mobile grating and rotates, and Moire fringe is also moved thereupon.Utilize photoelectric tube, change-over circuit and Displaying Meter, can calculate the Moire fringe quantity that moves past, measure the size of grating angle of rotation, thereby determine and read the measured object quality.

The fiber-optic grating sensor technology is to utilize insensitive original paper in the sensor---the optical spectrum of fiber grating reflection is to the sensitivity characteristic of temperature, strain, output optical spectrum analysis of input light source excitation and the physical quantity finished fiber-optic grating sensor by inner each functional module of optical fiber grating sensing network analyser convert, and provide the information such as temperature, ess-strain, pressure, displacement of each monitoring point with chart and digital form.The appearance of fiber grating sensing technology has brought unprecedented vitality for the development of sensor.

7. gyro ceremony LOAD CELLS

Gyro ceremony sensor, rotor is contained in the inner frame, stablizes rotation with angular velocity omega around X-axis.Inner frame connects with outside framework through bearing, and can tilt to rotate around transverse axis Y.Outside framework connects with support through universal coupling, and can rotate around Z-axis Z.Armature spindle (X-axis) be not subjected to external force to make time spent maintenance horizontality.One end of armature spindle is done the time spent being subjected to external force (P/2), produces and rotates (precession) around Z-axis Z.Angular velocity of precession ω is directly proportional with external force P/2, measures ω by the method that detects frequency, can obtain the external force size, and then obtains the quality of the measured object that produces this external force.

The gyro ceremony sensor response time fast (5 seconds), no hysteresis phenomenon, good temp characteristic (3ppm), vibration effect is little, and the accurate precision height of frequency measurement is so can obtain the high resolution (1/100000) and the high accuracy of measuring (1/30000～1/60000).

Summary of the invention

Technical matters to be solved by this invention is to have overcome the automatic vehicle classification difficulty and the slow-footed problem of weighing that prior art exists, and a kind of on-line automatic pre-judgement method and system of overloading wagon based on lorry operation sound and vibration signal are provided.

For solving the problems of the technologies described above, the present invention has adopted following technical scheme to realize: this method comprises the vehicle identification and pre-two steps of judging that overload, and described overload judges that in advance step comprises following flow process:

1) utilizes the vibration signal of vibration transducer collection vehicle in travelling;

2) utilize the method judgement of end-point detection to have or not the vehicle that travels, do not have the vehicle that travels, signal does not enter next step;

3) voice signal from the driving vehicle gathered judges whether to be goods carrying vehicle, is not goods carrying vehicle, and signal does not enter next step;

4) when being goods carrying vehicle by the judgement of vehicle identification step, promptly the vibration signal to goods carrying vehicle carries out signal denoising, frame intercepting and normalized Signal Pretreatment;

5) signal through Signal Pretreatment is carried out the wavelet packet calculation of parameter:

(1) each sample is carried out three layers of WAVELET PACKET DECOMPOSITION, extract the signal characteristic X of the 3rd layer of 8 frequency channel _3j(j=0,1, ..., 7),

(2), extract the signal of each frequency band range to the reconstruct of WAVELET PACKET DECOMPOSITION coefficient.With S _3jExpression X _3jReconstruction signal;

(3) ask the gross energy of each band signal, because input signal is a random signal, its output also is random signal, S _3j(j=0,1, ..., 7) and corresponding energy is E _3j(j=0,1 ..., 7), then have:

$E_{3j}=\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{\left|x_{\mathrm{jk}}\right|}^2$

Wherein, x _Jk(j=0,1 ..., 7; K=1,2 ..., n) expression reconstruction signal S _3jThe amplitude of discrete point;

(4) structure energy vector space

Structure energy vector is:

T＝[E ₃₀，E ₃₁，E ₃₂，E ₃₃，E ₃₄，E ₃₅，E ₃₆，E ₃₇]；

(5) adopt energy normalized, promptly each parameter quadratic sum is that one method is carried out normalized to the energy vector after the normalization, within data map to 0～1 scope,

Order: $E={\left(\underset{j=0}{\overset{7}{\mathrm{\Σ}}}{\left|E_{3j}\right|}^2\right)}^{\frac{1}{2}}$

$T^{\′}=[\frac{E_{30}}{E},\frac{E_{31}}{E},\frac{E_{32}}{E},\frac{E_{33}}{E},\frac{E_{34}}{E},\frac{E_{35}}{E},\frac{E_{36}}{E},\frac{E_{37}}{E}]$

Energy vector after vector T ' the be normalization, $\frac{E_{3j}}{E}=T_j,$ Then:

T′＝[T ₀，T ₁，T ₂，T ₃，T ₄，T ₅，T ₆，T ₇]；

6) the BP network overload is judged in advance

(1) sample that will learn in the learning phase input, till network convergence, this stage is exactly constantly to adjust each neuronic strength of joint, makes it can approach pairing output on the meaning of least square, and learning phase takes offline mode to carry out;

(2) at cognitive phase, calculate for given input, judged the result in advance.

For the pre-judgement system of the overloading wagon on-line automatic pre-judgement method of a kind of enforcement based on lorry operation sound and vibration signal is provided, the present invention has adopted following technical scheme to realize: this system is made up of signal acquisition module, signal processing module and microcomputer interface module three parts, signal acquisition module is that electric wire is connected with signal processing module, and signal processing module is that electric wire is connected with the microcomputer interface module.

Described signal acquisition module is by sound transducer, vibration transducer, first modulate circuit, second modulate circuit and model are that the two passage A/D converters of THS10064 are formed, sound transducer is connected with the input end electric wire of first modulate circuit, the output terminal of first modulate circuit and model are that the pin AINP electric wire of the two passage A/D converters of THS10064 is connected, vibration transducer is connected with the input end electric wire of second modulate circuit, the output terminal of second modulate circuit and model are that the pin BINP electric wire of the two passage A/D converters of THS10064 is connected, and model is that the output terminal of the two passage A/D converters of THS10064 is connected with the signal processing module electric wire.

Described signal processing module comprises that the model that computer program is installed is digital signal processor DSP, the FLASH that model is AM29LV800B, the random access device RAM that model is IS61LV12816 of TMS320VC5402.Model is to be respectively electric wire between digital signal processor DSP and the model of TMS320VC5402 is the two passage A/D converters of THS10064, FLASH that model is AM29LV800B, model is IS61LV12816 random access device RAM and the microcomputer interface module to be connected.

Described microcomputer interface module is that the high speed USB chip of Cy7c68013 is formed by model, the input end of microcomputer interface module and model are that an end electric wire of the digital signal processor DSP of TMS320VC5402 is connected, and the output terminal of microcomputer interface module is connected with the input end electric wire of PC.

The model that computer program is installed described in the technical scheme is that the digital signal processor DSP of TMS320VC5402 is meant: the computer sub of being extracted by vehicle sound characteristic parameter is installed, the computer sub of neural network vehicle classification, the computer sub that the vehicle weight characteristic parameter extracts, the model of the computer program that the computer sub that pre-computer sub of judging of neural network overload of vehicle and result dispose is formed is the digital signal processor DSP of TMS320VC5402; Described model is that digital signal processor DSP and the model of TMS320VC5402 is that the random access device RAM electric wire of IS61LV12816 is connected and is meant: model is the pin D (0:15) of the digital signal processor DSP of TMS320VC5402, A (0:15) and HDO are respectively the pin I/O (0:15) of the random access device RAM of IS61LV12816 with model, A (0:15) is connected with the A16 electric wire, and model is the pin R/W of the digital signal processor DSP of TMS320VC5402, MSTRB#, DS# and XF are that the pin WE/ of the random access device RAM of IS61LV12816 is connected with the CE/ electric wire by two with door and model respectively; Described model is that digital signal processor DSP and the model of TMS320VC5402 is that the FLASH electric wire of AM29LV800B is connected and is meant: model is the pin D (0:15) of the digital signal processor DSP of TMS320VC5402, A (0:15), BIO# and A (16:18) are respectively the pin D (0:15) of the FLASH of AM29LV800B with model, A (0:15), RY/BY# is connected with A (16:18) electric wire, and model is the pin R/W# of the digital signal processor DSP of TMS 320VC5402, MSTRB#, DS# and XF are the pin OE# of the FLASH of AM29LV800B with door with model by two not gates and three respectively, WE# is connected with the CE# electric wire; Described model is that digital signal processor DSP and the model of TMS320VC5402 is that the output terminal electric wire of the two passage A/D converters of THS10064 is connected and is meant: model is the pin IOSTRB# of the digital signal processor DSP of TMS320VC5402, A15, BI0/, BCLKX0 and D0-D9 are respectively the pin CS0/ of the two passage A/D converters of THS10064 with model, CS1, DATA_AV, GONV_GLK is connected with the D0-D9 electric wire, and model is that pin IS# and the R/W# of the digital signal processor DSP of TMS320VC5402 is that the pin R/W# electric wire of the two passage A/D converters of THS10064 is connected by one with door and model; Described model is that the digital signal processor DSP of TMS320VC5402 is connected with microcomputer interface module electric wire and is meant: model is that digital signal processor DSP and the model of TMS320VC5402 is that the high speed USB chip electric wire of Cy7c68013 is connected, for this reason, model is the pin PS of the digital signal processor DSP of TMS320VC5402, DS, IS, RW, XF, IOSTRB, MSTRB, BIO, READY, INT0, INT1, A (0-15) and D (0-7) are respectively the pin CPLD_PS of the complex programmable logic device (CPLD) of XC95144 with model earlier, CPLD_DS, CPLD_IS, CPLD_RW, CPLD_XF, IOSTRB, MSTRB, CPLD_BIO, READY, INT0, INT1, A (0-15) is connected with D (0-7) electric wire, model is the pin CY_IFCLK of the complex programmable logic device (CPLD) of XC95144, CY_CLKOUT, CY_SLWR, CY_SLRD, CY_CTL (0-2), PA (0-2), USB_INT1, FIFOADR (0-1), PKTEND, PA7, CY_PA (0-7) and CY_PD (0-7) are respectively the high speed USB pin of chip CY_IFCLK of Cy7c68013 again with model, CY-CLKOUT, CY_SLWR, CY_SLRD, CY_CTL (0-2), PA (0-2), USB_INT1, FIFOADR (0-1), PKTEND, PA7, CY_PA (0-7) is connected with CY_PD (0-7) electric wire, model is the pin A (0-18) of the complex programmable logic device (CPLD) of XC95144, D (0-8), FLASH_CE is respectively again the pin A (0-18) of the FLASH of AM29LV800B with model, D (0-8), the FLASH_CE electric wire connects, and model is the pin A (0-18) of the complex programmable logic device (CPLD) of XC95144, D (0-8), RAM_CE, RAM_WE, RAM_RD is respectively again the pin A (0-18) of the random access device RAM of IS61LV12816 with model, D (0-8), RAM_CE, RAM_WE, the RAM_RD electric wire connects.Compared with prior art the invention has the beneficial effects as follows:

One. the vehicle weight simulated experiment

Between sound that sends when this experiment is intended to prove the different load-carryings of automobile and its operation and the vibration signal close getting in touch arranged.

Toy car dead weight used in the experiment is 30kg, and the car body deadweight is 7.2kg.Under lab, we have carried out test unloaded, that goods 5.3kg is housed, voice signal under three kinds of situations of goods 12kg is housed to toy car.Toy car uses shock sensor to carry out signals collecting by the road surface of poly (methyl methacrylate) plate.The signal of gathering is carried out carrying out wavelet packet analysis after 1/10 decimation in frequency, and the result is as follows:

Unloaded (a):

23.3860 44.3300 1.8302 28.8040 0.2298 0.3774 0.5802 0.4626

55.9120 34.9160 0.5721 7.9456 0.0922 0.1475 0.2459 0.1683

28.3130 45.0590 2.9478 21.6040 0.2351 0.3519 0.8160 0.6729

46.0670 42.8220 0.6906 9.4452 0.1494 0.2388 0.3409 0.2464

29.8930 40.8530 1.8226 25.6910 0.2573 0.3769 0.4968 0.6096

44.9190 42.6600 1.0075 10.2600 0.1552 0.2639 0.4314 0.3028

Fully loaded (b):

19.7250 57.0810 2.4698 16.8850 0.5735 0.9278 1.0247 1.3117

36.2290 50.2170 1.2329 9.9678 0.2728 0.6142 0.8779 0.5888

26.6340 51.0580 1.7331 17.3030 0.5550 0.6517 0.9739 1.0907

26.9670 54.9390 2.5446 11.4770 0.7178 0.9057 1.5289 0.9207

19.8370 59.3930 2.8335 12.4400 0.5359 0.9452 1.9654 2.0495

43.1520 42.0890 1.6995 9.9520 0.4897 0.7862 1.1849 0.6463

Half-full year (c):

40.5330 35.4770 2.2525 19.3020 0.3296 0.5027 0.8152 0.7875

37.5230 52.0180 0.8621 8.2608 0.1855 0.3141 0.5309 0.3054

45.5680 38.0410 1.3723 12.3140 0.3149 0.5598 0.7465 1.0824

47.9950 41.9300 0.9715 7.6548 0.2113 0.3134 0.6053 0.3180

18.1890 59.0950 1.9240 17.1030 0.5294 0.7741 1.0790 1.3064

38.5880 47.6020 1.5171 10.4910 0.2429 0.4441 0.6587 0.4557

Mean value:

(a)38.0817 41.7733 1.4785 17.2916 0.1865 0.2927 0.4852 0.4104

(b)28.7573 52.4628 2.0856 13.0041 0.5241 0.8051 1.2593 1.1013

(c)38.0660 45.6938 1.4832 12.5209 0.3023 0.4847 0.7393 0.7092

Variance:

(a)161.3758 13.4176 0.8130 84.0078 0.0041 0.0085 0.0399 0.0417

(b)86.3803 38.0240 0.3836 10.9453 0.0211 0.0210 0.1718 0.2891

(c)111.3227 80.0687 0.2894 22.5939 0.0156 0.0300 0.0378 0.1765

The U assay:

Uab?1.4510 3.6507 1.3594 1.0778 5.2121 7.3020 4.1213 2.9422

Ubc?1.6216 1.5258 1.7984 0.2044 2.8381 3.4758 2.7824 1.4073

Uac?0.0023 0.9932 0.0112 1.1318 2.0210 2.3974 2.2323 1.5671

The data analysis result:

From above data as can be seen, when significance level is defined as α=0.05, has three eigenvectors and satisfy requirement greater than 1.96.Therefore, can think tentatively whether this overload of analyzing for on-highway motor vehicle has certain identification, particularly the truck overload often reaches more than 3 to 5 times of automotive dead weight in the reality, its signal characteristic will be more obvious, should be able to produce a desired effect in highway field experiment in the future.

Two. beneficial effect compared with prior art of the present invention

1. discern on the basis in vehicle based on the on-line automatic pre-judgement method and system of overloading wagon of lorry operation sound and vibration signal, can estimate the load-carrying degree of same model, whether vehicle overloaded judge in advance, send overweight suspicion alarm, significantly reduce the quantity of the vehicle of need accurately weighing, increase work efficiency.Overload suspicion threshold value can require to adopt low threshold value and high threshold dual mode to be provided with according to the traffic administration decision-making.Low threshold mode guarantees that all overweight vehicles all excite the alarm of overload suspicion, but may produce the not overload false-alarm of overloaded vehicle, thereby increases the vehicle fleet size of weighing; The high threshold mode guarantees that all excite the vehicle of overweight suspicion alarm all to overload, but may produce the false dismissal of overloaded vehicle.Simulated experiment shows, when vehicle load difference 30% is above, utilizes the Vehicular vibration signal just can effectively discern, and satisfies the pre-basic demand of judging of overload of vehicle.

2. owing to adopt acoustic signal and (ground) vibration signal based on the on-line automatic pre-judgement method and system of the overloading wagon of lorry operation sound and vibration signal, compare with existing weighing technology, belong to and the non-contacting metering system of car body, and equipment cost is low, the signal Processing amount is little, do not destroy road surface (vibration transducer only needs to keep good contact to get final product with ground), easily install, safeguard, more help the realization that online in real time is measured.

Based on the on-line automatic pre-judgement method and system of overloading wagon of lorry operation sound and vibration signal through further furtheing investigate and be converted into technical products, very wide application prospect will be arranged.According to noted earlier, by the end of 2007, national highway total kilometrage was 180.98 ten thousand kilometers, wherein 29745 kilometers of highways, 29903 kilometers of Class I highways.2007, national automobile pollution was 1.598 hundred million, and the highway rotation volume of goods transport amounts to accumulative total 11257.6 hundred million ton kilometres.As seen, overload transportation management, weight metering charging all have huge current demand to the rapid weighing technology.

Description of drawings

The present invention is further illustrated below in conjunction with accompanying drawing:

Fig. 1 is based on the signal processing flow schematic block diagram of the on-line automatic pre-judgement method of overloading wagon of lorry operation sound and vibration signal;

Fig. 2 is based on the structure composition schematic block diagram of the on-line automatic pre-judgement of lorry operation sound and the overloading wagon of vibration signal system;

Fig. 3 is that model is the digital signal processor DSP of TMS320VC5402 and the random access device RAM line connection diagram that model is IS61LV12816;

Fig. 4 is that model is the digital signal processor DSP of TMS320VC5402 and the FLASH line connection diagram that model is AM29LV800B;

Fig. 5 is that model is the digital signal processor DSP of TMS320VC5402 and the two passage A/D converter line connection diagrams that model is THS10064;

Fig. 6 is that model is the digital signal processor DSP of TMS320VC5402 and the complex programmable logic device (CPLD) line connection diagram that model is XC95144;

Fig. 7-A is that model is the line connection diagram of complex programmable logic device (CPLD) and the FLASH that model is AM29LV800B of XC95144;

Fig. 7-B is that model is the line connection diagram of complex programmable logic device (CPLD) and the random access device RAM that model is IS61LV12816 of XC95144;

Fig. 8 is that model is the complex programmable logic device (CPLD) of XC95144 and the high speed USB chips wire connection diagram that model is Cy7c68013;

Fig. 9 is based on three layers of BP network that have a middle layer (hidden layer) in the on-line automatic pre-judgement method of overloading wagon of lorry operation sound and vibration signal;

Figure 10 is based on wavelet decomposition synoptic diagram in the on-line automatic pre-judgement method of overloading wagon of lorry operation sound and vibration signal;

Figure 11 is based on wavelet packet decomposing schematic representation in the on-line automatic pre-judgement method of overloading wagon of lorry operation sound and vibration signal;

Embodiment

Below in conjunction with accompanying drawing the present invention is explained in detail:

The present invention proposes a kind of operation sound that produces when utilizing vehicle ' at first discerns vehicle, whether be truck vehicle after, the method and system of the on-line automatic measurement that running noises that produces when utilizing vehicle ' and ground vibration signal are judged in advance to the overweight situation of vehicle if determining fast.

I. the design concept of the on-line automatic pre-judgement method and system of overloading wagon and major technique feature

The present invention relates to a kind of on-line automatic pre-judgement method and system of overloading wagon that adopt self-editing computer program based on lorry operation sound and vibration signal.The self-editing computer program model of packing into is in the digital signal processing chip DSP of TMS320VC5402, on-line automatic pre-this computer program of judgement system implementation of overloading wagon is realized the on-line automatic pre-judgement method of overloading wagon based on lorry operation sound and vibration signal.

One. the on-line automatic pre-judgement method of overloading wagon

The design concept of the on-line automatic pre-judgement method of overloading wagon

Consult Fig. 1, computer program judges in advance that by vehicle sound characteristic parameter extraction computer sub, neural network vehicle classification computer sub, vehicle weight characteristic parameter extraction computer sub, neural network overload of vehicle computer sub and result dispose computer sub and form.Vehicle sound characteristic parameter extracts computer sub and carries out computing to crossing the vehicle sounds signal, extract vehicle sound characteristic value of consult volume, the vehicle sound characteristic value of consult volume composition characteristic vector that extracts is as the neural network input vector, by the neural network identification type of vehicle that has trained.The vehicle weight characteristic parameter extracts computer sub and carries out computing to crossing the Vehicular vibration signal, extract car weight vibration performance value of consult volume, car weight vibration performance value of consult volume that extracts and the type of vehicle value composition characteristic vector that obtained be as the neural network input vector, whether vehicle overloaded according to pre-determined threshold criteria by the neural network that has trained and judge in advance.The result disposes computer sub and according to pre-determined disposal method signal processing results is carried out site disposal (as: showing toll amount, the action of control magnetic card Fare Collection System, the warning of overload suspicion etc.) or by the Internet net result transmission carried out long-range disposal to the appointed place.

Vehicle sound characteristic parameter extracts and adopts the AR model to carry out model parameter calculating, calls AR model parameter counting subroutine during calculating, and vehicle feature AR model order n preestablishes.Vehicle weight vibration performance parameter adopts wavelet packet channel energy value, and the WAVELET PACKET DECOMPOSITION level preestablishes.

System at first carries out the AR model parameter value calculating of n rank to the vehicle sounds signal during operation, the parameter sets that draws constitutes the n dimension vehicle proper vector of vehicle, use the BP neural network as sorter then, the characteristic parameter vector is as input vector (n dimension), the vehicle kind is carried out vehicle identification as output vector (1 dimension) to vehicle.Afterwards the Vehicular vibration signal is carried out WAVELET PACKET DECOMPOSITION, draw each wavelet packet passage relative energy value as the vehicle load proper vector and and vehicle kind value constitute m+1 and tie up the vehicle weight proper vector, with the BP neural network as determinant, the characteristic parameter vector is as input vector (m+1 dimension), overload condition is judged as output vector (1 dimension Boolean quantity), overload of vehicle is judged in advance.The system that implements computer program carries out site disposal or transfers to remote port vehicle and the load-carrying situation result who draws.

The major technique feature of the on-line automatic pre-judgement method of overloading wagon

1. end-point detection

Have or not the judgement of vehicle sound, belong to the end-point detection problem from the angle of signal Processing.The method of end-point detection is a lot, and this method adopts the short-time energy method to detect the vehicle acoustical signal, gets rid of no vehicle sound section.The short-time energy method is a kind of basic skills that acoustical signal is analyzed, and the energy function of signal x (t) is defined as:

E(t)＝E(t)＝∫x ²(t)dt，

To x ²(t) do that definite integral obtains between at a time in the section be signal x (t) at the energy value of this section in the time, the power function that just should be in the time period on geometric meaning and the size of the area of time shaft encirclement.The energy value of signal when vehicle sound is arranged (area that function and time shaft surround) will be obviously energy value (area of function and time shaft encirclement) during greater than no vehicle sound.So we can provide an energy threshold Eth according to the energy function of signal, energy value is the audible signal section greater than Eth's, and less than Eth is noiseless signal segment.Concrete steps are as follows: at first the acoustical signal that obtains is carried out the branch frame, obtain energy value in every frame according to the energy function formula, make its short-time energy spectrum.Suppose that former frames are noise frames, calculate their the average energy value A, with 1.5 times of A as threshold value.Select the length T of suitable integrating range again.When signal at time t ₁Its amplitude is during greater than 1.5 times of A, just from t ₁Begin energy function is carried out integration, integration lengths is T, and the integrated value that obtains is exactly that measured signal is at t ₁To t ₁Energy value between the+T.If this energy value illustrates at t less than energy threshold Eth ₁To t ₁No vehicle sound between the+T, i.e. unvoiced segments, this segment signal does not enter next step processing.

2. signal Processing

Signal Pretreatment comprises denoising, frame intercepting, three contents of normalization.

(1) signal denoising

In the gatherer process of acoustical signal, can be subjected to nature and artificial various interference inevitably, before signal is further handled, be necessary acoustical signal is carried out necessary denoising, to strengthen useful signal, reduce the interference of hazardous noise signal, improve signal to noise ratio (S/N ratio), thereby reduce the difficulty of subsequent treatment.

(2) frame intercepting

In experiment, in order to guarantee the integrality of data, the data volume of the acoustic signal of system acquisition is bigger.General when vehicle by the time, measure from beginning from the sensor larger distance, when vehicle leaves larger distance till, the generally lasting tens seconds left and right sides time of image data.When deal with data, real useful that segment signal when just vehicle is near sensor.Because when vehicle distances was far away, the signal to noise ratio (S/N ratio) between vehicle noise signal and the ground unrest was lower, does not have availability.Therefore we wish and can intercept one section higher useful signal of signal to noise ratio (S/N ratio) by computing machine, so that further analyze.

The amplitude of vehicle signal should be constant concerning himself in vehicle traveling process, be that envelope is contour, but be not such in the actual measurement, because the decay of energy, amplitude is zero when remote, have an envelope peak in the time of closely, survey the envelope of curve of decay to two so peak value can appear in the actual signal that records.When we handle at vehicle signal, the intercepting of numerical point with the spectrum peak be that axis of symmetry is selected, the numerical point of taking as much as possible, to be without loss of generality.

But based on the requirement of online statistics, the processing time of voice signal is short more good more, so can not unconfinedly choose a lot of points, chooses 4096 of frame lengths at the signal peak place in the experiment and intercepts in the experiment.

(3) normalization

Automobile sound is constant to extraradial energy in the unit interval, but the size of the sound intensity is with to leave sound source far and near relevant, in order to eliminate because of the influence of factors such as distance, orientation to original noise, we carry out normalized to noise signal and power spectrum.

Normalization relates to many dissimilar computings, its purpose is to make problem to point to relativeness, make the input statistic in the time, frequency, keep as far as possible when orientation or other quantitative changes evenly, so that can on the basis of " homogeneous background ", tell this situation of signal plus noise immediately, in addition, normalized second effect of noise is to emphasize (giving prominence to) certain signal, makes it be matched with " certain human factor " that operating personnel have a preference for.Normalized the 3rd effect is to make the dynamic range of treatment capacity suitable.The present invention adopts the power normalizing, and its definition is as follows:

The power normalization function f of energy function f (x) ₀(x):

$f_0\left(x\right)=\frac{f\left(x\right)}{A}$

Wherein

$A={\left[{\∫}_{-\∞}^{\∞}{f}^2\left(x\right)\mathrm{dx}\right]}^{1/2}$

Can finish the power normalization of signal by following formula.

3.AR model parameter is calculated

The modernism that spectrum is estimated mainly is based on the parameter model of stochastic process, therefore, also can be referred to as parameter model method or abbreviation model method.The Wold decomposition theorem is thought: any generalized stationary random process all can resolve into part and definite part of a completely random.The determinacy stochastic process be one can its past the stochastic process predicted fully of a unlimited sampling value.An inference of Wold decomposition theorem is: if power spectrum is continuous fully, so any ARMA process or AR process can be with the MA procedural representations of an infinite order, and the theorem that Kolmogorov proposes has similar conclusion: any ARMA or MA process can be with the AR procedural representations of an infinite order.Native system adopts the AR model to carry out the vehicle signature analysis, and the great amount of samples of two kinds of vehicles is carried out the parameter model analysis, uses the species characteristic parameter of the mean value of vehicle model parameter as this kind vehicle.Satisfying difference equation between the output of this model and the input is:

$x\left(n\right)=-\underset{k=1}{\overset{p}{\mathrm{\Σ}}}{a}_{k}x(n-k)+u\left(n\right)---\left(1\right)$

Input stimulus u (n) is that average is zero, and variance is σ ²White noise sequence, this model is called p rank autoregressive model or abbreviates AR (p) model as, its transition function is:

$H_{\mathrm{AR}}\left(z\right)=\frac{1}{A\left(z\right)}=\frac{1}{1+\underset{k=1}{\overset{p}{\mathrm{\Σ}}}{a}_k{z}^{-k}}---\left(2\right)$

The model output power spectrum is:

$S_{\mathrm{xx}}\left(z\right)=\frac{{\mathrm{\σ}}^2}{A\left(z\right)A\left(z^{-1}\right)}$ Or $S_{\mathrm{xx}}\left(e^{\mathrm{j\ω}}\right)=\frac{{\mathrm{\σ}}^2}{{\left|A\left(e^{\mathrm{j\ω}}\right)\right|}^2}=\frac{{\mathrm{\σ}}^2}{{|1+\underset{k=1}{\overset{p}{\mathrm{\Σ}}}{a}_k{e}^{-\mathrm{j\ωk}}|}^2}---\left(3\right)$

This is an all-pole modeling.

Enough the AR model of high exponent number can be described the vehicle feature of certain vehicle operation sound fully, but not all parameter can be used as the characteristic quantity of distinguishing different automobile types, have only those just vehicle classification to be had contribution for the different automobile types mean value parameter that there were significant differences, promptly for vehicle was differentiated, some parameter was redundant.We can suppose, because individual difference makes the discreteness that certain parameter had of vehicle feature belong to normal distribution.Therefore, the work that whether has a significant difference of certain parameter of judging different automobile types can be finished by u-test.

Can be used for carrying out the vehicle proper vector that parameter sets that vehicle differentiates constitutes vehicle, Zhang Chengyi feature space.In this space, if the distance between the same model proper vector is significantly less than the distance between the different automobile types proper vector, then this space can be used as the feature space that vehicle is differentiated.Whether the space of opening in order to investigate is effective for vehicle classification, as sorter, the characteristic parameter vector is as input vector with the BP neural network for we, and the vehicle kind is as output vector, the sample of being gathered is carried out vehicle classification, to detect the effect that the different parameters set is differentiated for vehicle.

4.BP neural network classification

Consult Fig. 9, because neural network has the unexistent advantage of some conventional arts: good fault-tolerant ability, classification capacity are strong, parallel processing capability and self-learning capability.Thereby adopting the neural network recognition method is a kind of good selection.Neural network commonly used at present mainly contains BP network, Hopfield network, Ko-honen network etc., because the complicacy of neural network self, select for use which kind of type network not have optimized mode, mainly decide by the sample type of classifying, quantity at neural network.Native system is used comparatively ripe at present and widely used feed-forward (Back Propagation is called for short BP) network.

The feed-forward network typically refers to the multilayer feedforward neural network based on error backpropagation algorithm (BP algorithm).At present in the practical application of artificial neural network, the neural network model of the overwhelming majority is to adopt network and its version, it also is the core of feedforward network, has embodied the part of the elite of artificial neural network.There is nearly 90% Application of Neural Network to be based on the BP algorithm according to statistics.Different with perceptron and linear neural network is that the transport function that the neuron of BP network adopts is sigmoid type differentiable function normally, so can realize any Nonlinear Mapping between input and output.

The BP network is made of input layer, hidden layer and output layer, and the neuron between the adjacent layer is totally interconnected, does not have connection with the neuron in one deck.Shown in Figure 9 is the realization of introducing the BP algorithm with three layers of BP network with middle layer (hidden layer).The unit number of input layer, middle layer and output layer is respectively N, L and M, is input as X ₀, X ₁..., X _N-1The middle layer is output as h ₀, h ₁..., h _L-1The actual y that is output as of network ₀, y ₁..., y _M-1d ₀, d ₁..., d _M-1Expression training sample desired output.Input block i is V to the weights of temporary location j _IjTemporary location j is W to output unit k weights _JkUse θ _kAnd ψ _jThe threshold values of representing output unit and temporary location respectively.

The realization of BP network is divided into two stages, i.e. learning phase and cognitive phase.At learning phase, the sample that input will be learnt, according to the weight of network initial setting, threshold value and transition function calculate the neuronic output of each layer, and this upwards carries out from bottom; Determine whether weight, threshold value are made amendment by the error between ideal output and the top output, this modification has been carried out downwards from high level: two processes are carried out repeatedly, and till network convergence, this is a learning phase.The study of weight is exactly constantly to adjust each neuronic strength of joint, makes it can approach pairing output on the meaning of least square; Cognitive phase, calculate for given input this moment, obtains recognition result.

5. wavelet packet calculation of parameter

(1) Wavelet Packet Theory

Consult Figure 10 and Figure 11, wavelet packet analysis (Wavelet Packet Analysis) is a kind ofly can provide a kind of meticulousr analytical approach for signal, it is divided frequency band at many levels, there is not the HFS of segmentation to continue to divide to wavelet analysis, and can be according to the analyzed signal feature, adaptive selection frequency band corresponding makes it to be complementary with signal spectrum.One frequency spatial decomposition is as shown in Figure 10 and Figure 11 when desirable small echo and wavelet packet.

This shows that wavelet packet can decompose the high frequency Subspace Decomposition again, it is very different with the subdivision of small echo to the space.From Figure 10 and Figure 11 as can be seen, the y-bend subtree of each wavelet packet tree all corresponding initial subspace, for the signal of finite energy, wavelet packet basis can provide the method for a kind of specific signal coding and reconstruction signal according to each sub-band information.

(2) the wavelet packet characteristic of vehicle load vibration signal

Feature extraction is the key problem in pattern-recognition or the classification.Concerning identification or classification, key does not lie in complete description scheme, but effective characteristic of division in the extraction pattern.So-called effectively characteristic of division is exactly the bigger feature of different mode class difference.But these features are difficult for observed in the primitive character territory usually or detection.Feature extraction is exactly the method by conversion, and these key characters are shown at transform domain, removes the insignificant information of classifying.The pattern that generally is identified or classifies all is the pattern of non-stationary or jump signal, as voice, radar and seismic signal etc., in these signals, usually comprise when long the low frequency and the signal of high frequency different scale in short-term, the feature that is used for classifying often is included in local time frequency signal, thereby the wavelet packet algorithm has good application potential in feature extraction.

Feature extraction is the key problem in pattern-recognition or the classification, directly has influence on the design and the performance of sorter.Key does not lie in complete description scheme concerning identification or classification, but effective characteristic of division in the extraction pattern.So-called effectively characteristic of division is exactly the bigger feature of different mode classification difference.But these features are difficult for observed in the primitive character territory usually or detection.Feature extraction is exactly the method by conversion (normally linear transformation), and these important features are shown at transform domain, removes the insignificant information of classifying, and so original higher dimensional space is become the feature space of low-dimensional.

The wavelet packet calculation of parameter:

[1] each sample is carried out three layers of WAVELET PACKET DECOMPOSITION, extract the signal characteristic X of the 3rd layer of 8 frequency channel _3j(j=0,1, ..., 7).

[2], extract the signal of each frequency band range to the reconstruct of WAVELET PACKET DECOMPOSITION coefficient.With S _3jExpression X _3jReconstruction signal.

[3] ask the gross energy of each band signal.Because input signal is a random signal, its output also is random signal.If S _3j(j=0,1, ..., 7) and corresponding energy is E _3j(j=0,1 ..., 7), then have:

$E_{3j}=\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{\left|x_{\mathrm{jk}}\right|}^2$

Wherein, x _Jk(j=0,1 ..., 7; K=1,2 ..., n) expression reconstruction signal S _3jThe amplitude of discrete point.

[4] structure energy vector space.

Structure energy vector is:

T＝[E ₃₀，E ₃₁，E ₃₂，E ₃₃，E ₃₄，E ₃₅，E ₃₆，E ₃₇]

Because the one dimension of neural network input is represented a feature, when the input of neural network is multidimensional, want recognized patterns that a plurality of features are arranged, when the data of these a plurality of features differ greatly, during as several magnitude, just need normalization, become same order, in case the low feature of some numerical value is submerged, the while is also for convenient data processing, the energy vector is carried out normalized, within data map to 0～1 scope.Method for normalizing has multiple, and native system adopts energy normalized, and promptly each parameter quadratic sum is one after the normalization.

Order: $E={\left(\underset{j=0}{\overset{7}{\mathrm{\Σ}}}{\left|E_{3j}\right|}^2\right)}^{\frac{1}{2}}$

$T^{\′}=[\frac{E_{30}}{E},\frac{E_{31}}{E},\frac{E_{32}}{E},\frac{E_{33}}{E},\frac{E_{34}}{E},\frac{E_{35}}{E},\frac{E_{36}}{E},\frac{E_{37}}{E}]$

Energy vector after vector T ' the be normalization.Order $\frac{E_{3j}}{E}=T_j,$ Then:

T′＝[T ₀，T ₁，T ₂，T ₃，T ₄，T ₅，T ₆，T ₇]

Between sound that sends when the different load-carryings of automobile and its operation and the vibration signal close getting in touch arranged.

The 3rd layer of wavelet-packet energy parameter of each sample extraction during to the different load-carrying of vehicle obtained the average and the variance of corresponding energy parameter.Have only those parameters therefore, must judge which parameter of different load conditions has significant difference as characteristic parameter in order to identification with significant difference.Whether certain parameter of judging different load conditions has the work of significant difference can be finished by u-test.

Two. the design concept and the major technique feature of the on-line automatic pre-judgement of overloading wagon system

Consult Fig. 2, the on-line automatic pre-judgement of overloading wagon system is by signal acquisition module, signal processing module, and microcomputer interface module three parts are formed.Signal acquisition module is made up of a sound transducer, vibration transducer, first modulate circuit, second modulate circuit and one two a passage A/D converter, and sound, the vibration signal be responsible for when vehicle crossed are nursed one's health, gathered and sampled result is sent in the digital signal processor DSP.Signal processing module is that the FLASH of AM29LV800B, the random access device RAM that model is IS61LV12816 form by digital signal processor DSP, model, this part data that signal acquisition module is sent here is handled, and comprises vehicle operating voice signal and vibration signal end-point detection, to choosing of measurement data and normalization, to extraction of oversampled signals etc.The microcomputer interface module is that the high speed USB chip of Cy7c68013 is formed by model, and the system of being responsible for is connected with microcomputer, and realization result disposal is carried out at remote port.

One end of first modulate circuit links to each other with sound transducer, the other end is received the pin AINP of A/D converter, other control line (output terminal) and the model of A/D converter is that the digital signal processor of TMS320VC5402 is connected, so that carrying out with model is that the digital signal processor of TMS320VC5402 is communicated by letter, model is that the digital signal processor DSP of TMS320VC5402 is that the FLASH of AM29LV800B, the random access device RAM that model is IS61LV12816 link to each other with model, so that carry out the storage of data and program, link to each other with PC by the microcomputer interface module then.Signal Processing is carried out in model is the digital signal processor DSP of TMS320VC5402.Signal Processing work is finished at PC or in model is the digital signal processor DSP of TMS320VC5402 according to system's set-up mode, and promptly the result disposes at the scene or remote port carries out.

II. the concrete enforcement of the on-line automatic pre-judgement method and system of overloading wagon

One. the on-line automatic pre-judgement method of overloading wagon

1) the pre-judgement system of the on-line automatic pre-judgement method of overloading wagon is implemented in installation and debugging.Especially sound transducer is installed in road side or top, vibration transducer is installed in the road side and well contacts (adopting bolt that method or bonding installation method are installed) in the road surface.

2) utilize the vibration signal of vibration transducer collection vehicle in travelling;

3) utilize the method judgement of end-point detection to have or not the vehicle that travels.

Have or not the judgement of vehicle sound, belong to the end-point detection problem from the angle of signal Processing.This method adopts the short-time energy method to detect the driving vehicle acoustical signal, gets rid of no vehicle sound section.Can provide an energy threshold Eth according to the energy function of signal, energy value is the audible signal section greater than Eth's, proof has the operational vehicle signal to enter next step processing, less than Eth is noiseless signal segment, proof does not have the operational vehicle signal, this segment signal does not enter next step and handles, and turns back to the vibration signal of vibration transducer collection vehicle in travelling.

4) voice signal from the driving vehicle gathered judges whether to be goods carrying vehicle.

Enter this step for energy value greater than Eth for the audible signal section, operational vehicle is promptly arranged, should further judge whether it is goods carrying vehicle.Judge whether for goods carrying vehicle be to finish by the vehicle identification step.Being, entering next step (the pre-Signal Pretreatment judged in the step of overload), is not to return the step that vibration transducer is gathered vibration signal.

5) when being goods carrying vehicle by the judgement of vehicle identification step, promptly the vibration signal to goods carrying vehicle carries out signal denoising, frame intercepting and normalized Signal Pretreatment.

Signal denoising is in order to strengthen useful signal, to reduce the interference of hazardous noise signal, raising signal to noise ratio (S/N ratio), thereby the difficulty of minimizing subsequent treatment.Why taking frame intercepting step, is because when deal with data, real useful that segment signal when just vehicle is near sensor.Because when vehicle distances was far away, the signal to noise ratio (S/N ratio) between vehicle noise signal and the ground unrest was lower, does not have availability.Therefore we wish and can intercept one section higher useful signal of signal to noise ratio (S/N ratio) by computing machine, so that further analyze.Automobile sound is constant to extraradial energy in the unit interval, but the size of the sound intensity is with to leave sound source far and near relevant, in order to eliminate because of the influence of factors such as distance, orientation to original noise, we carry out normalized to noise signal and power spectrum.

6) signal through Signal Pretreatment is carried out the wavelet packet calculation of parameter:

(1) each sample is carried out three layers of WAVELET PACKET DECOMPOSITION, extract the signal characteristic X of the 3rd layer of 8 frequency channel _3j(j=0,1, ..., 7),

(2), extract the signal of each frequency band range to the reconstruct of WAVELET PACKET DECOMPOSITION coefficient.With S _3jExpression X _3jReconstruction signal;

(3) ask the gross energy of each band signal, because input signal is a random signal, its output also is random signal, S _3j(j=0,1, ..., 7) and corresponding energy is E _3j(j=0,1 ..., 7), then have:

$E_{3j}=\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{\left|x_{\mathrm{jk}}\right|}^2$

Wherein, x _Jk(j=0,1 ..., 7; K=1,2 ..., n) expression reconstruction signal S _3jThe amplitude of discrete point;

(4) structure energy vector space

Structure energy vector is:

T＝[E ₃₀，E ₃₁，E ₃₂，E ₃₃，E ₃₄，E ₃₅，E ₃₆，E ₃₇]；

(5) adopt energy normalized, promptly each parameter quadratic sum is that one method is carried out normalized to the energy vector after the normalization, within data map to 0～1 scope,

Order: $E={\left(\underset{j=0}{\overset{7}{\mathrm{\Σ}}}{\left|E_{3j}\right|}^2\right)}^{\frac{1}{2}}$

$T^{\′}=[\frac{E_{30}}{E},\frac{E_{31}}{E},\frac{E_{32}}{E},\frac{E_{33}}{E},\frac{E_{34}}{E},\frac{E_{35}}{E},\frac{E_{36}}{E},\frac{E_{37}}{E}]$

Energy vector after vector T ' the be normalization, $\frac{E_{3j}}{E}=T_j,$ Then:

T′＝[T ₀，T ₁，T ₂，T ₃，T ₄，T ₅，T ₆，T ₇]；

7) the BP network overload is judged in advance

The realization of BP network is divided into two stages, i.e. learning phase and cognitive phase:

(1) sample that will learn in the learning phase input, according to the weight of network initial setting, threshold value and transition function calculate the neuronic output of each layer, and this upwards carries out from bottom; Determine whether weight, threshold value are made amendment by the error between ideal output and the top output, this modification has been carried out downwards from high level: two processes are carried out repeatedly, till network convergence, this stage is exactly constantly to adjust each neuronic strength of joint, makes it can approach pairing output on the meaning of least square.Learning phase takes offline mode to carry out.

(2) at cognitive phase, calculate for given input, judged the result in advance: goods carrying vehicle is overload or non-overloading.

8) judge result's on-the-spot or long-range disposal in advance

In this step, the result disposes computer sub and according to pre-determined disposal method the pre-result of judgement that overloads is carried out site disposal, as: show toll amount, the action of control magnetic card Fare Collection System, the warning of overload suspicion etc.; Or will overload by the Internet net and pre-judge that result transmission carries out long-range disposal to the appointed place.

Two. the on-line automatic pre-judgement of overloading wagon system

The on-line automatic pre-judgement of overloading wagon system is made up of signal acquisition module, signal processing module and microcomputer interface module three parts.

Consult Fig. 2 and Fig. 5, described signal acquisition module is made up of sound transducer, vibration transducer, first modulate circuit, second modulate circuit and two passage A/D converters, sound transducer is connected with the input end electric wire of first modulate circuit, the output terminal of first modulate circuit and model are that the pin AINP electric wire of the two passage A/D converters of THS10064 is connected, vibration transducer is connected with the input end electric wire of second modulate circuit, and the output terminal of second modulate circuit and model are that the pin BINP electric wire of the two passage A/D converters of THS10064 is connected.Model is that the output terminal of the two passage A/D converters of THS10064 is connected with the signal processing module electric wire, be that model is that pin CS0/, CS1, DATA_AV, GONV_GLK and the D0-D9 of the two passage A/D converters of THS10064 is that pin IOSTRB#, A15, BIO/, BCLKX0 and the D0-D9 electric wire of the digital signal processor DSP of TMS320VC5402 is connected with model respectively, model is that the pin R/W# of the two passage A/D converters of THS10064 is that the pin IS# of the digital signal processor DSP of TMS320VC5402 is connected with the R/W# electric wire by one with door and model.The pin RD/ of the two passage A/D converters of THS10064 connects the 3.3V power supply.

Consult the described signal processing module of Fig. 2 and comprise that the model that computer program is installed is that the digital signal processor DSP of TMS320VC5402, the FLASH that model is AM29LV800B, random access device RAM, the model that model is IS61LV12816 are the complex programmable logic device (CPLD) of XC95144.Model is that digital signal processor DSP and the model of TMS320VC5402 is the two passage A/D converters of THS10064, FLASH that model is AM29LV800B, model is IS61LV12816 random access device RAM is respectively electric wire and is connected, and model is that the digital signal processor DSP of TMS320VC5402 is that the complex programmable logic device (CPLD) of XC95144 is connected with microcomputer interface module electric wire by model.Model is that the complex programmable logic device (CPLD) of XC95144 is that the FLASH of AM29LV800B, the random access device RAM electric wire that model is IS61LV12816 are connected with model respectively.

Consult Fig. 2 and Fig. 3, model is that pin D (0:15), A (0:15) and the HDO of the digital signal processor DSP of TMS320VC5402 is that pin I/O (0:15), the A (0:15) of the random access device RAM of IS61LV12816 is connected with the A16 electric wire with model respectively, and model is that pin R/W, MSTRB#, DS# and the XF of the digital signal processor DSP of TMS320VC5402 is that the pin WE/ of the random access device RAM of IS61LV12816 is connected with the CE/ electric wire by two with door and model respectively.Model is pin OE/, UB, the LB/ ground connection (GND) of the random access device RAM of IS61LV12816.

Consult Fig. 2 and Fig. 4, model is that pin D (0:15), A (0:15), BIO# and the A (16:18) of the digital signal processor DSP of TMS320VC5402 is that pin D (0:15), A (0:15), RY/BY# and A (16:18) electric wire of the FLASH of AM29LV800B is connected with model respectively, and model is that pin R/W#, MSTRB#, DS# and the XF of the digital signal processor DSP of TMS320VC5402 is that pin OE#, the WE# of the FLASH of AM29LV800B with CE# electric wire be connected with three with door and model by two not gates respectively.Model is that pin RST#, the BYTE# of the FLASH of AM29LV800B connects the 3.3V power supply.

Consult Fig. 2, Fig. 6 to Fig. 8, realize that model is that digital signal processor DSP and the model of TMS320VC5402 is that the high speed USB chip electric wire of Cy7c68013 is connected, for this reason, model is the pin PS of the digital signal processor DSP of TMS320VC5402, DS, IS, RW, XF, IOSTRB, MSTRB, BIO, READY, INT0, INT1, A (0-15) and D (0-7) are at first respectively the pin CPLD_PS of the complex programmable logic device (CPLD) of XC95144 with model, CPLD_DS, CPLD_IS, CPLD_RW, CPLD_XF, IOSTRB, MSTRB, CPLD_BIO, READY, INT0, INT1, A (0-15) is connected with D (0-7) electric wire, then, model is the pin CY_IFCLK of the complex programmable logic device (CPLD) of XC95144, CY_CLKOUT, CY_SLWR, CY_SLRD, CY_CTL (0-2), PA (0-2), USB_INT1, FIFOADR (0-1), PKTEND, PA7, CY_PA (0-7) and CY_PD (0-7) are respectively the high speed USB pin of chip CY_IFCLK of Cy7c68013 again with model, CY_CLKOUT, CY_SLWR, CY_SLWR, CY_CTL (0-2), PA (0-2), USB_INT1, FIFOADR (0-1), PKTEND, PA7, CY_PA (0-7) is connected with CY_PD (0-7) electric wire, model is the pin A (0-18) of the complex programmable logic device (CPLD) of XC95144, D (0-8), FLASH_CE is respectively again the pin A (0-18) of the FLASH of AM29LV800B with model, D (0-8), the FLASH_CE electric wire connects, and model is the pin A (0-18) of the complex programmable logic device (CPLD) of XC95144, D (0-8), RAM_CE, RAM_WE, RAM_RD is respectively again the pin A (0-18) of the random access device RAM of IS61LV12816 with model, D (0-8), RAM_CE, RAM_WE, the RAM_RD electric wire connects.

Claims (7)

1. the on-line automatic pre-judgement method of overloading wagon that adopts computer program comprises the vehicle identification step, it is characterized in that, the on-line automatic pre-judgement method of overloading wagon also comprises the pre-judgement of overload step, and described overload judges that in advance step comprises following flow process:

1) utilizes the vibration signal of vibration transducer collection vehicle in travelling;

2) utilize the method judgement of end-point detection to have or not the vehicle that travels, do not have the vehicle that travels, signal does not enter next step;

3) voice signal from the driving vehicle gathered judges whether to be goods carrying vehicle, is not goods carrying vehicle, and signal does not enter next step;

4) when being goods carrying vehicle by the judgement of vehicle identification step, promptly the vibration signal to goods carrying vehicle carries out signal denoising, frame intercepting and normalized Signal Pretreatment;

5) signal through Signal Pretreatment is carried out the wavelet packet calculation of parameter:

(1) each sample is carried out three layers of WAVELET PACKET DECOMPOSITION, extract the signal characteristic X of the 3rd layer of 8 frequency channel _3j(j=0,1, ..., 7);

(2), extract the signal of each frequency band range, with S to the reconstruct of WAVELET PACKET DECOMPOSITION coefficient _3jExpression X _3jReconstruction signal;

(3) ask the gross energy of each band signal, because input signal is a random signal, its output also is random signal, S _3j(j=0,1, ..., 7) and corresponding energy is E _3j(j=0,1 ..., 7), then have:

$E_{3j}=\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{\left|x_{\mathrm{jk}}\right|}^2$

Wherein, x _Jk(j=0,1 ..., 7; K=1,2 ..., n) expression reconstruction signal S _3jThe amplitude of discrete point;

(4) structure energy vector space

Structure energy vector is:

T＝[E ₃₀，E ₃₁，E ₃₂，E ₃₃，E ₃₄，E ₃₅，E ₃₆，E ₃₇]；

(5) adopt energy normalized, promptly each parameter quadratic sum is that one method is carried out normalized to the energy vector after the normalization, within data map to 0～1 scope,

Order: $E={\left(\underset{j=0}{\overset{7}{\mathrm{\Σ}}}{\left|E_{3j}\right|}^2\right)}^{\frac{1}{2}}$

$T^{\′}=[\frac{E_{30}}{E},\frac{E_{31}}{E},\frac{E_{32}}{E},\frac{E_{33}}{E},\frac{E_{34}}{E},\frac{E_{35}}{E},\frac{E_{36}}{E},\frac{E_{37}}{E}]$

Energy vector after vector T ' the be normalization, $\frac{E_{3j}}{E}=T_j,$ Then:

T′＝[T ₀，T ₁，T ₂，T ₃，T ₄，T ₅，T ₆，T ₇]；

6) the BP network overload is judged in advance

(1) sample that will learn in the learning phase input, till network convergence, this stage is exactly constantly to adjust each neuronic strength of joint, makes it can approach pairing output on the meaning of least square, and learning phase takes offline mode to carry out;

(2) at cognitive phase, calculate for given input, judged the result in advance.

2. system that implements the on-line automatic pre-judgement method of the described overloading wagon of claim 1, it is characterized in that, this system is made up of signal acquisition module, signal processing module and microcomputer interface module three parts, signal acquisition module is that electric wire is connected with signal processing module, signal processing module is that electric wire is connected with the microcomputer interface module

Described signal acquisition module is by sound transducer, vibration transducer, first modulate circuit, second modulate circuit and model are that the two passage A/D converters of THS10064 are formed, sound transducer is connected with the input end electric wire of first modulate circuit, the output terminal of first modulate circuit and model are that the pin AINP electric wire of the two passage A/D converters of THS10064 is connected, vibration transducer is connected with the input end electric wire of second modulate circuit, the output terminal of second modulate circuit and model are that the pin BINP electric wire of the two passage A/D converters of THS10064 is connected, and model is that the output terminal of the two passage A/D converters of THS10064 is connected with the signal processing module electric wire;

Described signal processing module comprises that the model that computer program is installed is digital signal processor DSP, the FLASH that model is AM29LV800B, the random access device RAM that model is IS61LV12816 of TMS320VC5402, and model is that digital signal processor DSP and the model of TMS320VC5402 is the two passage A/D converters of THS10064, FLASH that model is AM29LV800B, model is IS61LV12816 random access device RAM is respectively electric wire with the microcomputer interface module and is connected;

Described microcomputer interface module is that the high speed USB chip of Cy7c68013 is formed by model, the input end of microcomputer interface module and model are that an end electric wire of the digital signal processor DSP of TMS320VC5402 is connected, and the output terminal of microcomputer interface module is connected with the input end electric wire of PC.

3. according to the system of the on-line automatic pre-judgement method of the described overloading wagon of the described enforcement of claim 2 claim 1, it is characterized in that the digital signal processor DSP that the described model that computer program is installed is TMS320VC5402 is meant: the computer sub of being extracted by vehicle sound characteristic parameter is installed, the computer sub of neural network vehicle classification, the computer sub that the vehicle weight characteristic parameter extracts, the model of the computer program that the computer sub that pre-computer sub of judging of neural network overload of vehicle and result dispose is formed is the digital signal processor DSP of TMS320VC5402.

4. according to the system of the on-line automatic pre-judgement method of the described overloading wagon of the described enforcement of claim 2 claim 1, it is characterized in that, described model is that digital signal processor DSP and the model of TMS320VC5402 is that the random access device RAM electric wire of IS61LV12816 is connected and is meant: model is the pin D (0:15) of the digital signal processor DSP of TMS320VC5402, A (0:15) and HD0 are respectively the pin I/O (0:15) of the random access device RAM of IS61LV12816 with model, A (0:15) is connected with the A16 electric wire, and model is the pin R/W of the digital signal processor DSP of TMS320VC5402, MSTRB#, DS# and XF are that the pin WE/ of the random access device RAM of IS61LV12816 is connected with the CE/ electric wire by two with door and model respectively.

5. according to the system of the on-line automatic pre-judgement method of the described overloading wagon of the described enforcement of claim 2 claim 1, it is characterized in that, described model is that digital signal processor DSP and the model of TMS320VC5402 is that the FLASH electric wire of AM29LV800B is connected and is meant: model is the pin D (0:15) of the digital signal processor DSP of TMS320VC5402, A (0:15), BIO# and A (16:18) are respectively the pin D (0:15) of the FLASH of AM29LV800B with model, A (0:15), RY/BY# is connected with A (16:18) electric wire, and model is the pin R/W# of the digital signal processor DSP of TMS320VC5402, MSTRB#, DS# and XF are the pin OE# of the FLASH of AM29LV800B with door with model by two not gates and three respectively, WE# is connected with the CE# electric wire.

6. according to the system of the on-line automatic pre-judgement method of the described overloading wagon of the described enforcement of claim 2 claim 1, it is characterized in that, described model is that digital signal processor DSP and the model of TMS320VC5402 is that the two passage A/D converter electric wires of THS10064 are connected and are meant: model is the pin IOSTRB# of the digital signal processor DSP of TMS320VC5402, A15, BIO/, BCLKX0 and D0-D9 are respectively the pin CS0/ of the two passage A/D converters of THS10064 with model, CS1, DATA_AV, GONV_GLK is connected with the D0-D9 electric wire, and model is that pin IS# and the R/W# of the digital signal processor DSP of TMS320VC5402 is that the pin R/W# electric wire of the two passage A/D converters of THS10064 is connected by one with door and model.

7. according to the system of the on-line automatic pre-judgement method of the described overloading wagon of the described enforcement of claim 2 claim 1, it is characterized in that, described model is that the digital signal processor DSP of TMS320VC5402 is connected with microcomputer interface module electric wire and is meant: model is that digital signal processor DSP and the model of TMS320VC5402 is that the high speed USB chip electric wire of Cy7c68013 is connected, for this reason, model is the pin PS of the digital signal processor DSP of TMS320VC5402, DS, IS, RW, XF, IOSTRB, MSTRB, BIO, READY, INT0, INT1, A (0-15) and D (0-7) are respectively the pin CPLD_PS of the complex programmable logic device (CPLD) of XC95144 with model earlier, CPLD_DS, CPLD_IS, CPLD_RW, CPLD_XF, IOSTRB, MSTRB, CPLD_BIO, READY, INT0, INT1, A (0-15) is connected with D (0-7) electric wire, model is the pin CY_IFCLK of the complex programmable logic device (CPLD) of XC95144, CY_CLKOUT, CY_SLWR, CY_SLRD, CY_CTL (0-2), PA (0-2), USB_INT1, FIFOADR (0-1), PKTEND, PA7, CY_PA (0-7) and CY_PD (0-7) are respectively the high speed USB pin of chip CY_IFCLK of Cy7c68013 again with model, CY_CLKOUT, CY_SLWR, CY_SLRD, CY_CTL (0-2), PA (0-2), USB_INT1, FIFOADR (0-1), PKTEND, PA7, CY_PA (0-7) is connected with CY_PD (0-7) electric wire, model is the pin A (0-18) of the complex programmable logic device (CPLD) of XC95144, D (0-8), FLASH_CE is respectively again the pin A (0-18) of the FLASH of AM29LV800B with model, D (0-8), the FLASH_CE electric wire connects, and model is the pin A (0-18) of the complex programmable logic device (CPLD) of XC95144, D (0-8), RAM_CE, RAM_WE, RAM_RD is respectively again the pin A (0-18) of the random access device RAM of IS61LV12816 with model, D (0-8), RAM_CE, RAM_WE, the RAM_RD electric wire connects.

Images