JP2994088B2 - Fire detector - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fire detecting device, and more particularly to a fire detecting device for detecting a fire from environmental odor.

[0002]

2. Description of the Related Art Recently, various ventilation systems have been adopted in offices and factories in fire detection, and it has become difficult to detect a fire using a conventional smoke sensor under ventilation flow.
In recent years, fires in computers and electronic equipment have been reported, and abnormalities are discovered and suppressed before smoke spreads indoors, that is, when electronic circuit boards and the like in the equipment are smoked. The desire to do so is growing. That is, if there is an effective way to quickly find environmental anomalies before a fire, the likelihood of a fire occurring is greatly reduced. When the material is gradually heated, an odor is generated prior to the generation of smoke. Therefore, if this burning odor can be selectively detected, it is possible to detect a fire at a very early stage.

[0003] Researches for identifying the type of odor using an odor sensor have been actively conducted.
At present, it is impossible with simple equipment, and detection is performed as a molecular weight using expensive and complicated equipment such as gas chromatography and mass spectrometer. In addition, a gas sensor using stannic oxide SnO ₂ that is commonly used has little selectivity to odor, and fire detection is not performed by identifying a burning smell using this sensor.

[0004]

DISCLOSURE OF THE INVENTION The present invention provides a method for accurately discriminating between odors in the environment such as odors of organic solvents which are easily used and easy to respond to odor sensors and burnt odors.
It is an object of the present invention to make fire detection due to burnt odor with high accuracy without misinformation.

[0005]

Therefore, according to the present invention, various types of odor sensors are used by using a plurality of types of thin film odor sensors and means which is good at pattern identification such as back propagation of a neural network. Is detected with high accuracy for fire detection.

More specifically, according to the present invention, a plurality of different types of odor sensors and sensor signals from the plurality of odor sensors are inputted to determine the accuracy of the burning odor in the environmental odor. Means for outputting a signal representing the accuracy of the burnt odor, for example, a neural network, and fire determining means for performing a fire determination based on the output signal representing the accuracy of the burnt odor. A fire detection device is provided.

Further, according to the present invention, a plurality of different types of odor sensors and sensor signals from the plurality of odor sensors are inputted, and the accuracy of the burning odor in the odor in the environment and the environment other than the burning odor are improved. Determine the accuracy of the smell inside,
Means for outputting a signal representing the accuracy of the burning smell and a signal representing the accuracy of the odor in the environment other than the burning smell, for example, a neural network; a signal representing the accuracy of the outputted burning smell; A fire detection device for making a fire judgment based on a signal representing the accuracy of odor in an environment other than the above.

[0008] The accuracy of the odor in the environment other than the burning odor can be the accuracy of the odor of the organic solvent.
Further, the plurality of odor sensors may be composed of cells having metal oxide films of the same material having different film thicknesses, and the metal oxide of the same material may be composed of stannic oxide (SnO 2). ₂ ) can be.

[0009]

[Action] Based on signals from multiple types of odor sensors,
Accuracy of burnt smell is accurately obtained by using a means that is good at pattern identification such as a neural network, and the accuracy of burnt smell and / or
Alternatively, since fire detection is performed based on the accuracy of odor in other environments, fire detection can be performed quickly and accurately.

In the case where a plurality of types of odor sensors are composed of cells having metal oxide films of the same material having different film thicknesses and the metal oxide of the same material is stannic oxide, As a result of making two kinds of such odor sensors and conducting experiments, based on the sensor signals from these two kinds of odor sensors, the burning odor during smoking and the odor of organic solvents such as alcohol are determined. It turned out to be clearly identifiable.

Therefore, a system for discriminating a burning smell generated from the very beginning of a fire and an odor existing in an ordinary environment by using two or more thin film sensors using a discriminating means such as a neural network is created. By doing so, it is possible to reliably detect an extremely early smoldering fire.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to an example in which odor is determined based on sensor signals from two sensor units having films of the same material having different film thicknesses.

In this embodiment, the odor sensor shown in FIG. 1 is used. In the odor sensor shown in FIG. 1, Pt (platinum) has electrodes 12 and 14 on an alumina substrate 10.
Pt is also sputtered on the surface on the opposite side of the alumina substrate 10 for use as the heater 18. Lead wires 24, 26 and 28 are coupled to the Pt electrodes 12, 14 and 16 of the sensor, respectively.

Further, two cells having different thicknesses of SnO ₂ (tin oxide) film are formed on the alumina substrate 10 by sputtering.
O ₂ has a thickness of 2500 ° and the second cell 22
It has a thickness of 1000 ° of nO ₂ . First cell 20
Is used as a thick film first sensor unit S1, and the second cell 22 is used as a thin film second sensor unit S2. FIG. 2 shows an electric circuit configuration of the odor sensor.

Next, the operation of the present invention using the odor sensor having such a configuration will be described.

A voltage is applied to the heater 18 of the odor sensor, and the surface on the side on which SnO ₂ is
FIG. 3 shows the response of the thick film sensor unit S1 and the thin film sensor unit S2 to various odors at the temperature of 00 ° C., that is, the reading of the voltage values at the outputs OUT1 and OUT2. In FIG. 3, the vertical axis represents the sensor unit output OUT1 having a large film thickness, and the horizontal axis represents the sensor unit output OUT2 having a small film thickness.

In FIG. 3, reference numerals 1 to 10 indicate the locus of sensor output for odors generated when these materials are smoked, and reference numerals 11 to 25 indicate sensor outputs for odors generated in a natural state without being smoked. It is a locus of. Especially number 13
Materials 25 to 25 are for odors generated by organic solvents. As can be seen from FIG. 3, the sensor output when the materials Nos. 1 to 10 are smoked has a higher sensitivity with the output from the thick film sensor unit S1 than with the output from the thin film sensor unit S2. It stands almost along the vertical axis in FIG.

On the other hand, the odor generated by the organic solvents Nos. 13 to 25 is more sensitive to the thin film sensor S2, and the output of the thick film sensor S1 rises slightly later with the lapse of time. The appearance of coming is shown. Thus, the odor emitted from the materials of Nos. 13 to 25 is shown in FIG.
The curve is along the horizontal axis. In addition, the tobacco sucker of No. 11 and the coffee powder of No. 12 show the rising state of the materials 1 to 10 more than that of the organic solvent material.
Is closer to the smoked version of this material, presumably because these materials were once burned or smoked.

As described above, when the sensor sections having different film thicknesses as shown in FIG. 1 are used, the output value patterns of the sensor section differ with a certain regularity depending on the quality or type of odor. By identifying
In the type of odor, in particular, in the embodiment of FIG. 3, it is possible to distinguish between the burnt odor generated when the material is smoked and the odor of the organic solvent.

Therefore, for example, the second thin film sensor section S2
As the output value of the first thick film sensor unit S1 becomes higher than the output value of the first thick film sensor unit S1, the accuracy of the burning smell becomes higher and the accuracy of the odor of the organic solvent becomes lower. The first thick film sensor S1 and the second thick film sensor S2 are configured such that the higher the output value of the second thin film sensor unit S2 is, the lower the accuracy of the burning smell and the accuracy of the odor of the organic solvent are. By creating a definition table that defines the accuracy of the combination of the output values of the thin film sensor S2, and learning and memorizing the contents of the definition table defined in this way by, for example, a neural network, the neural network can First and second in the environment
The output values of the sensors S1 and S2 are read, and it is possible to determine an arbitrary odor existing in the environment from the pattern determination of the read values.

FIG. 4 shows an example of such a definition table. In principle, this definition table indicates that the smaller the angle between the line drawn from the origin to that point and the vertical axis is for a point on an arc line having the same distance from the origin in FIG. The larger the angle between the horizontal axis and the line drawn from the origin to the point, the greater the accuracy of the organic solvent, and the constant the angle between the line drawn from the origin and the vertical axis The point is set so that the greater the distance from the origin to that point, the greater the accuracy of both the burning odor and the odor of the organic solvent. It is made with some experience.

In the definition table of FIG. 4, the column of "input" includes a second sensor section (thin film sensor section) S2 having a thickness of 1000 nm SnO _{2 and} a first sensor section (thin film section) having a film thickness of 2500 nm SnO _2. The analog sensor level from the thick film sensor unit S1 is shown as being converted into a value of 0 to 1 (0 to 1000 mV of the sensor output corresponds to 0 to 1). In the column, the definition value D1 of the accuracy of the burning odor and the definition value D2 of the accuracy of the odor of the organic solvent corresponding to 26 combinations of the sensor levels from the two sensors are converted into values of 0 to 1. Respectively. In addition, the other two columns of the four columns in FIG. 4 show calculated values obtained when the learning is completed and a neural network that has sufficiently memorized the contents of the definition table performs the calculation, that is, the values obtained from the two sensors. When the 26 patterns of the sensor level are input, the calculated value C1 of the accuracy of the burning odor calculated and output by the learned neural network and the calculated value C2 of the accuracy of the odor of the organic solvent are also shown. ing.

A neural network is conceptually shown in FIG. FIG. 5 is for explaining the operation of a neural network performed inside a computer or the like, and such a structure does not actually exist. The so-called neural network of the back propagation method is disclosed in
Although described in detail in Japanese Patent Application Laid-Open No. 105299, the neural network used here has a similar effect.

As shown conceptually in FIG. 5, the neural network in this embodiment has two sensor units S1 and S2.
, Two input layers INi (i = 1, 2) receiving the analog data from S2 and five intermediate layers IMj, respectively.
(J = 1 to 5) and two output layers OTk (k = 1, 2) for outputting the calculated values of the burning odor and the odor of the organic solvent, respectively.
2) that connects each input layer and each intermediate layer.
The zero line and the ten lines connecting each intermediate layer and each output layer have ten weights ω _ij and ten weights ω _ij, respectively, depending on the degree of contribution to the output value at the output layers OT ₁ and OT _2. ν _jk is given.

To let the neural network learn the contents of the definition table is equivalent to the sensor level 26 shown in FIG.
The weights ω _ij and ν _jk are adjusted so that the values output from the output layer when the same combination patterns are input from the input layer of the neural network match the corresponding defined values in FIG. That is. This is actually performed using a computer, and a flowchart showing such operations performed by the computer is shown in FIG. 6 as a net structure creation program.

That is, the first and second sensor inputs of the definition table of FIG. 4 are respectively input to the input layers IN ₁ and I ₁ by the net structure creation program shown in the flowchart of FIG.
N ₂ and output layers OT ₁ and OT based on their inputs.
_The value output from ₂ is the defined value D of the burning odor accuracy in FIG.
1 and the definition value D2 of the accuracy of the odor of the organic solvent, respectively, and the weighting values ω _ij and ν _jk assigned to each connection line are changed so that the error of each comparison result is minimized. The weight value in which the error of the comparison result becomes minimum or zero is stored in a memory (hereinafter, referred to as a weight value memory) and used. In this way, it is possible to instruct the net structure of FIG. 5 a very close approximation to the entire function of the definition table of FIG. When the contents of the definition table are taught in the net structure in this manner, the calculation is performed by using the weights stored in the weight memory for the inputs of the sensors S1 and S2. It is possible to accurately obtain the calculated values of the burning odor and the accuracy of the odor of the organic solvent.

Now, the weight ω _ij between the input layer INi and the intermediate layer IMj and the weight ν _jk between the intermediate layer IMj and the output layer OTk (i = 1, 2, j = 1 to 5) , K = 1, 2)
Have positive, zero, and negative values, respectively. If the input value in the input layer INi is represented by INi, the intermediate layer I
The total sum NET ₁ (j) of inputs to Mj is expressed by the following equation 1.

[0028]

(Equation 1)

This value NET ₁ (j) is converted to, for example, sigmoid (s
igmoid) function to convert the value to 0-1 and convert it to IMj
Is expressed by the following equation (2).

[0030]

(Equation 2)

Similarly, the total sum NET ₂ (k) of inputs to the output layer OTk is expressed by the following equation (3).

[0032]

(Equation 3)

If this value NET ₂ (k) is similarly converted to a value of 0 to 1 by a sigmoid function and is represented by OTk, it is expressed by the following equation (4).

[0034]

(Equation 4)

As described above, the relationship between the input values IN ₁ and IN ₂ and the output values OT ₁ and OT ₂ in the net structure shown in FIG. 5 is expressed by Equations 1 to 4 using weighting values. Here, γ ₁ and γ ₂ are adjustment coefficients of the sigmoid curve, and in this embodiment, γ ₁ = 1.0 and γ = 1.2 are appropriately selected.

In the net structure creation program, first, when one of the 26 combinations of sensor input patterns in the definition table of FIG. 4 is given to the input layer,
The actual outputs OT ₁ and OT ₂ calculated from the above-described equations 1 to 4 and output from the output layer are compared with the burnt smell definition value D1 and the organic solvent smell definition value D2 in FIG. 4, respectively.
Sum of each error in each output layer at that time Em
(m = 1 to 26) is represented by the following equation 5.

[0037]

(Equation 5)

Here, OTk is a value obtained by the above equation (4). The value E obtained by summing the sum of errors Em for all of the 26 combinations in the definition table of FIG.

[0039]

(Equation 6)

Finally, by repeating the net structure creation program shown in FIG. 6 several times, the weighting value stored in the weighting value memory is changed so that the value E in equation 6 becomes zero or minimum. Adjustment and change, finally,
The weighting value memory stores the weighting value at which the value E is zero or minimum. By using this weighting value memory, or by copying the contents of this weighting value memory to another RAM or ROM, it can be used when performing odor determination or fire detection.

FIGS. 7 and 8 show the accuracy output when various combinations of sensor inputs are given using a neural network which has already completed the education of the contents of the definition table of FIG. On the axis is the output OUT of the thin film sensor S2
2 is converted to 0 to 1, the X-axis indicates the value obtained by converting the output OUT1 of the thick film sensor S1 to 0 to 1, and the Z-axis indicates the accuracy of the burning odor of 0 in FIG. In the case of FIG. 8, values of 0 to 1 are shown for the accuracy of odor of the organic solvent. The meshes in FIGS. 7 and 8 represent the magnitude of the accuracy calculated by the neural network for an arbitrary sensor input, and the lines or dotted lines on the mesh indicate the burning smell of a specific substance or the odor of a specific organic solvent. Represent. FIG. 9 shows the maximum accuracy value for the odor for each substance, that is, the maximum value on the Z axis in%. Nos. 1 to 11 correspond to cases where odor was generated by heating and smoking at a rate of 10 ° C./min.

When the education of the definition table of FIG. 4 for the net structure conceptually shown in FIG. 5 is completed, that is, when the adjustment of the weight value of each one is completed, the actual odor discrimination or fire detection is performed. Is provided with an input value from each sensor unit to a net structure by a net structure calculation program described later, and a value obtained from each output layer is calculated by using the above-described equations 1 to 4, thereby obtaining a burnt smell. The calculated value and the calculated value of the accuracy of the odor of the organic solvent are obtained. For example, when performing a fire detection, the calculated value of the burning odor thus obtained may be compared with a certain reference value, and if it is equal to or greater than the reference value, it may be determined that a fire has occurred.

FIG. 5 shows an example of a fire detection apparatus which performs a fire monitoring by making the neural network conceptually shown in FIG. 5 learn the contents of the definition table shown in FIG. 5 and discriminating the odor in the environment based on the learning result. It is shown in FIG. Fire detection apparatus in the embodiment of FIG. 10, the fire receiver RE via the signal and power line L is connected to a plurality of fire detectors DE ₁ ~DE _N, each fire detector is the fire receiver This system returns a status signal such as a fire signal to a fire receiver in response to a polling signal from the RE. The function of performing calculations based on a neural network structure as shown in FIG. The figure shows a configuration in which each fire detector is provided and the result of the determination is transmitted to the fire receiver RE. Of course, it is also possible that each fire sensor sends the sensor data as it is to the fire receiver RE, and the odor or fire discrimination is performed on the fire receiver RE side. Note that the fire detector DE ₁ ~DE _N is the same configuration shows an internal structure in detail only the fire detector DE _1.

In the fire detector DE ₁ shown in FIG.
U1 is a microprocessor. The microprocessor MPU1 has a working memory RAM1, a program storage memory ROM11, various constant storage memory ROM1.
2, a weight value memory ROM 13 storing weight values copied from the weight value memory described in FIG.
A self-address memory ROM 14, a transmitting / receiving unit TRX1 for transmitting / receiving to / from the fire receiver RE, and the like are connected and shown. In addition, the first and second odor sensor units S1 and S1 described with reference to FIGS. S2 is shown connected as a fire sensor FS. A switch DIP for setting the self-address indicated by a dotted line can be connected as needed. Note that IF11 to IF15 are interfaces.

FIG. 1 shows the operation of the fire detection device shown in FIG.
This will be described with reference to the flowchart shown in FIG. The fire monitoring operation for the nth fire sensor DEn will be described. When the data reading time comes (Y in step 706), data is read from the fire sensor FS, that is, the first sensor unit S1 and the second sensor unit S2. The read sensor levels are temporarily stored in the work area RAM 1 and converted into values INi (i = 1, 2) of 0 to 1 (steps 708 to 714), and then the net structure shown in FIG. The calculation program 800 is executed.

In the net structure calculation program 800, NET ₁ (j) is calculated according to the above-described equation 1 (step 803), and is converted into the value of IMj according to the equation 2 (step 804). All of the IM to IM ₁ ~IM ₅
When the value of j is determined (Y in step 805), NET ₂ (k) is calculated using the values of IMj according to the above-described equation 3 (step 808), and is calculated as OTk according to equation 4.
(Step 809). When the value of all the OTk of OT ₁ and OT ₂ are determined (Y of step 810), returns to the flowchart of FIG. 11. The values of OT ₁ and OT ₂ represent the actually measured values of the accuracy of the burning smell and the accuracy of the odor of the organic solvent, respectively.

Therefore, in each step of FIG.
T ₁ is compared with a reference value A (for example, A = 0.4 can be set) of the fire probability read from the various constant table storage area ROM 12, and if OT ₁ ≧ A (Y in step 716) , The fire signal is set (step 718), and OT ₂ is compared with the reference value B of the odor of the organic solvent which is also read from the storage area ROM12,
If OT ₂ ≧ B (Y in step 720), an unpleasant signal is set (step 722). These set signals are transmitted and received the next time a call is made from the fire receiver RE (Y in step 702).
Then, it is sent to the fire receiver via the transmission line L (step 704).

In the embodiment shown in FIG. 11, attention is paid only to the burning smell.
In the above description, a fire is determined to have occurred when the degree of burnt smell is, for example, 0.4 or more. However, the following fire determination method may be used. 1. If the accuracy level exceeds 0.4 within a certain period of time after the accuracy output of the burning odor changes, it is determined that a fire has occurred. In this way, the fire can be determined by determining the burning smell and the odor of the organic solvent. 2. Method of determining fire using each accuracy level of burnt odor accuracy and organic solvent odor accuracy (Part 1): For example, if the burnt odor accuracy level is 0.15 or more and the organic solvent odor accuracy level is When the fire determination level is 0.01 or less, the burnt odor accuracy level exceeds 0.15 within a certain time after the change in the burnt odor accuracy output appears, and the accuracy level of the organic solvent odor is within 0.01. If there is, judge it as a fire. In this manner, the fire determination can be performed more accurately, and, for example, by simply comparing the accuracy of the burnt odor with the reference value as in the embodiment of FIG. In this case, it is determined that a fire has occurred, but such misinformation can be avoided. 3. Method of determining fire using each accuracy level of burnt odor accuracy and organic solvent odor accuracy (Part 2): Calculates burnt odor accuracy and organic solvent odor accuracy obtained within a certain period of time If the total value exceeds a certain level, for example, 0 (in this case, the output of the accuracy of the burning odor is required), or if the total value exceeds, for example, 0.10. In other words, the accuracy of the burning odor and the accuracy of the odor of the organic solvent are separately added, and the added value of the accuracy of the odor of the organic solvent is subtracted from the added value of the accuracy of the burning odor. It is determined whether or not it has been reached. In this case, when subtracting the added value of the accuracy of the smell of the organic solvent from the added value of the accuracy of the burning smell, the added value of the accuracy of the smell of the organic solvent may be multiplied by a coefficient.

In the above embodiment, the odor sensor is composed of cells having two different thicknesses of SnO _2. However, the material of the odor sensor is not limited to SnO _2. Instead, other various materials may be used, and odor sensors made of different types of materials may be mixed and used.

Also, in order to determine the type of odor based on sensor signals from a plurality of types of odor sensors, a neural network using a back propagation method has been described. Other means that can perform pattern identification based on the odor signal described above, such as a table, may be used. Further, the processing may be performed by fuzzy inference.

[0051]

As described above, according to the present invention, based on signals from a plurality of types of thin film odor sensors, various types of odors can be accurately detected by means which is good at pattern identification such as back propagation of a neural network. Since the fire is detected by performing good detection, there is an effect that such fire detection can be performed quickly and accurately. Also,
When a plurality of types of thin film odor sensors are composed of cells having metal oxide films of the same material having different film thicknesses, such an odor sensor can be manufactured easily and compactly. is there.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram showing an odor sensor used in an embodiment of the present invention.

FIG. 2 is an electric circuit diagram of the odor sensor shown in FIG.

FIG. 3 is a graph for explaining characteristics of the odor sensor shown in FIG. 1;

FIG. 4 is a diagram showing a definition table that defines and defines the accuracy of burnt smell and the smell of an organic solvent from the graph showing the characteristics of FIG. 3;

FIG. 5 is a diagram for conceptually explaining a neural network used in the embodiment of the present invention.

FIG. 6 is a flowchart for explaining a net structure creation program for causing the neural network shown in FIG. 5 to store the definition table shown in FIG. 4;

7 is a diagram showing the accuracy of burnt odor output from the neural network when various sensor values are given to the neural network created by the program shown in FIG. 6;

8 is a diagram showing the accuracy of the odor of the organic solvent output from the neural network when various sensor values are given to the neural network created by the program shown in FIG.

9 is a diagram showing the maximum value of the odor accuracy of various substances in FIGS. 8 and 9. FIG.

FIG. 10 is a block circuit diagram showing a fire detection device according to one embodiment of the present invention.

FIG. 11 is a flowchart for explaining the operation of FIG. 10;

FIG. 12 is a flowchart illustrating a net structure calculation program executed during the operation of FIG. 11;

[Explanation of symbols]

S1 first sensor portion S2 second sensor section RE fire receiver DE ₁ ~DE _N plurality of fire detectors MPU1 microprocessor ROM11 program storing memory ROM12 various constants table storage area ROM13 duplex bid of a thin film of thick Memory

Claims (5)

(57) [Claims] 1. A plurality of different types of odor sensors, and a sensor signal from the plurality of odor sensors, a signal representing the accuracy of the burning odor in an environmental odor and representing the accuracy of the burning odor. And a fire discriminating means for judging a fire based on the output signal representing the accuracy of the burning odor. 2. A plurality of different types of odor sensors, and sensor signals from the plurality of odor sensors are input, and the accuracy of the burning odor in the environmental odor and the accuracy of the odor in the environment other than the burning odor are determined. Means for outputting a signal representing the accuracy of the burning odor and a signal representing the accuracy of the odor in the environment other than the burning odor; and a signal representing the accuracy of the output burning odor and the output of the signal representing the accuracy of the burning odor. And a fire determining means for determining a fire based on a signal representing the accuracy of the odor. 3. The fire detection device according to claim 2, wherein the accuracy of the odor in the environment other than the burning odor is the accuracy of the odor of the organic solvent. 4. The fire detecting device according to claim 1, wherein said plurality of odor sensors are cells having metal oxide films of the same material having different film thicknesses. 5. The fire detecting device according to claim 4, wherein the metal oxide of the same material is stannic oxide (SnO ₂ ).