CN102506983A - Weighing error automatic compensation method of vehicle scale - Google Patents

Abstract

The invention discloses a weighing error automatic compensation method of a vehicle scale, which comprises three aspects of composite neural network construction, neural network off-line training and vehicle scale on-line weighing. First, constructing three sub-neural networks according to priori knowledge; then utilizing standard weights of different tonnage and a weighing signal collecting circuit to train a sample and transmitting the sample to an upper computer, utilizing training software to finish off-line training of the three sub-neural networks, obtaining a corresponding parameter, and downloading the parameter to a lower computer to prepare for the on-line measurement of the vehicle scale; when the vehicle scale performs on-line measurement, utilizing the lower computer to first obtain a weighing signal vector through the weighing signal collecting circuit to serve as input of the three sub-neural networks and calculating the output of each sub-neural network; roughly estimating the weight of a tested load, and automatically obtaining the output weight of a composite neural network; and merging the output weighing of the sub-neural networks, obtaining a final weighing result, and simultaneously finishing the weighing error compensation of the vehicle scale. The weighing error automatic compensation method of the vehicle scale can achieve automatic compensation of an unbalance loading error and a linear error of the vehicle scale so as to greatly improve accuracy of the weighing result.

Description

Truck scale weighting error automatic compensating method

Technical field

Truck scale weighting error automatic compensating method; The present invention relates to a kind of truck scale pin-point accuracy method with error compensation of weighing; Specifically; Relate to a kind of neural net method that utilizes and carry out the truck scale high precision method with error compensation of weighing, belong to weighing system and detect and field of information processing, also can be applicable to other multisensor syste.

Technical background

Truck scale (comprising weighing apparatus, platform scale etc. on the ground) is as the important branch of weighing apparatus; Range of weighing by several tons to the hundreds of ton even go up kiloton; It is the main weighing-appliance that factories and miness, businessman etc. are used for the bulk supply tariff metering; Its department such as trade, communications and transportation, industrial and mining enterprises that are widely used in storing in a warehouse, along with the development of commercial production and communications and transportation, the demand of truck scale is with increasing.Existing truck scale also has multiple functions such as monitoring, management except that function of measuring, its modern management to enterprise is significant.

Owing to reasons such as prices, present analog electronic vehicle weighing apparatus occupies the leading position in truck scale market.Existing analog electronic vehicle weighing apparatus mainly is made up of load-bearing force transmission mechanism (scale body), simulation LOAD CELLS, weighting display instrument three big master units; Can accomplish the basic function of weighing of truck scale thus; It utilizes the sensor output signal accumulate mode, adds up in the output set of signals with each road LOAD CELLS in the simulation terminal box, obtains one and the proportional voltage signal of tested quality of loads; After signal condition, A/D conversion; Handle the acquisition weighing results by single-chip microcomputer, send demonstration, communication, accomplish weighing of tested load.The accuracy of truck scale weighing results is influenced by uneven loading error and linearity error etc.Uneven loading error is because truck scale receives various effect of non-linear, when tested load is on the truck scale loading end diverse location, and the inconsistent and error that produces of weighing result; Linearity error is because the characteristic of each road LOAD CELLS is inconsistent, causes the input and output and nonideal linear relationship of truck scale, thus the weighting error that produces.The uneven loading error of existing truck scale is what to separate with the linearity error compensation process; The uneven loading error compensation method has: resistor in the truck scale terminal box is regulated repeatedly by manual work in (1); Change the sensor passage gain of every road, the compensation uneven loading error, this method manually-operated is loaded down with trivial details; Inefficiency, compensation effect is poor; (2) utilize the position sensor test load to be carried in the position on the carrier; Realize the uneven loading error compensation and the fusion of weighing according to positional information, this method is fit to the electronic scale of little carrier, for the truck scale of big carrier; Require further improvement; This method need increase position sensor simultaneously, has increased the cost of truck scale, and is not easy to Project Realization; (3) adopt multiple linear regression analysis method; Promptly utilize standard test weight to be carried on each stressed fulcrum of truck scale (being LOAD CELLS); Obtain one group of linear equation, obtain the gain coefficient of each LOAD CELLS passage by finding the solution this equation, thereby realize the uneven loading error compensation; This method does not have the vehicle into account various non-linear factors influence of weighing, and compensation effect is not obvious; (4) adopt digital weighing sensor to realize the uneven loading error compensation, this method does not have the vehicle into account various non-linear factors influences of weighing, and compensation effect is poor, and the digital weighing sensor cost is high simultaneously.Existing truck scale linearity error compensation is after the uneven loading error compensation is accomplished; Utilize following method to accomplish: at first to utilize the standard test weight of Different Weight to be carried on the truck scale body successively; Obtain target weighing results and actual weighing results; Then actual weighing results is doubly taken advantage of correction factor, make it to equal the target weighing results.This method is based on the truck scale input-output on the basis that is linear relationship, differ bigger with actual, so compensation effect is relatively poor.

Summary of the invention

For overcoming the deficiency of prior art; The object of the present invention is to provide a kind of truck scale pin-point accuracy to weigh and error compensating method based on complex neural network; This method can be accomplished the compensation of truck scale weighting error automatically, exactly, improves the weighing results accuracy.

The factor affecting such as dispersiveness of internal stress that produces in rigidity and intensity, the truck scale processing that receives carrier and the installation process and mechanical deformation and scale error, LOAD CELLS sensitivity; When same tested load is carried in the carrier diverse location; Weighing results is inconsistent; Be that weighing results is relevant with the load(ing) point position, have uneven loading error.Truck scale has the multichannel LOAD CELLS, and the input-output characteristic of each road LOAD CELLS, sensitivity etc. are inconsistent, causes the input and output and nonlinear relationship of truck scale, thereby produces linearity error.The national standard of truck scale calibrating " JJG539-97 numeral self-indicating scale vertification regulation " regulation, the limits of error of the different weighing sections of truck scale are different, and when therefore measuring different loads, the accuracy that requires is different.

Based on above analysis, the truck scale weighting error automatic compensating method that the present invention proposes utilizes complex neural network to approach the nonlinear function between the truck scale input-output, weighs and error compensation thereby accomplish the truck scale pin-point accuracy.This method comprises complex neural network structure, neural network off-line training and three aspects of truck scale on-line weighing, and is specific as follows:

(1) complex neural network structure

According to national standard " the JJG539-97 numeral self-indicating scale vertification regulation " regulation of truck scale calibrating, the limits of error E of the different weighing sections of truck scale (generally being divided into 3 weighing sections) _PmaxDifference, promptly

$E_{p\max }=\left\{\begin{array}{cc}0.5e& M\&le;500e\\ 1.0e& 500e<M\&le;2000e\\ 1.5e& 2000e<M\&le;M_{\max }\end{array}\right.---\left(1\right)$

In the formula, M is the weighing of truck scale; M _MaxBe the maximum weighing of truck scale; E is a truck scale calibrating scale division value.Can know by formula (1), the weighing section that truck scale is different, its limits of error are different.Therefore according to this error character; Construct 3 sub neural networks (sub neural network 1, sub neural network 2 and sub neural network 3); Each sub neural network is responsible for the error compensation of corresponding weighing section; Each sub neural network structure is identical, is the three-layer network of N input 1 output, and sub neural network can be BP neural network, radial basis function neural network (RBFNN) or other feedforward neural network.

(2) neural network off-line training

The neural network off-line training comprises the training of sub neural network 1, sub neural network 2 and sub neural network 3.(weight of the accurate counterweight of bidding is y to utilize the standard test weight of different tonnages _j, j=1,2 ..., K), being carried in the diverse location on the truck scale loading end respectively, system's slave computer obtains N road LOAD CELLS output signal S through the separate weighing-up wave Acquisition Circuit in N road _i(i=1,2 ..., N), after the data pre-service, obtain weighing-up wave vector X _j(j=1,2 ..., K), and composing training sample (X _j, y _j), promptly

$(X_j,y_j)=\left(\begin{array}{ccccc}{x}_{11}& x_{21}& ...& x_{N1}& y_1\\ x_{12}& x_{22}& ...& x_{N2}& {\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">y</figure-callout>}}_2\\ .& .& .& .& .\\ .& .& .& .& .\\ .& .& .& .& .\\ x_{1j}& x_{2j}& ...& x_{\mathrm{Nj}}& y_j\\ .& .& .& .& .\\ .& .& .& .& .\\ .& .& .& .& .\\ x_{1K}& x_{2K}& ...& x_{\mathrm{NK}}& y_K\end{array}\right)---\left(2\right)$

Slave computer is sent to host computer (being PC) with training sample through RS232 or other serial communication interface, and antithetical phrase neural network 1, sub neural network 2 and sub neural network 3 carry out off-line training respectively to utilize the training software of host computer.Training is accomplished the back and is obtained each sub neural network parameter, and pass through serial communication interface with these parameter downloads to slave computer, for the truck scale on-line measurement is prepared.

The data of weighing pre-service adopt a kind of digital filtering method to accomplish, Q LOAD CELLS output voltage of this method continuous sampling, and it is obtained S by series arrangement from small to large _Im, remove forward and backward each M sampled value, to all the other Q-2M S _ImCarry out mean filter, promptly

$x_i=\frac{1}{Q-2M}\underset{m=M+1}{\overset{Q-M}{\mathrm{\&Sigma;}}}{S}_{\mathrm{im}}---\left(3\right)$

Truck scale only needs the complex neural network training when installing with annual test for the first time, need not when weighing at ordinary times to train.

(3) truck scale on-line weighing and error compensation

After the sub neural network structure was accomplished with off-line training, truck scale can carry out on-line weighing and error compensation, and its step is following:

(a) slave computer obtains N road LOAD CELLS output signal at first through the separate weighing-up wave Acquisition Circuit in N road, after the data pre-service shown in the formula (3), obtains N road weighing-up wave vector X, for on-line weighing is prepared.

(b), calculate the output y of each sub neural network with the input of weighing-up wave vector X as sub neural network 1,2,3 _i(i=1,2,3).

(c) utilize weighing-up wave vector X, the weight of the tested load of guestimate, and, obtain complex neural network output weights W through weights regulator self-adaptation according to the regulation of national standard " JJG539-97 numeral self-indicating scale vertification regulation ".If w ₁, w ₂, w ₃Component for W then has

${\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">w</figure-callout>}}_1=\left\{\begin{array}{ccc}1& \mathrm{if}& {\tilde{y}}_i\&le;500e\\ 0& \mathrm{if}& 500e<{\tilde{y}}_i\&le;2000e\\ 0& \mathrm{if}& 2000e<{\tilde{y}}_i\&le;M_{\max }\end{array}\right.---\left(4\right)$

${\mathrm{<figure-callout\; id="2"\; label="w"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">w</figure-callout>}}_2=\left\{\begin{array}{ccc}0& \mathrm{if}& {\tilde{y}}_i\&le;500e\\ 1& \mathrm{if}& 500e<{\tilde{y}}_i\&le;2000e\\ 0& \mathrm{if}& 2000e<{\tilde{y}}_i\&le;M_{\max }\end{array}\right.---\left(5\right)$

${\mathrm{<figure-callout\; id="3"\; label="w"\; filenames="HSA00000601474100031.png,HSA00000601474100041.png"\; state="\{\{state\}\}">w</figure-callout>}}_3=\left\{\begin{array}{ccc}0& \mathrm{if}& {\tilde{y}}_i\&le;500e\\ 0& \mathrm{if}& 500e<{\tilde{y}}_i\&le;2000e\\ 1& \mathrm{if}& 2000e<{\tilde{y}}_i\&le;M_{\max }\end{array}\right.---\left(6\right)$

$\left\{\begin{array}{c}W=\left(\begin{array}{c}{w}_1\\ w_2\\ w_3\end{array}\right)\\ \underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{w}_j=1\end{array}\right.---\left(7\right)$

The guestimate of tested load is accomplished through the weighing results estimator, and it can adopt structure identical with sub neural network and training algorithm, also can adopt other structure and algorithm, but because the estimated result error is bigger, can not be as final weighing results value.

(d) the truck scale weighing results obtains.Output y according to each sub neural network _iWith complex neural network output weights, the output weighting fusion with sub neural network obtains final weighing results, and accomplishes the compensation of truck scale weighting error simultaneously, promptly

$y=w_1{y}_i+w_2{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">y</figure-callout>}}_2+w_3{\mathrm{<figure-callout\; id="3"\; label="y"\; filenames="HSA00000601474100031.png,HSA00000601474100041.png"\; state="\{\{state\}\}">y</figure-callout>}}_3=\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{w}_j{y}_j---\left(8\right)$

LOAD CELLS output signal in N of the present invention road adopts the separate signal acquisition circuit in N road, and utilizes the data preprocessing method shown in the formula (3) to obtain, and its principle comprises weighing-up wave conditioning, analog to digital conversion (A/D conversion) etc. shown in accompanying drawing 2.

Slave computer of the present invention is high-performance single-chip microcomputer, dsp processor or other embedded system device, constitutes master control borad with other circuit, and is installed in the truck scale body below, is convenient to the truck scale on-line measurement.

Host computer of the present invention is a PC, is used for off-line training truck scale error compensation model, the line parameter optimization of going forward side by side.

Master control borad of the present invention must be carried out protection against the tide, anti-electromagnetic interference (EMI), anti-thunderbolt processing, in order to avoid influence the accuracy of tested load weighing results.

Compared with present technology the present invention has following advantage: the present invention can realize the automatic compensation of truck scale uneven loading error and linearity error, the accuracy that has improved weighing results greatly.

Description of drawings

Fig. 1 is a truck scale complex neural network training FB(flow block) of the present invention.

Fig. 2 is a truck scale on-line weighing FB(flow block) of the present invention.

Fig. 3 is a complex neural network structural drawing of the present invention.

Fig. 4 is a truck scale master control board theory diagram of the present invention.

Fig. 5 is weighting error compensation effect figure of the present invention, and wherein (a) is the forward and backward weighing results comparison diagram of compensation, (b) is the forward and backward weighing results graph of errors comparison diagram of compensation.

Wherein, 1, first via LOAD CELLS, 2, first via modulate circuit, 3, first via analog to digital conversion circuit; 4, the second road LOAD CELLS, 5, the second road modulate circuit, 6, the second road analog to digital conversion circuit; 7, N road LOAD CELLS, 8, N road modulate circuit, 9, N road analog to digital conversion circuit; 10, slave computer, 11, display circuit, 12, keyboard circuit, 13 host computers (PC).

Embodiment

The present invention proposes a kind of truck scale weighting error automatic compensating method.Be described further below in conjunction with accompanying drawing 1～4, but not as qualification of the present invention.

Embodiment one

Set in the present embodiment, truck scale has 8 road LOAD CELLSs (N=8), and range is 40 tons, and the max cap. of every road LOAD CELLS is 20 tons, and the number of divisions is 4000, and calibrating scale division value e and actual graduation value d are 10kg.The high-performance single-chip microcomputer MSP430F449 that adopts TI company is as slave computer 10.

(1) complex neural network structure.In the present embodiment, sub neural network adopts radial basis function neural network (RBFNN), and the basis function of RBFNN adopts Gaussian function.According to method shown in the formula (1), construct 3 sub neural networks, be example with sub neural network 1, the sub neural network of this moment is the network of one 8 input 1 output, so it is output as

${\mathrm{<figure-callout\; id="1"\; label="y"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">y</figure-callout>}}_1=b_1+\underset{j=1}{\overset{m1}{\mathrm{\&Sigma;}}}{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">h</figure-callout>}}_{1,\mathrm{<figure-callout\; id="1"\; label="j\; w"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">j</figure-callout>}}{w}_{}{}_{1,j}=\underset{j=0}{\overset{m1}{\mathrm{\&Sigma;}}}{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">h</figure-callout>}}_{1,\mathrm{<figure-callout\; id="1"\; label="j\; w"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">j</figure-callout>}}{w}_{}{}_{1,j}={W_1}^T{H}_1---\left(9\right)$

In the formula, the number m of hidden neuron ₁Confirm as 9 by experiment; W ₁Be the weight vector of sub-RBFNN, W ₁=(w _1,0, w _1,1, w _1,2..., w _{1, m}) ^T, w wherein _{N, 0}=1; H ₁Be RBF vector, H ₁=(h _1,0, h _1,1, h _1,2..., h _{1, m}) ¹, h wherein _1,0=b ₁b ₁Be the output layer bias,

${\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">h</figure-callout>}}_{1,j}=\exp (-\frac{{\left|\right|X-{\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">C</figure-callout>}}_{1,j}\left|\right|}^2}{{2\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">r</figure-callout>}}_{1,j}^2})j=\mathrm{1,2},...m_1---\left(10\right)$

In the formula, X is an input vector, X=(x ₁, x ₂, x ₈) ^Tr _{1, j}It is the expansion constant of j node; C _{1, j}Be the center vector of j node, C _{1, j}=(c _{11, j}, c _{12, j}..., c _{18, j}) ^T, || || be Euclidean distance.For sub-RBFNN2, sub-RBFNN3 same analysis is arranged.

(2) complex neural network training.Utilize the standard test weight (as 0.5 ton, 1 ton, 3 tons, 6 tons, 12 tons, 24 tons, 36 tons etc.) of different tonnages to be carried in the diverse location on the truck scale loading end; 140 group of 8 road load cell signal gathered, through the data preprocessing method (Q=50 shown in the formula (3) through LOAD CELLS, modulate circuit, analog to digital conversion circuit and slave computer 10 by system; M=5); Obtain 140 groups of training samples and test sample book, wherein 84 groups are used for each sub-RBFNN training, and 56 groups are used for each sub-RBFNN test.Slave computer 10 is sent to host computer 13 through serial communication interface with these samples, and host computer 13 utilizes training software to carry out the sub neural network off-line training, and the hidden neuron number of each sub-RBFNN is 12.After network finishes training, host computer 13 with the parameter of each sub-RBFNN (like expansion constant R _n, center vector C _n, weight matrix W _nWith output layer biasing b _n) download in the slave computer 10, for truck scale online weighing and error compensation are prepared.Truck scale complex neural network training method is as shown in Figure 1.

(3) truck scale online weighing and error compensation.When truck scale online weighing and error compensation, 8 road LOAD CELLSs output signal is gathered through LOAD CELLS, modulate circuit, analog to digital conversion circuit and slave computer 10 by system; Through the data preprocessing method (Q=50 shown in the formula (3); M=5), obtain weighing-up wave vector X, the weight of the tested load of guestimate; And according to the regulation of national standard " JJG539-97 numeral self-indicating scale vertification regulation ", self-adaptation obtains complex neural network output weights W; Simultaneously with weighing-up wave vector X as sub-RBF neural network 1,2,3 input, according to the parameter that is kept at the sub-RBFNN1 in the slave computer 10, sub-RBFNN2, sub-RBFNN3 (like R _n, C _n, W _nAnd b _n), calculate the output y of RBFNN1, sub-RBFNN2, sub-RBFNN3 respectively ₁, y ₂, y ₃Utilize formula (8) to accomplish truck scale on-line weighing and weighting error compensation at last, obtain the weighing results of tested load.The truck scale method of weighing materials online is as shown in Figure 2.Utilize the national standard " JJG539-97 numeral self-indicating scale vertification regulation " of truck scale calibrating, the truck scale that adopts this method is carried out on-site proving, table 1 uneven loading error verification result is for table 2 is the linearity error verification result.

Table 1 uneven loading error verification result

Table 2 linearity error verification result

In the table 1,1# representes that 1 is known as the loading zone that retransmits the sensor position, other identical meaning arranged.Can find out that by table 1,2 adopt the permissible error of the truck scale uneven loading error of this method much smaller than national Specification, less than the permissible error of national Specification, the error compensation effect is obvious equally for linearity error.

Embodiment 2

Set in the present embodiment, truck scale has 6 road LOAD CELLSs (N=6), and range is 40 tons, and the max cap. of every road LOAD CELLS is 20 tons, and the number of divisions is 4000, and calibrating scale division value e and actual graduation value d are 10kg.The DSP TMS320VC5502 that adopts TI company is as slave computer 10.

(1) complex neural network structure.In the present embodiment, sub neural network adopts BP neural network (BPNN), the latent layer excitation function f of BPNN ₁Adopt sigmoid function, output layer excitation function f ₂Adopting linear function, is example with sub neural network 1, and the sub neural network of this moment is the network of one 6 input 1 output, so it is output as

${\mathrm{<figure-callout\; id="1"\; label="y"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">y</figure-callout>}}_1=W^2{F}_1+b^2=\frac{W^2}{1+e^{-(W^{1}X+b^1)}}+b^2$

$=\underset{j=1}{\overset{m1}{\mathrm{\&Sigma;}}}(\frac{w_{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="HSA00000601474100041.png"\; state="\{\{state\}\}">j</figure-callout>}}^2}{1+e^{-(\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}^1{x}_i+b_i^1)}}+b^2)---\left(11\right)$

In the formula, the number m of hidden neuron ₁Confirm as 5 by experiment; W ¹, W ²Be respectively the BPNN input layer to latent layer, latent layer of weight matrix to output layer; b ¹, b ²Be respectively the bias of the bias vector sum output layer of latent layer.F ¹Be latent layer functions vector (being sigmoid function output).For sub-BPNN2, sub-BPNN3 same analysis is arranged.

(2) complex neural network training.Utilize the standard test weight (as 0.5 ton, 1 ton, 3 tons, 6 tons, 12 tons, 24 tons, 36 tons etc.) of different tonnages to be carried in the diverse location on the truck scale loading end; 112 group of 6 road load cell signal gathered, through the data preprocessing method (Q=50 shown in the formula (3) through LOAD CELLS, modulate circuit, analog to digital conversion circuit and slave computer 10 by system; M=5); Obtain 112 groups of training samples and test sample book, wherein 70 groups are used for each sub-BPNN training, and 42 groups are used for each sub-BPNN test.Slave computer 10 is sent to host computer 13 through serial communication interface with these samples, and host computer 13 utilizes training software to carry out the sub neural network off-line training, and the hidden neuron number of each sub-BPNN is 5.After network finishes training, host computer 13 with the parameter of each sub-BP (like W ¹, W ², b ¹, b ²) download in the slave computer 10, for truck scale online weighing and error compensation are prepared.Truck scale complex neural network training method is as shown in Figure 1.

(3) truck scale online weighing and error compensation.When truck scale online weighing and error compensation, 6 road LOAD CELLSs output signal is gathered through LOAD CELLS, modulate circuit, analog to digital conversion circuit and slave computer 10 by system; Through the data preprocessing method (Q=50 shown in the formula (3); M=5), obtain weighing-up wave vector X, the weight of the tested load of guestimate; And according to the regulation of national standard " JJG539-97 numeral self-indicating scale vertification regulation ", self-adaptation obtains complex neural network output weights W; Simultaneously with weighing-up wave vector X as sub-BP neural network 1,2,3 input, according to the parameter that is kept at the sub-BPNN1 in the slave computer 10, sub-BPNN2, sub-BPNN3 (like W ¹, W ², b ¹, b ²), calculate the output y of BPNN1, sub-BPNN2, sub-BPNN3 respectively ₁, y ₂, y ₃Utilize formula (8) to accomplish truck scale online weighing and error compensation at last, obtain the weighing results of tested load.The truck scale method of weighing materials online is as shown in Figure 2.Utilize the national standard " JJG539-97 numeral self-indicating scale vertification regulation " of truck scale calibrating, the truck scale that adopts this method is carried out on-site proving, table 3 uneven loading error verification result is for table 4 is the linearity error verification result.

Table 3 uneven loading error verification result

Table 4 linearity error verification result

In the table 3,1# representes that 1 is known as the loading zone that retransmits the sensor position, other identical meaning arranged.Can find out that by table 3,4 adopt the permissible error of the truck scale uneven loading error of this method much smaller than national Specification, less than the permissible error of national Specification, the error compensation effect is obvious equally for linearity error.

The content of not doing in the instructions of the present invention to describe in detail belongs to this area professional and technical personnel's known prior art.

Claims (8)

1. truck scale weighting error automatic compensating method; It is characterized in that: utilize complex neural network to approach the nonlinear function between the truck scale input-output; Thereby accomplishing the truck scale pin-point accuracy weighs and error compensation; This method comprises complex neural network structure, neural network off-line training and three aspects of truck scale on-line weighing, and concrete steps are following:

(1) according to priori, promptly the error character of national standard " JJG539-97 numeral self-indicating scale vertification regulation " regulation is constructed 3 sub neural networks;

(2) utilize the standard test weight of different tonnages; Be carried in the diverse location on the truck scale loading end respectively, system's slave computer obtains N road LOAD CELLS output signal through the separate weighing-up wave Acquisition Circuit in N road; After the data pre-service, obtain weighing-up wave vector X _j(j=1,2 ..., K), and composing training sample (X _j, y _j); Slave computer transfers to host computer through communication interface with training sample, utilizes training software, accomplishes 3 sub neural network off-line trainings, obtains relevant parameters, and is downloaded to slave computer, for the truck scale on-line measurement is prepared;

(3) when truck scale online weighing and error compensation, slave computer obtains N road LOAD CELLS output signal at first through the separate weighing-up wave Acquisition Circuit in N road, after the data pre-service, obtains N road weighing-up wave vector X; With the input of weighing-up wave vector X, calculate the output y that obtains each sub neural network as sub neural network 1,2,3 _i(i=1,2,3); The weight of the tested load of guestimate, and according to the regulation of national standard " JJG539-97 numeral self-indicating scale vertification regulation ", self-adaptation obtains complex neural network output weights W, promptly

$w_1=\left\{\begin{array}{ccc}1& \mathrm{if}& {\tilde{y}}_i\&le;500e\\ 0& \mathrm{if}& 500e<{\tilde{y}}_i\&le;2000e\\ 0& \mathrm{if}& 2000e<{\tilde{y}}_i\&le;M_{\max }\end{array}\right.$

$w_2=\left\{\begin{array}{ccc}0& \mathrm{if}& {\tilde{y}}_i\&le;500e\\ 1& \mathrm{if}& 500e<{\tilde{y}}_i\&le;2000e\\ 0& \mathrm{if}& 2000e<{\tilde{y}}_i\&le;M_{\max }\end{array}\right.$

$w_3=\left\{\begin{array}{ccc}0& \mathrm{if}& {\tilde{y}}_i\&le;500e\\ 0& \mathrm{if}& 500e<{\tilde{y}}_i\&le;2000e\\ 1& \mathrm{if}& 2000e<{\tilde{y}}_i\&le;M_{\max }\end{array}\right.$

$\left\{\begin{array}{c}W=\left(\begin{array}{c}{w}_1\\ w_2\\ w_3\end{array}\right)\\ \underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{w}_j=1\end{array}\right.$

In the formula, w ₁, w ₂, w ₃3 components for W; Output y according to each sub neural network _iWith complex neural network output weights, the output weighting fusion with sub neural network obtains final weighing results, and accomplishes the compensation of truck scale weighting error simultaneously, promptly

$y=w_1{y}_i+w_2{y}_2+w_3{y}_3=\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{w}_j{y}_j.$

2. truck scale error character according to claim 1; Be by national standard " the JJG539-97 numeral self-indicating scale vertification regulation " defined of truck scale calibrating; It is divided into 3 different calibrating sections with the total range of truck scale, and the permissible error of each calibrating section is different.

3. complex neural network model according to claim 1 is made up of sub neural network maker, sub neural network, estimator, weights regulator etc.

4. can be radial basis function neural network (RBFNN), BP neural network (BPNN), perhaps other feedforward neural network according to claim 1 and the described sub neural network of claim 3.

5. identical with the structure of described 3 sub neural networks of claim 4 according to claim 1, claim 3, all be the three-layer network of N input 1 output, wherein N is the number of truck scale LOAD CELLS.

6. the data preprocessing method of weighing according to claim 1 is a kind of digital filtering method, Q LOAD CELLS output voltage of this method continuous sampling, and it is obtained S by series arrangement from small to large _Im, remove forward and backward each M sampled value, to all the other Q-2M S _ImCarry out mean filter, promptly

$x_i=\frac{1}{Q-2M}\underset{m=M+1}{\overset{Q-M}{\mathrm{\&Sigma;}}}{S}_{\mathrm{im}}$

7. the weighing-up wave Acquisition Circuit that N according to claim 1 road is separate; Comprise weighing-up wave modulate circuit, analog to digital conversion (A/D conversion) circuit etc., wherein modulate circuit comprises weighing-up wave amplifying circuit, weighing-up wave filtering circuit and impedance matching circuit.

8. slave computer according to claim 1 is high-performance single-chip microcomputer, dsp processor or other embedded system devices; Described host computer is a PC.

Images