CN101100073A - Artificial board production line formaldehyde concentration online monitoring system - Google Patents

Abstract

The computerized intelligent on-line monitoring system for the harmful volatile amount of artificial board and its product in timber industry includes front end devices, formaldehyde monitoring network based one CAN bus and one computerized control terminal. One self-learning expert model containing multiple variable uncertainty system is established by means of modern control theory, neuron, etc for the statistical treatment and analysis of acquired monitoring data and the remote diagnosis and calibration of in-situ units.

Description

Artificial board production line formaldehyde concentration online monitoring system

Technical field

The present invention relates to monitoring and the control technology and the method for control wood-based plate environmental index in a kind of production process, be used for monitoring in process of production and controlling each concerned process steps, make the environmental index such as formaldehyde of product reach optimization, satisfy national wood-based plate and discharge the compulsory standard of limiting the quantity of, assurance China people's is healthy and safe.The invention belongs to Wood-based Panel Production manufacturing and detection range, the burst size of methanal and the concentration of each process procedure in the Monitoring and Controlling building board production process in real time, simultaneously according to the result of monitoring and the decision analysis system with self-learning function of invention, hold the relation of each production process volatile matter burst size and product qualified rate, set up the relational model between the qualified and process of producing product of product, and then realize that formaldehyde discharges the optimum control of limiting the quantity of, or each procedure optimum control scheme is proposed, reach from the environmental index of the source control wood-based plate of Wood-based Panel Production.

Background technology

The total output of China's wood-based plate has reached 6,000 ten thousand M3, has become a genuine Wood-based Panel Production big country, and its annual production has been in first of the world Wood-based Panel Production big country.But the formaldehyde that discharges in wood-based plate and the goods thereof is one of the most serious pollutant.The people of long-term contact formaldehyde can cause oral cavity, throat, skin and gastral cancer.Because the economy of formaldehyde and gummed property also do not find even more ideal substitute at present in wood-based plate, diluent.Yet under the general normal fitting case, the formaldehyde release period in the artificial board reaches 3-15, and therefore, limiting its burst size in containing formaldehyde products just becomes the focus that various countries pay close attention to.

At present, the Wood-based Panel Production enterprise of China in producing the wood-based plate process, be technological parameter with each procedure by after the technological requirement adjustment and fixing, produce again, and the problem that can not occur when producing is done to adjust timely.For example the applying glue concentration of baking temperature, resin added, mixing speed, different aspects, speed of service of slab or the like, and do by this sample loading mode always.But, after the national standard of China's artificial board formaldehyde burst size is implemented, need the urgent problem that solves just to occur in July calendar year 2001.This be because, the fixing way of wood-based plate parameters in the past, if parameter needs to adjust and in time do not adjust aborning in process of production, maximum may be exactly that final product degradation is sold.And formaldehyde discharges the national Specification of limiting the quantity of, and the product that formaldehyde exceeds standard must not be sold on market without exception.Can not the dynamic adjustments parameter if so make in building board production process, will cause into hundred or go up scrapping of kilostere wood-based plate to enterprise, cause the loss of great amount of manpower and material resources, financial resources etc. simultaneously.Therefore artificial board enterprise just is badly in need of online formaldehyde monitoring of a cover and control technology, controls the production of wood-based plate, makes it to adjust each technological parameter dynamically, reaches the optimum control of burst size of methanal and production technology.

The present invention is according to the present situation of China's timber industry, the problem that runs at wood-processing industry, studying the drying of Wood-based Panel Production in great detail, applying glue, mat formation, on the basis of operations such as moulding, research and develop out the concentration of formaldehyde monitoring network, on-line monitoring and each operation concentration of formaldehyde of control, the decision analysis system that simultaneously invention is had self-learning function, hold the relation of each production process volatile matter burst size and product qualified rate, and then realize that formaldehyde discharges the optimum control of limiting the quantity of, or each procedure optimum control scheme is proposed, reach from the environmental index of the source control wood-based plate of Wood-based Panel Production.

Achievement of the present invention can be applicable to various Wood-based Panel Production enterprise, control by formaldehyde monitoring net, reach the optimum control of production technology, thereby according to having guaranteed that enterprise can produce qualified wood-based panel product, the tremendous economic loss that the artificial board formaldehyde of avoiding producing exceeds standard and brings to enterprise; Achievement of the present invention also can be applicable to public arenas such as furniture sales field, storehouse simultaneously, to realize the Monitoring and Controlling of its concentration of formaldehyde, ventilates automatically when concentration over-standard etc., thereby guarantees that people have the PE of a health.Still there is not the on-line monitoring technique that similar techniques is carried out concentration of formaldehyde such as artificial board product line, furniture sales field, warehouse at present both at home and abroad.

Summary of the invention

The present invention is considered as occasion with workshop or wood-based plate procedure, utilizes electric biochemical sensor to receive the analog quantity of the concentration of formaldehyde of occasion, and is translated into data volume.Micro computer receives these data, by setting up the self study expert model of system, these data of analyzing and processing, by handling, analyze the data of each sampled point, and set up between the finished product and get in touch, behind the neuroid algorithm that process project team sets up, the regression algorithm, provide per pass operation best parameter coupling or provide optimum control scheme.

The present invention is directed to the wood-based plate and the product enterprise in China timber industry field, product formaldehyde discharges limits the quantity of, and the computer on-line monitoring technique of a kind of intelligent wood-based plate and goods unwanted volatile burst size thereof is provided.It is by fore device (mostly being 110 concentration of formaldehyde sensors most), constitutes based on CAN bus formaldehyde monitoring net, computer control terminal (host computer) three parts.As shown in Figure 1.

Fore device of the present invention is formed by deciding the on-the-spot machine that potentiometric sensor, single-chip microcomputer constitute, and has collection, parameter setting, data demonstration, communication interface, data storage, power management, execution, warning eight big major functions.The acquisition function module adopts decides the current potential electrochemical sensor, satisfies the transient change of polluting the burst accident scene, can change very at pollutant levels to determine the result rapidly fast, can adapt to the complexity of pollution components.In earlier stage change rapidly at dusty gas, the later stage changes characteristics slowly, and adopting disappears trembles filter method and the method that the arithmetic average filter method combines, and gives full play to the strong point of two kinds of filtering methods, thereby improves system accuracy.Sensor circuit designs can not only be carried out field statistics to Monitoring Data and be handled, and has a data-interface that links to each other with background computer, background computer can be according to last forward data, and carry out analysis-by-synthesis in conjunction with the nature and the social each side situation of this area, from lot of data, disclose its Changing Pattern, predict its variation tendency.Communication interface adopts USB to link to each other with background computer, image data is uploaded to background computer carries out data statistic analysis and modeling.

The second portion of invention is the transmission network based on the CAN bus, comprising the selection of transmission means, the selection of transmission channel, the formulation of communications protocol, i.e. transmission medium, signal form, coded system, medium access mode, wrong detection, frame structure, bit rate.Transmission network has wired and wireless two kinds of patterns.The slave computer control that transmission network is made up of the ARM chip comprises signal acquisition module, communication module, self study Analysis of Policy Making and control module in the lower computer system.Utilize methods such as modern control theory, neuron, make up the self study expert model that contains the multivariable uncertain system Monitoring Data of gathering is carried out statistical disposition, analysis, each on-the-spot machine is carried out remote diagnosis and calibration.

Host computer constitutes control terminal by the PC computer, and major function is that the various signals with slave computer show by different forms with the data of result of calculation, following biography user's various command.Make the parameter analysis that the slave computer user sets, each output optimal value of computing system, thereby make the control parameter of each output node reach optimization, make the environmental index of wood-based plate reach ideal value in process of production.

Description of drawings

Fig. 1 is a formaldehyde monitoring network schematic diagram;

Fig. 2 is based on the formaldehyde monitoring net building-block of logic of CAN bus;

Fig. 3 is based on the formaldehyde monitoring net hardware structure diagram of CAN bus;

Fig. 4 is expert's self learning model schematic diagram;

Fig. 5 is that the input data normalization is handled;

Fig. 6 is that the output data normalization is handled;

Fig. 7 is the generalized regression nerve networks structure chart.

The specific embodiment

As shown in Figure 2, based on the distributed formaldehyde monitoring system of CAN bus mainly by center monitoring main frame (host computer), the PC-CAN adapter, intelligent node is formed.

Operation monitoring software on the center monitoring main frame can be monitored in real time to each workshop section in the wood-based plate manufacture process, comprises warning, ventilation and data record.Simultaneously on the center monitoring main frame, utilize methods such as modern control theory, neuron, be built with the self study expert model that contains the multivariable uncertain system, this model can be found out non-linear relation in existing complicated record data, and can predict, reach the optimization of process.For the center monitoring main frame as a node on the CAN bus, the data card that need have total line traffic control function: the PC-CAN bus adapter, on this basis just can be the center monitoring PC as host computer.Simultaneously, utilize the adapter system that makes to be easy to and other production management department's networkings, be convenient to unified management and running.The PC-CAN bus adapter adopts four-way CAN interface card, and four passages connect corresponding separately network intelligence nodes, extends between factory, test carriage respectively, product storehouse and sample exhibition room.Each network intelligence node is given upper PC main frame with the data upload of collection in worksite.Simultaneously, but the state of each network intelligence node of upper PC main frame distributed earth inquiry control.Each intelligent node all other intelligent nodes on network of active at any time sends information.Each node is installed in the scene, scattered distribution, and the field data of being gathered is passed through the CAN bus transfer to the mster-control centre.The quantity of the next intelligent node depends on the scale that control is on-the-spot, and the multipotency of each CAN passage articulates 110 CAN nodes.As increase repeater, can also increase the node number in theory.

Figure 3 shows that in artificial plate of the present invention and the goods unwanted volatile on-line monitoring technique thereof hardware structure diagram based on the on-the-spot machine of CAN bus formaldehyde monitoring net.

As shown in Figure 3, the function of embedded microcontroller is the control peripheral functional modules, is responsible for the initialization of CAN bus control unit, realizes communication tasks such as the reception of data and transmission by control CAN bus control unit; The function of CAN communication interface is to carry out independently interactive communication with host computer; The function of liquid crystal display be show the system time, the concentration of formaldehyde value and the various error message of intelligent node address, system's baud rate, collection; The function of the data acquisition module environment content of formaldehyde analog signal that to be embedded microcontroller sense the electrochemistry formaldehyde sensor is converted to digital quantity after by signal condition, handles, stores and communicates by letter; The function of warning device is to set limit value if the concentration of formaldehyde value that collects surpasses, and starts ring and reports to the police, or start interlocking equipment; The function of power module is that alternating current 220V changes direct current 24V network switching power supply whole system is powered, and the system that guarantees simultaneously can reliably working, and improves antijamming capability; The function of parameter setting is by 8 toggle switch intelligent node address ID and system transmissions baud rate to be set; The function of managing that resets is initialization embedded microcontroller chip and CAN bus control unit, allows system reset by force system is reinitialized, thereby change normal operation over to; The function of data storage is storing system information and historical data; The function of real-time clock is to select year, month, day, hour, min according to system needs, is used for improving data, and as the foundation of data storage location.

Figure 4 shows that expert's self learning model schematic diagram, according to the concentration of formaldehyde of gathering, resin added, ambient temperature and humidity, pressure, electric current, ventilate, data such as warning are carried out the Treatment Analysis of data by linear regression algorithm or neuroid algorithm, technological parameter according to each operation of result output control comprises resin added, ambient temperature and humidity, pressure, electric current, fan, control informations such as warning.

The regression algorithm that adopts is specific as follows:

Given one group of measurement data (xi, yi), i=0,1,2 ..., m}, based on the principle of least square, (x A), approaches or the match given data it best to try to achieve functional relation f between variable x and the y.(x A) is called model of fit to f, is some undetermined parameters.

Its guiding theory makes model of fit and actual observed value in the weighted sum of squares minimum of the residual error of each point for selecting parameter A, promptly asks f* (x) to make $\underset{i=0}{\overset{m}{\mathrm{\Σ}}}\mathrm{\ω}\left(x_i\right){(f*\left(x_i\right)-y_i)}^2=\min \underset{i=0}{\overset{m}{\mathrm{\Σ}}}\mathrm{\ω}\left(x_i\right){(f*\left(x_i\right)-y_i)}^2,$ ω (x _i) 〉=0 is called power, its reflection data (xi, yi) proportion of shared data in experiment.The curve of using this method match is called the least square fitting curve.

The curve form of for example establishing employing is:

y＝a+bLn(x)

According to principle of least square method, the mean square error that should make matched curve is that least mean-square error is defined as: $Q=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(y_i-(a+\mathrm{bLn}\left(x_i\right)))}^2$

Then ask the partial derivative of a Q respectively, and allow them equal zero a, b.

$\left\{\begin{array}{c}\frac{\∂Q}{\∂a}=0\\ \frac{\∂Q}{\∂b}=0\end{array}\right.$

$\left\{\begin{array}{c}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}2(y_i-(a+\mathrm{bLn}\left(x_i\right)))*(-1)=0\\ \underset{i=1}{\overset{n}{\mathrm{\Σ}}}2(y_i-(a+\mathrm{bLn}\left(x_i\right)))*(-\mathrm{Ln}\left(x_i\right))=0\end{array}\right.$

So obtain equation group

$\left\{\begin{array}{c}\mathrm{na}+\left(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Ln}\left(x_i\right)\right)b=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{y}_i\\ \left(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Ln}\left(x_i\right)\right)a+\left(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\left(\mathrm{Ln}\left(x_i\right)\right)}^2\right)b=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Ln}\left(x_i\right)y_i\end{array}\right.$

The subsequent treatment of curve match is the performance of check curve match, mainly comprises:

Make one-variable linear regression by emulation and object vector, in the coefficient correlation of coming do linear regression between curve match output and the desired value between calculated curve fitting result and the desired value.

Calculate the average relative error between fitting result and the desired value.

The neuroid algorithm that adopts is as follows:

In the generalized regression nerve networks modeling process, since bigger for the mutual difference in input sample data codomain interval, so before entering the neuron operational network, input, output sample at first will enter standardization.By standardization, input vector and target output vector can be quantified as zero-mean and deviation and be 1 standard vector.

The input data normalization is handled, and sees Fig. 5.

When carrying out data output, the thresholding in the time of need being reduced into input to the data of standardization, the output data normalization is handled, and sees Fig. 6.

The Modeling Calculation process

As seen it is a two-layer neutral net by Fig. 7 generalized regression nerve networks structure chart.Ground floor is radially basic neuron layer, and it is input as n1, is output as a1.The second layer is the network linear layer, and it is input as n2, is output as a2.R is the element number of input vector, and S is the element number of output vector, and Q is that the neuron number of ground floor, the second layer and input/target are to number.Ai1 is the output of i Gaussian function, IW1, the 1st, the weights of basic unit radially, LW2, the 1st, the weights of output layer.

Input/object vector number (being sample number) among the neuron number of generalized regression neuroid ground floor and the P (input sample) as many.The weight matrix IW1 of ground floor, 1 is set to P ' (transposed matrix of input sample).Each neuronic network input nI is input vector P and weight vector IW1, the product of Euclidean distance between 1 and threshold value b1.

Calculate the Euclidean distance between input vector and the weight vector:

$\mathrm{dist}({P^{\′}}_{Q\×R},P_{R\×Q})$

$=\mathrm{dist}(\left[\begin{array}{cccc}{P}_{11}& P_{11}& \mathrm{\Λ}& P_{1R}\\ P_{21}& P_{21}& \mathrm{\Λ}& P_{2R}\\ M& M& \mathrm{\Λ}& M\\ P_{Q1}& P_{Q1}& \mathrm{\Λ}& P_{\mathrm{QR}}\end{array}\right],\left[\begin{array}{cccc}{P}_{11}& P_{11}& \mathrm{\Λ}& P_{1Q}\\ P_{21}& P_{21}& \mathrm{\Λ}& P_{2Q}\\ M& M& \mathrm{\Λ}& M\\ P_{R1}& P_{R1}& \mathrm{\Λ}& P_{\mathrm{RQ}}\end{array}\right])$

$={\left[\begin{array}{ccccc}0& d_{12}& d_{13}& \mathrm{\Λ}& d_{1Q}\\ d_{21}& 0& d_{23}& \mathrm{\Λ}& d_{2Q}\\ d_{31}& d_{32}& 0& \mathrm{\Λ}& d_{3Q}\\ M& M& M& & M\\ d_{Q1}& d_{Q2}& d_{Q3}& \mathrm{\Λ}& 0\end{array}\right]}_{Q\×Q}$

The distance of euclidean between i row vector of dij representing matrix P ' and j column vector of matrix P in the following formula.So the matrix of computing gained is a main diagonal element is 0 symmetrical matrix.

After the Euclidean distance that obtains input vector and weight vector,, just can calculate neuronic network input n1 if can determine the threshold value b1 of ground floor again.

In the generalized regression nerve networks model, threshold value b1 is set at 0.8326/spread.Spread is meant radially basic neuronic width, and this width also is " smoothing factor " or " bandwidth ".The user selects spread when the design generalized regression nerve networks, the spread default value is 1, in the generalized regression nerve networks design, along with the increase of spread value, radially basic neuronic response range can enlarge, and the smoothness between each neuron function is also better, the spread value obtains less, make that then function shape is narrower, make with the nearer input of weights Euclidean distance just might be near 1 output, and insensitive to the response of other input.

Just obtain the network input n1 of ground floor this moment, computing formula is:

$n^1=\mathrm{dist}(P^{\′},P)\·*b_1$

$={\left[\begin{array}{ccccc}0& b_{12}*d_{12}& b_{13}*d_{13}& \mathrm{\Λ}& b_{1Q}*d_{1Q}\\ b_{21}*d_{21}& 0& b_{23}*d_{23}& \mathrm{\Λ}& b_{2Q}*d_{2Q}\\ b_{31}*d_{31}& b_{32}*d_{32}& 0& \mathrm{\Λ}& b_{3Q}*d_{3Q}\\ M& M& M& M& M\\ b_{Q1}*d_{Q1}& b_{Q2}*d_{Q2}& b_{Q3}*d_{Q3}& \mathrm{\Λ}& 0\end{array}\right]}_{Q\×Q}$

The neuronic output of ground floor a1 is the RBF of network input n1, that is:

a ¹＝radbas(dist(P′，P).*b ¹)

In the generalized regression nerve networks model, use the form of gaussian kernel function (Gaussian kernelfunction) as basic function, if establish:

n＝dist(P′，P).*b ¹

Then:

$a^1=\mathrm{radbas}(\mathrm{dist}(P^{\′},P).*b^1)=\mathrm{radbas}\left(n\right)=e^{{-n}^2}$

Through RBF, neuron node is output as a1, begins to enter the network linear layer.The network linear layer has the neuron that equates with input object vector number equally.At first will be in the network linear layer through a normalization process device, and then enter common linear neuron.

The normalization process device be input as a1, its output n2 is the cum rights output of input vector.The weight vector LW2 of network linear layer, 1 is set to T (output sample matrix).Computing formula is as follows:

$n^2=\mathrm{normprod}(T,a^1)$

$=\mathrm{normprod}(\left[\begin{array}{cccc}{t}_{11}& t_{12}& \mathrm{\Λ}& t_{1Q}\\ t_{21}& t_{22}& K& P_{2Q}\\ M& M& K& M\\ t_{S1}& t_{S2}& K& t_{\mathrm{SQ}}\end{array}\right],\left[\begin{array}{cccc}{a}_{11}& a_{12}& \mathrm{\Λ}& a_{1Q}\\ a_{21}& a_{22}& \mathrm{\Λ}& a_{2Q}\\ M& M& & M\\ a_{Q1}& A_{Q2}& \mathrm{\Λ}& a_{\mathrm{QQ}}\end{array}\right])$

$={\left[\begin{array}{cccc}\frac{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{t}_{1j}*a_{j1}}{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{a}_{j1}}& \frac{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{t}_{1j}*a_{j2}}{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{a}_{j2}}& \mathrm{\Λ}& \frac{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{t}_{1j}*a_{\mathrm{jQ}}}{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{a}_{\mathrm{jQ}}}\\ M& M& \mathrm{\Λ}& M\\ \frac{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{\mathrm{ts}}_j*a_{j1}}{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{a}_{j1}}& \frac{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{\mathrm{ts}}_j*a_{j2}}{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{a}_{j2}}& \mathrm{\Λ}& \frac{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{\mathrm{ts}}_j*a_{\mathrm{jQ}}}{\underset{j=1}{\overset{j=Q}{\mathrm{\Σ}}}{a}_{\mathrm{jQ}}}\end{array}\right]}_{S\×Q}$

Linear neuron be input as n2, its output a2 is the linearity output of input vector.Computing formula is:

a ²＝purelin(n ²)＝n ²

According to above-mentioned Modeling Calculation process, with the concentration of formaldehyde of gathering, resin added, ambient temperature and humidity, pressure, electric current ventilates, and data such as warning are carried out the generalized regression nerve networks modeling.Simultaneously the capability of fitting of network is tested, mainly comprises:

By artificial network, and the output of reduction network, then output of the network after the reduction and object vector are returned, come the performance of supervising network training.In the coefficient correlation of coming do linear regression between network response and the desired value between computing network response and the desired value.

Calculate the average relative error between fitting result and the desired value.

Whether matched curve can embody virgin curve ground variation tendency more exactly.

The effect example:

This technology is by the concentration of formaldehyde monitoring to building board production process, provide the relation of production process volatile matter burst size and product qualified rate, and then realize that formaldehyde discharges the optimum control of limiting the quantity of, or each procedure optimum control scheme proposed, and can dynamically adjust the parameters of equipment, reach from the environmental index of the source control wood-based plate of Wood-based Panel Production, final products as shown in Figure 7.For the enterprise that uses native system reduces the product loss of hundreds and thousands of cubic meters, bring huge benefit to enterprise.

Claims (8)

1. monitoring system of the unwanted volatile burst size of wood-based plate and goods thereof being carried out on-line monitoring, it comprises: be used for image data and data are carried out the fore device that field statistics is handled, be used for transmission network that data are analyzed and transmitted, and be used to represent various signals, the result of calculation of slave computer and pass the computer control terminal of user command down based on the CAN bus.

2. monitoring system as claimed in claim 1, wherein, described fore device is by deciding potentiometric sensor and single-chip microcomputer is formed; Described transmission network based on the CAN bus has line or wireless mode, and the slave computer control by the ARM chip is formed is used to analyze data, sets up self study expert model and transfer of data.

3. monitoring system according to claim 2, wherein unwanted volatile is a formaldehyde.

4. monitoring system according to claim 2, wherein the data of fore device collection comprise: concentration of formaldehyde, resin added, applying glue be degree, density, intensity, hot pressing temperature, hardening time, ambient temperature and humidity, pressure and/or electric current all.

5. monitoring system according to claim 2, wherein adopting disappears trembles method that filter method combines with the arithmetic average filter method to being handled by the signal of deciding the potentiometric sensor acquisition.

6. monitoring system according to claim 1 and 2 wherein adopts linear regression algorithm or neuroid algorithm that data are handled.

7. monitoring system according to claim 6, in the linear regression algorithm therein, given one group of measurement data ((xi, yi), i=0,1,2 ..., m}, based on the principle of least square, (x A), approaches it or the match given data best to try to achieve functional relation f between variable x and the y.

8. monitoring system according to claim 6, in the neuroid algorithm therein, carry out standardization according to acquired data after, use gaussian kernel function as basic function, the data that obtained are carried out the modeling of generalized regression neuroid.

Images