JP7280734B2 - gas leak alarm - Google Patents

Description

The present invention relates to a gas leak alarm that detects and warns of a gas leak of city gas or LP gas based on the output of a gas detection element.

2. Description of the Related Art Conventionally, gas leak alarms have been widely used to detect and warn of leaks of city gas and LP gas (propane gas) used in ordinary homes and the like.

The gas alarm detects CO gas and hydrocarbon gas as monitored gases, and the overheating temperature of the heater of the gas detection element is periodically switched between the detection temperature of CO gas and the detection temperature of hydrocarbon gas. monitor alternately for CO gas or hydrocarbon gas, and light or flash the CO gas alarm lamp when detecting CO gas, and light or flash the hydrocarbon gas alarm lamp when detecting hydrocarbon gas. Furthermore, an alarm sound is output from the speaker in accordance with lighting or blinking of the alarm lamp.

Japanese Patent Application Laid-Open No. 2002-228613 JP-A-2002-230661 WO2016-136434 WO2016-031080 publication

By the way, natural gas, which is the raw material of city gas and LP gas, has no smell, and an odorant is added for safety purposes so that even a very small amount of leakage can be detected quickly. Therefore, even if a slight gas leak occurs in a kitchen or the like, if there is someone present, it is possible to detect the peculiar smell of the gas leak and quickly and appropriately deal with it.

However, conventional gas leak alarms detect hydrocarbon gas, which is the main component of city gas and LP gas, and issue an alarm. The gas concentration of hydrocarbon gas must exceed a predetermined concentration, and the gas leak alarm is installed on the ceiling surface of the kitchen or on a wall surface close to the ceiling surface, so it takes time to detect a gas leak and issue an alarm. It may take

By the way, in recent years, odor sensors have been put into practical use that are capable of detecting and distinguishing a specific odor from a collection of complex odors, like the human sense of smell (Patent Documents 3 and 4). . In this way, if we have a sensor that can detect and discriminate odors that correspond to the human sense of smell, we will be able to detect and distinguish the odors of city gas and LP gas, which have not been seen in the past, such as the smell of odorants added for safety purposes. It can be combined with leakage monitoring, can quickly and reliably determine gas leakage and issue an alarm, and can reliably prevent erroneous alarms due to miscellaneous gases other than the monitored gas.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a gas leak alarm that can determine gas leaks with high accuracy by combining the element of odor with the determination of gas leaks.

(gas leak alarm)
The present invention is a gas leak alarm,
gas detection means for detecting a predetermined monitoring target gas;
an odor detection means for detecting the odor of a predetermined odorant added to the gas to be monitored;
gas leak determination means for determining a gas leak based on the detection output of the gas detection means and the detection output of the odor detection means and outputting a gas leak alarm;
is provided.

(Odor sensor and odor determination part)
The odor detection means is
an odor sensor that outputs multiple types of odor detection values according to the adsorption of multiple odorants;
an odor determination unit that inputs multiple types of odor detection values output from the odor sensor and determines the odor of the odorant;
Prepare.

(Determination of odor type by multilayer neural network)
The odor determination unit is composed of a multi-layered neural network. Input information is a combination of multiple types of odor detection values from the odor sensor, and the expected output value is a combination of odorant odors. It learns and estimates and outputs the odor of the odorant when multiple types of odor detection values from the odor sensor are input as input information.

(Gas leak alarm for both gas detection and odor detection)
The gas leakage determining means determines a gas leakage and outputs a gas leakage alarm when both the detection output of the gas to be monitored by the gas detection means and the detection output of the odor of the odorant by the odor detection means are determined. .

(preliminary alarm by gas detection or odor detection)
The gas leakage determining means determines a warning of gas leakage and detects gas leakage when determining either one of the detection output of the gas to be monitored by the gas detection means and the detection output of the odor of the odorant by the odor detection means. Output a preliminary alarm.

(Gas leak alarm by gas detection or smell detection)
The gas leakage determination means determines gas leakage and outputs a gas leakage alarm when it determines the detection output of the gas to be monitored by the gas detection means or the detection output of the odor of the odorant by the odor detection means.

(Effects of multi-layer neural network configuration and learning)
A multilayer neural network is a coupled neural network,
Learning is performed by back propagation based on the error between the value output when the learning information is input and the expected value of the predetermined value.

(Gas leak alarm using only the odor detection part)
In another aspect of the invention, a gas leak alarm comprising:
odor detecting means for outputting a plurality of types of odor detection values according to the adsorption of the odorants of the monitoring target gas contained in the monitoring target gas and the odorant added to the monitoring target gas;
It is composed of a multi-layer neural network, and each combination of a plurality of types of odor detection values of the odor detection means, which is input information, and the odor of the monitored gas and odorant, which are output expected values, is input as learning information. an odor discrimination means that learns by deep learning and determines whether or not there is an odor of the gas to be monitored and the odorant when a plurality of types of odor detection values of the odor detection means is input as input information, and outputs the result;
a gas leak determination means for determining a gas leak based on the determination result of the smell of the monitored gas and the odorant by the smell determination means and outputting a gas leak alarm;
is provided.

(Determination of gas leak alarm)
The gas leakage determination means is
when both the detection of the gas to be monitored and the detection of the odor of the odorant are determined by the odor determining means, the gas leak is determined and a gas leak alarm is output;
In this case, when either the detection of the gas to be monitored or the detection of the odor of the odorant is determined, it is determined that gas leakage should be cautioned, and a preliminary warning of gas leakage is output.

Further, the gas leakage determination means may determine gas leakage and output a gas leakage alarm when detection of the gas to be monitored by the odor determination means or detection of the odor of the odorant is determined.

(basic effect)
The present invention is a gas leak alarm device comprising gas detection means for detecting a predetermined gas to be monitored, odor detection means for detecting an odor of a predetermined odorant added to the gas to be monitored, and gas detection means. A gas leak determination means for determining a gas leak based on the detection output of and the smell detection means and outputting a gas leak alarm is provided. When the gas is detected as the target gas, the odor of the specified odorant added to the monitoring target gas is also detected to determine the presence or absence of gas leakage. Gas leakage can be determined with accuracy.

(Effect of odor sensor and odor determination part)
Further, the odor detection means includes an odor sensor that outputs a plurality of types of odor detection values according to the adsorption of a plurality of odorants, and an odorant sensor that receives the plurality of types of odor detection values output from the odor sensor and detects the odor of the odorant. Further, the odor determination unit is configured by a multi-layer neural network, and includes multiple types of odor detection values of the odor sensor as input information and the odor of the odorant as an output expected value. are input as learning information and learned by deep learning, and when multiple types of odor detection values from the odor sensor are input as input information, the odor of the odorant is estimated and output. Like the sense of smell, the odor of odorants added to city gas or LP gas is detected from a collection of complex odors, and it is used to judge gas leaks by sharply distinguishing them, making it possible to judge gas leaks with high accuracy.

(Effect of alarming gas leakage with both gas detection and odor detection)
Further, the gas leakage determination means determines a gas leakage and outputs an alarm when both the detection output of the gas to be monitored by the gas detection means and the detection output of the odor of the odorant by the odor detection means are determined. As a result, it is possible to make a more reliable determination of gas leakage.

(Effect of pre-warning by gas detection or odor detection)
Further, the gas leakage determination means determines a gas leakage caution and gas leakage warning when determining either one of the detection output of the gas to be monitored by the gas detection means and the detection output of the odor of the odorant by the odor detection means. Since the preliminary warning of leakage is output, the preliminary warning is issued at the stage before the gas leakage is judged, so that a slight gas leakage can be reported and a quick countermeasure can be taken.

(Effect of gas leak warning by gas detection or smell detection)
Further, the gas leakage determination means outputs a gas leakage alarm when determining the detection output of the gas to be monitored by the gas detection means or the detection output of the odor of the odorant by the odor detection means. By detecting hydrocarbon gas by the gas detection means, the smell of the odorant added to city gas or LP gas can be determined at an early stage, and a gas leak alarm can be output. can detect the odor of the odorant specific to the leaking gas and quickly and reliably warn of the gas leak.

In addition, the gas detection means and the odor detection means can be mutually backed up in the event of a failure, thereby improving the reliability of the gas leak alarm device.

(Effects of multi-layer neural network configuration and learning)
In addition, the multi-layer neural network is a coupled neural network, and since it learns by back propagation based on the error between the value output when learning information is input and a predetermined expected value, it is possible to determine, for example, the type of odor. , when a plurality of odor detection values output from the odor detection means are input as input information, an estimated value of the peculiar odor due to the odorant added to the city gas or LP gas is used as the output expected value. , and the weights and biases in the multi-layer neural network are learned to minimize the error between the output value and the expected value by back propagation processing. It is possible to discriminate the peculiar smell of the odorant and use it for highly accurate gas leak judgment.

(Effect of gas leak alarm using only odor detector)
According to another aspect of the present invention, there is provided a gas leak alarm device comprising a plurality of gas leak alarms according to the adsorption of the odorant of the monitoring target gas contained in the monitoring target gas and the odorant added to the monitoring target gas. An odor detecting means for outputting different types of odor detection values and a multi-layered neural network. The multiple types of odor detection values of the odor detecting means, which are input information, and the monitored gas and odorant, which are output expected values. Each combination with odor is input as learning information and learned by deep learning, and when multiple types of odor detection values of the odor detection means are input as input information, the presence or absence of the odor of the gas to be monitored and the odorant is determined. and a gas leakage judgment means for judging a gas leakage based on the judgment results of the monitored gas and the odor of the odorant by the odor judgment means and outputting a gas leakage alarm. The odor of hydrocarbon gas and odorant contained in gas and LP gas is determined by inputting a plurality of odor detection values output by the odor detection means, which is an odor sensor, into an odor determination unit composed of a multi-layer neural network. , and a combination of the results, it is possible to determine the gas leak with high accuracy and issue an alarm.

(Effect of determination of gas leak alarm)
Further, the gas leakage determination means determines gas leakage and outputs a gas leakage alarm when both the detection of the gas to be monitored by the odor determination means and the detection of the odor of the odorant are determined. When either one of the detection of the target gas and the detection of the odor of the odorant is determined, it is determined that there is a gas leakage caution and a preliminary warning of gas leakage is output, or the gas leakage determination means performs monitoring by the odor determination means. A gas leak alarm equipped with the above-described gas detection means and odor detection means is provided by determining a gas leak and outputting a gas leak alarm when detecting the target gas or detecting the odor of the odorant. The effect is obtained in the same manner as the gas leak determination means of the vessel.

Block diagram showing an embodiment of a gas leak alarm Explanatory diagram showing the odor detection values of 16 channels by the odor sensor in FIG. 1 in a radar chart Explanatory diagram showing the functional configuration of the odor determination unit in FIG. Explanatory diagram showing the functional configuration of the multi-layer neural network shown in FIG. Explanatory diagram showing a list of determination processing by the gas leak determination unit in FIG. Explanatory drawing showing another embodiment of an odor gas leak alarm device that detects a gas leak only by an odor detection unit. Explanatory diagram showing the functional configuration of the odor determination unit in FIG.

[Basic concept of the embodiment]
FIG. 1 is an explanatory diagram showing an embodiment of a gas leak alarm. The basic concept of the gas leak alarm device according to the present embodiment is that the gas detection unit 10, which is gas detection means, detects a predetermined monitored gas such as city gas or LP gas, and the odor detection unit 16, which is odor detection means, monitors the target gas. The odor of a predetermined odorant added to the target gas is detected, and a gas leak determination unit 30, which is a gas leak determination means, determines, based on the detection output of the gas detection unit 10 and the detection output of the odor detection unit 16, for example When both the detection output of the gas to be monitored and the detection output of the smell of the odorant are discriminated, a gas leak is judged and a gas leak alarm is output. When hydrocarbon gas is detected as a gas to be monitored, in addition to detecting the smell of odorants added to city gas and LP gas, the presence or absence of gas leaks can be detected, thereby reducing the influence of miscellaneous gases. Gas leakage can be determined with high accuracy without

The odor detection unit 16 also includes an odor sensor 18 that outputs a plurality of types of odor detection values according to the adsorption of a plurality of odorants, and an odor sensor 18 that inputs the plurality of types of odor detection values output from the odor sensor 18 to add odors. Further, the odor determination unit 20 is configured by a multi-layer neural network, and the odor determination unit 20 is configured by a plurality of types of odor detection values of the odor sensor 18 as input information and an output expected value. A combination of odors of a certain odorant is input as learning information and learned by deep learning, and when multiple types of odor detection values from the odor sensor 18 are input as input information, the odor of the odorant is estimated and output. By doing so, the odor of the odorant added to city gas and LP gas can be detected from a collection of complex odors, similar to the human sense of smell. This makes it possible to determine leakage.

[Overview of gas leak alarm]
As shown in FIG. 1, the gas leak alarm device of this embodiment comprises a gas detector 10, an odor detector 16, a controller 24, a display 26, and an audio alarm . The control unit 24 is composed of a computer circuit having a CPU, a memory, and various input/output ports, and implements the function of the gas leakage determination unit 30 by executing a program. The display unit 26 is provided with a CO gas alarm indicator lamp and a hydrocarbon gas alarm indicator lamp. The voice warning unit 28 is provided with a voice amplifier circuit and a speaker.

[Gas detector]
The gas detection unit 10 is provided with a gas sensor 12 and a gas detection circuit unit 14, and detects hydrocarbon gas ( _CH4 gas), which is the main component of city gas and LP gas, and CO gas generated by incomplete combustion. do. The gas sensor 12 is, for example, of a catalytic combustion type, and when gas comes into contact with a platinum wire through which an electric current is flowing, slight combustion occurs and the electrical resistance of the platinum wire increases, thereby detecting the gas.

The gas detection circuit unit 14 periodically automatically adjusts the temperature of the heater that heats the platinum wire provided in the gas sensor 12 to the temperature T1 for CO gas detection and the temperature T2 (>T1) for hydrocarbon gas detection. , the electric resistance of the platinum wire is detected at each switching timing, and the CO gas detection signal G1 and the hydrocarbon gas detection signal G2 are output to the control unit 24 .

The gas sensor 12 may be of a semiconductor type. A semiconductor gas sensor detects gas by reducing the resistance value of the semiconductor when the gas comes into contact with the surface of the semiconductor.

[Odor detector]
The odor detection unit 16 is composed of an odor sensor 18 and an odor determination unit 20. The odor determination unit 20 is provided with a multilayer neural network. Determine the odor of the added odorants.

(odorant)
The component of the odorant added to city gas or LP gas is, for example, a mixture of tertiary butyl mercaptan TBM and cyclohexene CH. Tertiary butyl _mercaptan TBM has the _{formula C4H10S} and cyclohexene CH has the formula _C6H10 .

Tertiary butyl mercaptan TBM is a kind of organic sulfur compound, has a strong odor, and becomes carbon dioxide, water, and sulfur oxides when burned. In addition, cyclohexene CH does not contain sulfur, and when burned, becomes carbon dioxide and water, and does not generate sulfur oxides.

(Odor sensor)
FIG. 2 is an explanatory diagram showing the odor detection values of 16 channels by the odor sensor 16 of FIG. 1 in the form of a radar chart. The odor sensor 18 provided in the odor detection unit 16 shown in FIG. 1 outputs multiple types of odor detection values x1 to x16 according to the adsorption of multiple odorants.

The odor sensor 18 is provided with a plurality of gas sensors in which a sensitive film for selectively adsorbing different odor molecules is formed on a quartz substrate, for example. The sensitive film of the gas sensor uses a material that selectively adsorbs gas components by reacting with each odor molecule, such as ammonia, toluene, acetone, and ethanol. The gas sensor changes the frequency of, for example, a crystal oscillator in proportion to the amount of adsorbed odor molecules, and outputs a detection value xi having a level corresponding to the amount of change in frequency. In this embodiment, for example, a 16-channel odor sensor 18 that outputs detection values x1 to x16 in response to 16 kinds of odor molecules is used.

When city gas or LP gas leaks, the odor sensor 18 comes into contact with the hydrocarbon gas and the odorant, and outputs a plurality of odor detection values x1 to x16 corresponding to the contact between the two. The odor sensor 18 also outputs a plurality of odor detection values x1 to x16 corresponding to the contact with CO gas, even when it comes into contact with CO gas generated by incomplete combustion or the like.

FIG. 2(A) shows a radar chart of the detection values x1 to x16 when the odorant comes into contact, and FIG. 2(B) shows a radar chart of the detection values x1 to x16 when the hydrocarbon gas comes into contact. FIG. 2C shows a radar chart of detection values x1 to x16 when CO gas comes into contact. Note that the radar chart in FIG. 2 is an example, and various patterns can be obtained depending on the type and arrangement of the 16-channel gas sensors.

As is clear from the radar chart in FIG. 2, the detection values x1 to x16 of the odor sensor 18 with a multi-channel configuration (multi-gas sensor configuration) visually show different patterns for odorant, hydrocarbon gas, and CO gas. This difference in visual patterns means that, for odorants, a specific odor can be detected and distinguished from a collection of complex odors, just like the human sense of smell.

Here, when city gas or LP gas leaks, an odorant is added at the same time. The pattern is a combination of FIG. 2(B).

In order to discriminate the odor of the odorant from the detection values x1 to x16 of the odor sensor 18, the odor detection values x1 to x16 from the odor sensor 18 are sent to an odor determination unit 20 having a multi-layer neural network 22. are typing.

The odor sensor 18 may have higher resolution such as 32 channels or 64 channels other than the 16 channels of this embodiment.

(Odor determination unit)
3 is an explanatory diagram showing the functional configuration of the odor determination unit in FIG. 1, and FIG. 4 is an explanatory diagram showing the functional configuration of the multilayer neural network shown in FIG.

As shown in FIG. 1, the odor determination unit 20 receives the odor detection values x1 to x16 from the odor sensor 18 and determines the odor of the odorant. An odor determination value Y2 of the odorant is output. Here, when the odor of an odorant is determined, (Y1, Y2)=(1, 0), and when the odor of a non-odorant is determined (when the odor of an odorant is not determined), (Y1 , Y2)=(0, 1).

The odor determination unit 20 has a multilayer neural network 22 . As shown in FIG. 3, the odor determination unit 20 equipped with the multilayer neural network 22 is composed of the multilayer neural network 22, the input unit 32, the learning control unit 34, and the learning information storage unit 36. These functions are as follows: It is realized by execution of a program by a CPU of a computer circuit corresponding to neural network processing.

Odor detection values x1 to x16 from the odor sensor 18 are input to the multi-layer neural network 22 via the input unit 32, and an odorant estimated odor value y1 and a non-odorant odor estimated value y2 are output.

The learning control unit 34 learns, as learning information, multiple sets of odor detection values x1 to x16 of the odor sensor 18 detected by experimental contact with the odorant itself and city gas or LP gas to which the odorant is added. The information storage unit 36 is caused to temporarily store and store the sensed odor values x1 to x16 stored in the learning information storage unit 36 and combinations of odorant odors and non-odorant odors, which are expected output values, with a teacher. Learning information is input to the multi-layer neural network 22 via the input unit 32, and weights and biases of the multi-layer neural network 22 are learned by a learning method such as the back propagation method (error backpropagation method).

When arbitrary detected odor values x1 to x16 are input from the odor sensor 18 to the multi-layer neural network 22 which has been trained using this supervised learning information, the estimated odor value y1 of the odorant and the estimated value y1 of the non-odorant are obtained. An estimated odor value y2 is output.

Here, the estimated odor value y1 of the odorant and the estimated odor value y2 of the non-odorant are ideally as follows.
(y1, y2) = (1, 0) for the smell of an odorant
(y1, y2) = (0, 1) for non-odorant odors
When the odor detection values x1 to x16 of the odor sensor 18 are actually input to the multi-layer neural network 46, the sum of the estimated values y1 and y2 is 1, and each takes a value between 0 and 1.

(fully connected neural network)
As shown in FIG. 4, the multi-layer neural network 46 is a fully-connected neural network and is composed of an input layer 42, a fully-connected neural network 44, an iteration of intermediate layers 46 and fully-connected 44, and an output layer 48. FIG.

Here, the fully-connected neural network performs multi-class classification for classifying the odor detection values x1 to x16, which are input information, into two classes, odorant odors and non-odorant odors. In 48, two units that are the same as the two classes of interest are arranged, and the inputs to these units are made to be outputs y1 and y2 with a sum of 1 using a softmax function. Outputs y1 and y2 will indicate the probability of belonging to that class.

(learning by backpropagation)
The multilayer neural network 22 shown in FIG. 4, which is composed of an input layer 42, a plurality of intermediate layers 46, and an output layer 48, has a plurality of units in each layer, which are combined with a plurality of units in other layers, and each unit A weight and a bias value are set in , and a vector product of a plurality of input values and the weight is calculated, the bias value is added to obtain a total sum, and this is passed through a predetermined activation function to be output to the unit of the next layer. , and forward propagation is performed to propagate the value until it reaches the final layer.

To change the weights and biases of such neural networks, we use a learning algorithm known as backpropagation. Backpropagation consists of supervised learning when the network is given a data set of input value x and expected output value (expected value) y, and unsupervised learning when only the input value x is given to the network. , and this embodiment performs supervised learning.

When performing backpropagation in supervised learning, use, for example, the mean square error function as the error comparing the estimated value y* and the expected value y, which is the result of forward propagation through the network. do.

Backpropagation uses the magnitude of the error between the estimated value y* and the expected value y to propagate values from back to front of the network while correcting weights and biases. The corrected amount for each weight and bias, treated as a contribution to the error, is computed by steepest descent, minimizing the value of the error function by varying the weight and bias values.

The procedure of learning by back propagation for a neural network is as follows.
(1) Input the input value x to the neural network and perform forward propagation to obtain the estimated value y*.
(2) Calculate the error using the error function based on the estimated value y* and the expected value y.
(3) Perform back propagation in the network while updating weights and biases.

This procedure is repeated with different combinations of input values x and expected values y until the weights and biases of the neural network have the lowest possible error, minimizing the value of the error function.

In the supervised learning control of the multilayer neural network 22 shown in FIG.
(y1, y2) = (1, 0) for the smell of an odorant
(y1, y2) = (0, 1) for non-odorant odors
is used as the output value (expected value), and back propagation is performed as described above.

(Implementation of odor determination part)
The learning of the multi-layer neural network 22 shown in FIG. 4 is not performed by the gas leak alarm, but is learned by another computer device, and the learned multi-layer neural network 22 is implemented as the odor determination unit 20 of the gas leak alarm. will do.

(Binarization of odor judgment value)
As shown in FIG. 3, the odorant odor estimation value y1 and the non-odorant odor estimation value y2 output from the multilayer neural network 22 are binarized by comparators 38 and 40, and the odorant odor determination value Y1 and non-odorant odor determination value Y2.

Since the estimated odor value y1 of the odorant and the estimated odor value y2 of the non-odorant output from the multi-layer neural network 22 have a sum of 1 and take values between 0 and 1, the comparator 38, 40 is binarized to indicate either 0 or 1.

Reference values 1/2 are set in comparators 38 and 40 by reference voltage sources 38a and 40a, and estimated values y1 and y2 are binarized into 2-bit judgment values (Y1 and Y2), and the odorant It indicates the presence or absence of odor.

[Gas leak determination part]
FIG. 5 is an explanatory diagram showing a list of determination processing by the gas leak determination unit in FIG. ” is indicated by ◯, and “none” corresponding to bit 0 is indicated by ▪.

Mode 1 is a case in which the CO gas detection value G1 from the gas detection unit 10 is determined, and the determination is "CO gas leakage", the CO gas alarm light of the display unit 26 blinks, and the voice alarm unit 28 An audible alarm or voice message indicating leakage is output.

Mode 2 is a case where the hydrocarbon gas detection value G2 from the gas detection unit 10 is determined, and the odor of the odorant is determined from the odor detection unit 16, and the determination is "hydrocarbon gas leak". The hydrocarbon gas warning lamp of the display unit 26 is flashed, and the voice warning unit 28 outputs a warning sound or a voice message indicating a hydrocarbon gas leak.

Mode 3 is a case where only the odor of the odorant from the odor detection unit 16 is discriminated. An alarm sound as a preliminary alarm indicating the caution of hydrocarbon gas leakage and a voice message for calling attention are output.

In mode 4, the odor of the odorant from the odor detection unit 20 is not determined, and only the hydrocarbon gas detection value G2 from the gas detection unit 10 is determined. CAUTION", the hydrocarbon gas warning lamp of the display unit 26 is turned on, and the audio warning unit 28 outputs an alarm sound as a preliminary warning to warn of hydrocarbon gas leakage and a voice message calling attention.

The gas leak determination by the gas leak determining unit 30 is not limited to FIG. 5, and modes 3 and 4 in FIG. 5 may be determined as hydrocarbon gas leak.

[Control operation of gas leak alarm]
The control operation of the gas leak alarm device of FIG. 1 will be explained as follows. If, for example, the city gas being used at the location where the gas leak alarm is installed leaks, the hydrocarbon gas comes into contact with the gas sensor 12, and the hydrocarbon gas detection value is sent from the gas detection circuit 14 to the controller 24. output.

At the same time, the odor detection values x1 to x16 of the odor sensor 18 are input to the multilayer neural network 22 of the odor determination unit 20, and the odor determination value Y1 of the odorant is output to the control unit 24.

The gas leakage determination unit 30 of the control unit 24 determines the gas leakage of hydrocarbon gas according to mode 2 of FIG. An alarm sound or voice message indicating caution is output.

[Embodiment in which gas leakage is determined only by an odor sensor]
FIG. 6 is an explanatory diagram showing another embodiment of a gas leak alarm that detects gas leakage only by the odor detection section, and FIG. 7 is an explanatory diagram showing the functional configuration of the odor determination section in FIG.

As shown in FIG. 6, this embodiment comprises an odor detection unit 100, a control unit 24, a display unit 26, and an audio alarm unit 28, and the gas detection unit 10 shown in the embodiment of FIG. 1 is removed. , the odor detection unit 100 detects the odors of CO gas, hydrocarbon gas, and odorants.

As shown in FIG. 7, the odor detection unit 50 is composed of a multilayer neural network 22, an input unit 32, a learning control unit 34, and a learning information holding unit 36. FIG. The multi-layered neural network 22 is controlled by the learning control unit 34 to generate a teacher that combines detected odor values x1 to x16, which are input information from the odor sensor 18, and CO gas, hydrocarbon gas, and odorant odors, which are expected output values. The learning information is input to the multi-layer neural network 22 via the input unit 32, and the weights and biases of the multi-layer neural network 22 are learned by a learning method such as the back propagation method (error backpropagation method). Let

Training of the multi-layer neural network 22 is performed by another computer device, and the trained multi-layer neural network 22 is installed in the gas leak alarm.

As a result, when the odor detection values x1 to x16 from the odor sensor 18 are actually input to the multi-layer neural network 22, the CO gas estimated value y1, the hydrocarbon gas estimated value y2, and the The estimated odor value y3 of the odorant is output, compared with the reference value 1/3 set by the reference voltage sources 52a, 54a, 56a of the comparators 52, 54, 56, and binarized to obtain the CO gas judgment value G1, As the hydrocarbon gas judgment value G2 and the odor judgment value Y1 of the odorant,
If CO gas is determined, (G1, G2, Y1) = (1, 0, 0)
When determining hydrocarbon gas, (G1, G2, Y1) = (0, 1, 0)
When the odor of the odorant is determined, (G1, G2, Y1) = (0, 0, 1)
is output.

Here, since the hydrocarbon gas and the odorant are components of city gas or LP gas, when city gas or LP gas leaks, the odor determination unit 20
(G1, G2, Y1) = (1, 1, 0)
to output

The gas leakage determination unit 30 provided in the control unit 24 of FIG. 6 is the same as that of the embodiment of FIG. A caution is determined and an alarm is issued.

[Modification of the present invention]
In the above embodiment, hydrocarbon gas and CO gas are used as monitoring target gases, but it can be applied to gas leak monitoring of appropriate monitoring target gases as necessary.

Moreover, the present invention includes appropriate modifications that do not impair its purpose and advantages, and is not limited by the numerical values shown in the above embodiments.

10: gas detection unit 12: gas sensor 14: gas detection circuit unit 16, 100: odor detection unit 18: odor sensors 20, 50: odor determination unit 22: multilayer neural network 24: control unit 26: display unit 28: voice Alarm unit 30: gas leak determination unit 32: input unit 34: learning control unit 36: learning information holding units 38, 40, 52, 54, 56: comparators 38a, 40a, 52a, 54a, 56a: reference voltage source 42: Input layer 44: Fully connected 46: Intermediate layer 48: Output layer

Claims (11)

gas detection means for detecting a predetermined monitoring target gas;
odor detection means for detecting the odor of a predetermined odorant added to the gas to be monitored;
gas leakage determination means for determining a gas leakage based on the detection output of the gas detection means and the detection output of the odor detection means and outputting a gas leakage alarm;
A gas leak alarm characterized by being provided with.
The gas leak alarm according to claim 1,
The odor detection means is
an odor sensor that outputs multiple types of odor detection values according to the adsorption of multiple odorants;
an odor determination unit that inputs the plurality of types of odor detection values output from the odor sensor and determines the odor of the predetermined odorant;
A gas leak alarm, comprising:
In the gas leak alarm according to claim 2,
The odor determination unit is configured by a multi-layer neural network, and receives as learning information a combination of the plurality of types of odor detection values of the odor sensor as input information and the combination of the odor of the odorant as an output expected value. is learned by deep learning, and when the plurality of types of odor detection values from the odor sensor are input as input information, the odor of the predetermined odorant is estimated and output. .
In the gas leak alarm according to claim 3,
The gas leakage determination means determines a gas leakage when determining both the detection output of the monitored gas by the gas detection means and the detection output of the odor of the odorant by the odor detection means. A gas leak alarm, characterized by outputting a leak alarm.
In the gas leak alarm according to claim 3,
The gas leakage determination means determines a gas leakage caution when determining either one of the detection output of the gas to be monitored by the gas detection means and the detection output of the smell of the odorant by the smell detection means. and outputting a preliminary alarm for gas leakage.
The gas leak alarm according to claim 1,
The gas leakage determination means determines gas leakage by determining the gas leakage when determining the detection output of the monitored gas by the gas detection means or the detection output of the odor of the odorant by the odor detection means. A gas leak alarm, characterized by outputting an alarm.
In the gas leak alarm according to claim 3,
The multilayer neural network is a coupled neural network,
A gas leak alarm that learns by back propagation based on an error between a value output when the learning information is input and an expected value of a predetermined value.
odor detection means for outputting a plurality of types of odor detection values in accordance with the adsorption of the odor substance of the monitoring target gas contained in the monitoring target gas and the odorant added to the monitoring target gas;
A multi-layered neural network configured to learn each combination of the plurality of types of odor detection values of the odor detection means as input information and the odors of the monitored gas and the odorant as expected output values. and learning by deep learning, and when the plurality of types of odor detection values of the odor detection means are input as input information, the presence or absence of the odor of the monitored gas and the odorant is determined and output. means and
gas leakage determination means for determining gas leakage based on the result of determination of the smell of the monitored gas and the odorant by the smell determination means and outputting a gas leakage alarm;
A gas leak alarm characterized by being provided with.
The gas leak alarm according to claim 8,
The gas leak determining means determines a gas leak and outputs a gas leak alarm when both the detection of the monitored gas and the detection of the odor of the odorant by the odor determining means are determined. gas leak alarm.
The gas leak alarm according to claim 9,
The gas leakage determination means determines a warning of gas leakage and issues a preliminary warning of gas leakage when either one of the detection of the gas to be monitored by the odor determination means and the detection of the odor of the odorant is determined. A gas leak alarm characterized by outputting.
The gas leak alarm according to claim 8,
The gas leakage determination means determines gas leakage and outputs a gas leakage alarm when detection of the monitored gas by the odor determination means or detection of the odor of the odorant is determined. gas leak alarm.

Images