WO1994014014A1

WO1994014014A1 - System for assessing air quality

Info

Publication number: WO1994014014A1
Application number: PCT/NL1993/000262
Authority: WO
Inventors: Petrus Gerardus Luscure; Philomena Margaretha Bluyssen
Original assignee: Nederlandse Organisatie Voor Toegepast-Natuurwetenschappelijk Onderzoek Tno
Priority date: 1992-12-07
Filing date: 1993-12-07
Publication date: 1994-06-23
Also published as: IL107907A0; NL9202116A

Abstract

System for assessing the presence of gases and physical components in air, comprising a number of different sensors (10A-N), each sensitive to one or more different gases and/or physical components and each able to supply a signal varying with the concentration of the respective gases or physical components concerned, and means (12) for processing the signals supplied by said sensors. Said processing means (12) are embodied as a neural network to which the sensors (10A-N) are connected, in which all sensor signals are combined into one single output value, which network is adapted and programmed during a learning period in which the output value obtained for predetermined air samples is repeatedly compared with a similar value supplied by a reference panel of trained human beings or animals for the same air sample until both values are within a predetermined margin identical.

Description

System for assessing air quality

The invention relates to a system for assessing the presence of gases or physical components in air comprising - a number of different sensors, each sensitive to one or more different gases and/or physical components and each able to supply a signal varying with the concentration of the respective gases or physical components concerned, means for processing the signals supplied by said sensors. The invention relates further to a method for calibrating such a system and to various applications of such a system.

A system of the above indicated type is known from the inter¬ national patent application WO 8501351. The purpose of this prior art system is to determine the presence and concentration of a number of selected toxic, combustible or reducing gases in an atmosphere. The gas sensors in this system are different from each of the others in their response to at least one of said selected gases. The means for processing the signals supplied by the various sensors are embodied as a processor for processing the signals from the various sensors and producing an indication of each of the gas concentrations.

In many cases it is not exactly known which specific gases have to be measured and under those circumstances a system for measuring the gas concentration of a number of selected gases is not the ideal instrument to use. One such case is the quality of air for instance in buildings. Practice has proven that measuring the concentration of a selected number of gases will not always provide a relation to the actual perceived air quality and/or reveal the actual cause of a so-called "sick building syndrome". At this moment a method is available which allows to quantify air quality, as experienced by a human being, in a scientifically justified and reproducible manner. According to this method a group of persons is trained to perform direct measurements of the air quality in a standard unity, the decipol, using the human nose. One decipol is defined as the observed air quality in a non-polluted space occupied by one standard person which represents for said space a load of one olf, and which space is ventilated with 10 1/s non-polluted air. One olf is defined as the effluents of one stan- dard person. For more information about this method the attention is drawn to articles in the magazine Klimaatbeheersing, volume 20, nr. 5, May 1991, pages 153-158 and volume 20, nr. 12, December 1991, pages 365-370. This method, whereby use is made of a group of trained per¬ sons, can be applied for instance for assessing the air quality inside buildings. Especially in case of a so called "sick building syndrome", i.e. in case users of a specific building have many complaints in relation to the air quality inside the building, this method appears to provide very useful information. It happens reg¬ ularly that, on the basis of chemical or physical measurements, the origin of the complaints caused by a deteriorated inner air quality can not be identified. The human senses however are in many cases superior in assessing the air quality than chemical analysis. Spe- cifically the perceived air quality can be judged better and faster by means of the human senses than is possible with a chemical analysis. The observations, carried out by a group of trained people, have therefor provided significant results.

Bringing into action a group of trained persons has however a number of disadvantages. To start with the group has to be trained thoroughly to such a level that the group as a whole is able to assess the observed air quality in an accurately and in a reproduc¬ ible manner. The amount of time and money involved with such a training is not insignificant. Further performing observations by means of such a group of trained people in a building is relatively time consuming and labour intensive, and therefor expensive.

So there is a need for a system by means of which the air quality can be assessed without the help of a group of trained people. Similar problems are encountered in the combat against drugs. Specially trained dogs are used to detect and find drugs during house searches, cargo checks on ships or planes, searches on air¬ fields or quays, especially during check-in or check-out procedures of luggage, searches in depots and warehouses, etc. Training a dog for this purpose is labour intensive and therefore very expensive. For coverage of all places where probably drugs could be smuggled a number of dogs would be needed far beyond the number of trained dogs available. Furthermore, there are many different types of drugs, each with their own "smell" and even a well trained dog is not always able to make the distinction between a drug and another substance. So, also in this field, there is a strong need for a system by means of which the presence of gases, volatile substances or physical particles emanating in an unknown combination and in unknown concentrations from unknown quantities of drugs can be assessed.

Another field in which such a system could be very useful is the detection of explosives, for instance on board of aeroplanes, in luggage depots, etc. Animals, especially dogs are used for this purpose also but even the very sensitive nose of a well-trained dog is not always capable of detecting a small amount of explosives, taking into account furthermore that there are many different types of explosives, each with their own "smell''. Resuming there is a need for a "general purpose" system for assessing the presence of gases, volatile substances or physical particles emanating in unknown combinations and in unknown concen¬ trations from various sources such as drugs, explosives or from unknown sources such as in the case of a sick building syndrome. The invention meets these various needs by providing in gen¬ eral a system for assessing the presence of gases or physical com¬ ponents in air of the type mentioned in the first paragraph of this specification, which is according to the invention characterised in that said processing means are embodied as a neural network to which the sensors are connected, in which all sensor signals are combined into one single output value, which network is adapted and programmed during a learning period in which the output value obtained for predetermined air samples is repeatedly compared with a similar value supplied by a reference panel of trained human beings or animals for the same air samples until both values are within a predetermined margin identical.

As such various sensors are known for assessing various gases, volatile substances or combinations thereof, as well as sensors for detecting physical components such as dust particles, smoke etc. These sensors generate signals which vary dependant on the concentration of the gases or physical components for which the respective sensor is sensitive. Even if a combination of such sen¬ sors is applied, each sensitive to another gas or volatile sub- stance or to another combination of gases, or sensitive to physical components, than ultimately only a series of measurements is obtained which does not provide information about presence of spe¬ cific drugs or explosives or other sources emanating gases, vol- atile substances or physical particles. However, by using according to the invention a specially adapted and programmed neural network to which the sensors are connected and in which all sensor signals are combined into one single output value, it has become possible to perform uniform and reproducible measurements without help of a reference panel (except for calibration) such as a specially trained group of people or animals such as dogs. (It is remarked that not only dogs are used for this sort of purposes but other animals such as pigs as well).

A preferred embodiment of the system according to the inven- tion is specifically destined to assess the air quality in closed rooms, buildings or other environments, whereby the reference panel consists of a panel of trained people, and whereby after the learn¬ ing period the system provides an output signal representing a measure for the total assessed air quality as would be experienced by a human being.

Another preferred embodiment of the system according to the invention is specifically destined to detect the presence of drugs, whereby the reference panel consists of one or more trained animals such as dogs. A further preferred embodiment of the system according to the invention is specifically destined to detect the presence of explosives, whereby the reference panel consists of one or more trained animals such as dogs.

The help of a reference panel such as a trained group of people or a trained dog is still necessary for calibrating the system according to the invention. Within the scope of the inven¬ tion is a method for calibrating a system according to the main claim, characterized in that the result of the observation by the system carried out in a specific space is compared with the result of an observation in the same space by a reference panel, where¬ after in case there is a difference in both results, the parameters of the neural network are adapted, which procedure is repeated until the appearing difference is within a predetermined margin. After adapting the parameters of the neural network during the "learning phase" such, that the output value of the system corresponds within predetermined margins with the output value generated by the reference panel, the system can be used further independent of said reference panel.

The invention relates furthermore to an apparatus for con¬ trolling the operation of a ventilation system, whereby use is made of a system according to the invention for determining the assessed air quality in the space or spaces, influenced by the ventilation system. In case the decipol-value of the assessed air quality in a space rises above a predetermined margin, than the refreshing capacity of the ventilation system could be increased temporarily such, that within a predetermined period the decipol-value of the assessed air quality is decreased underneath said margin. As such it is already known to apply a separate sensor, which is sensitive for one or more specific gases and/or physical compo¬ nents such as for instance C0₂, dust particles or smoke in a venti¬ lation system.

The application of a separate sensor in an apparatus for controlling the operation of a ventilation system has however the disadvantage that only a relatively restricted number of gases and/or physical components will be observed, which gases and/or physical components not all contribute in the same degree to the total air quality as will be experienced by a human being in a space in which the ventilation system should be active. By applying a number of different sensors, each sensitive to one or more dif¬ ferent gases and/or physical components in combination with a self adapting neural network with which the signals, derived from the various sensors, are combined into one single output signal repre- senting a measure for the assessed air quality as would be experi¬ enced by a human being, one obtains in a reproducible manner a measure for the perceived air quality, without the necessity to call the assistance of a trained group of people.

The invention will be explained hereinafter with reference to the attached drawings.

Figure 1 illustrates in the form of a block diagram a simple basic embodiment of a system according to the invention.

Figure 2 illustrates in the form of a flow diagram the method for calibrating the system illustrated in figure 1.

Figure 3 illustrates very schematically the application of a portable version of a system according to the invention.

Figure 4 illustrates very schematically the application of a system according to the invention inside a fixed gate-shaped hous¬ ing.

Figure 5 illustrates in the form of a block diagram the application of a system according to the invention for controlling a ventilation system. The system, which in the form of a block diagram is illus¬ trated in figure 1, comprises a number of sensors 10A, 10B, 10C, 10N. Each of these sensors is sensitive to the presence of one or more gases, volatile substances and/or physical components such as dust particles, smoke, etc, and supplies, in case one of the respective gases, volatile substances and/or physical components is present, a signal which varies with the concentration thereof.

Each of the sensors 10A 10N is sensitive to another gas or volatile substance or combination thereof, respectively another physical component such that all sensors together are able to observe a wide range of gases and/or physical components. The sen¬ sors 10A 10N are connected to a neural network which is sche¬ matically illustrated as block 12. The configuration of such a neural network is considered to be known to the expert in this field and will therefor not be explained in detail. The neural network 12 is configurated and programmed such that the signals, which are received from the various sensors 10A 10N are pro¬ cessed and combined into one single output value which will be available at the output 13. In the illustrated embodiment this single value will be presented through an indication unit 14 to the human observer. The indication unit 14 may consist of a display screen or a simple display on which the output value is made vis¬ ible, or may consist of an acoustic generator such as a buzzer or bell which will be activated in case the output value is above a predetermined level. In other embodiments which will be described hereinafter the single output value will be supplied from the out¬ put of the neural network for instance to a control input of a ventilation system, to a memory for (temporary) storage, or in general to another system in which the obtained value can be used or processed further.

It will be clear that the number of sensors 10A 10N is not restricted to four as in figure 1 , but that any arbitrary num¬ ber can be applied. Sensors of various type are known to the expert in this field so that a detailed discussion thereof is considered superfluous.

To obtain a properly functioning neural network 12 said net¬ work has to be calibrated. For that purpose the assistance of a reference panel is necessary which during the calibration procedure has to provide reference values.

In case the system has to be adapted and programmed for assessing the air quality the reference panel is formed by a group of trained people. The group of trained people provides not only an indication "that there is something wrong with the air quality" but provides furthermore a discrete value, expressed in decipol, of the assessed air quality.

In case the system has to be adapted and programmed for detecting the presence of drugs or explosives the reference panel is formed by one or more specially trained dogs or other animals (e.g. pigs also have a very sensitive nose and successful tests have been carried out with pigs). Such an animal only indicates if he "smells" drugs or explosives but is not able to quantify his observation. Nevertheless the mere indication of the presence of the sought drugs or explosives, provided by the animal, is as such sufficient to calibrate the underlying system although probably more measurements under different conditions will be needed to obtain a correctly adapted and programmed system.

The method for calibrating a system according to the inven¬ tion is schematically illustrated in figure 2. During the calibration phase or learning phase measurements are performed both by the system according to the invention as well by the panel of trained persons. The implementation of the measure¬ ments by the system is schematically illustrated in block 20, whereas the implementation of the measurements by the reference panel is illustrated in block 22. The results of the measurements in blocks 20 and 22 are compared in block 24 and in this block values are generated which are representative for the differences between both measurements. In block 26 it is determined if the difference values are larger or smaller than a predetermined limit value. If the limit value is exceeded, than the calibration process continues with block 28, in which according to a predetermined algorithm or learning process the parameters of the neural network are adapted. After adapting the parameters the process returns to block 20, in which again a measurement is performed by the system. As long as nothing changes in the space, in which the measurement is carried out, it is not necessary to repeat the measurement by the human panel in block 22. The new measurement values, resulting from block 20, are therefor each time compared with the determined measurements in block 22 and the adaption of the parameters of the neural network in block 22 continues until the different values generated in block 24 are underneath the predetermined limit values. If that situation is obtained than the question in block 26 will be answered with "yes" and for the time being the calibration phase is ended (block 30).

This calibration process will have to be repeated a large number of times each time under different circumstances until the system under unknown circumstances directly provides a correct output value, such as a decipol value for the assessed air quality which within predetermined limits corresponds with the decipol value determined by the human panel, or the correct decision wether or not there are drugs or explosives present in the measured space. At that moment the calibration phase or learning phase will have come to an end and the system can be used as a fully skilled measuring apparatus which does not need the assistance of a refer¬ ence panel.

As is illustrated very schematically in figure 3, depending on the application of the system it could be very convenient to install the neural network, the power supply unit therefor and the indication unit in a portable housing 40 which can be carried com¬ fortably by the person 42 performing the measurements. The assembly of the various different sensors can be mounted in a separate sen¬ sor head 44 which is for instance connected to one end of an elon- gated stick or stem 46 the other end of which is carried by the person 42. An electrical multi-conductor cable can be used to con¬ nect the sensors in the sensor head 44 to the neural network in the portable housing 40. Search operations for drugs or explosives on airports, in aeroplanes, in warehouses and depots, in luggage departments, in transport containers etc can be performed conveniently using the equipment illustrated in fig. 3. In other applications, such as the luggage check on airports, it could be more convenient to install the neural network, the power supply unit therefor and the indication unit in a fixed hous¬ ing 50, which in the embodiment illustrated schematically in figure 4 is shaped as a gate. The luggage to be inspected is for instance transported on a conveyor belt 56 through said gate shaped housing 50. Two luggage parcels 52 and 54 are shown in figure 4. The sen¬ sors of the system according to the invention are installed within the housing 50 such that they measure the atmosphere within the housing to detect the presence of drugs or explosives in the lug- gage which is momentarily inside the housing 50. To exclude as far as possible any outside influences on the measurement and to iso¬ late for that purpose the inner space inside the gate shaped hous¬ ing from the environment curtains could be used to close the entrance and exit of the housing 50. One of said curtains, which as such are known, is indicated by 58. The indication unit could be incorporated in a control console 60 which is mounted to the side of the housing 50.

It will be clear that the embodiment illustrated in figure 4 could be very well combined with the X-ray gate equipment applied for luggage checks on many airports.

The system according to the invention can be applied with special advantage for influencing the operation of a ventilation system. Figure 5 illustrates schematically the application of a system according to the invention for that purpose. The actual ventilation system is schematically illustrated in figure 5 with block 34. Because details of such a system are considered to be known to the expert in this field this system will not be described in more detail. At the left hand side of figure 5 the neural net¬ work 12 is illustrated with the thereto connected sensors 10A .... 10N. The sensors are installed within the room(s) or space(s) in which the ventilation system is active or within the duct(s) used for guiding the ventilation air to said room(s) or space(s). The output signal of the neural network 12 is supplied to one input of a comparator 32 which at its other input receives a reference value from an input node 36. The output signal of the comparator 32 is supplied to the ventilation system and used therein to vary the setting of the system in an adequate manner dependent on the momen- tary assessed air quality. As long as the decipol-value of the air quality at the output of the neural network is underneath the ref¬ erence value, set at the input node 36, the comparator 32 will supply a thereto corresponding output signal to the ventilation system 34 which in response thereto will not react at all or will eventually switch over to a lower refreshing capacity. However, as soon as the decipol-value at the output of the neural network 12 exceeds the limit value, set at point 36, than in response to the respective output signal of the comparator 32 the refreshing capac¬ ity of the system 34 will be adapted such that within a predeter- mined period the decipol value at the output of the neural network 12 will be decreased to under the limit value. It will be clear that the configuration of figure 5 can be considered as a feedback system with which the assessed air quality in the space influenced by the ventilation system, can be maintained underneath or at a predetermined decipol limit value. The advantage of such a con¬ figuration is especially found in the fact that on the one hand no energy will be lost by an excessively functioning ventilation sys¬ tem, whereas at the other hand the decipol-value of the assessed air quality will be maintained at all times underneath a predeter- mined limit value.

Although above a number of specific applications of the sys¬ tem are described there are other applications in which the system according to the invention may replace a specially trained person or group of persons, or a specially trained animals such as dogs, pigs, etc to detect the source of a combination of unknown gases an/or volatile substances and/or physical particles. Adapted and suitably programmed versions of the system according to the inven¬ tion could be used to detect the presence of truffles in the earth.

Claims

1. System for assessing the presence of gases and physical com¬ ponents in air, comprising - a number of different sensors, each sensitive to one or more different gases and/or physical components and each able to supply a signal varying with the concentration of the respective gases or physical components concerned, means for processing the signals supplied by said sensors, characterised in that said processing means are embodied as a neural network to which the sensors are connected, in which all sensor signals are combined into one single output value, which network is adapted and pro¬ grammed during a learning period in which the output value obtained for predetermined air samples is repeatedly compared with a similar value supplied by a reference panel of trained human beings or animals for the same air samples until both values are within a predetermined margin identical.

2. System according to claim 1, specifically destined to assess the air quality in closed rooms, buildings or other environments, whereby the reference panel consists of a panel of trained people, and whereby after the learning period the system provides an output signal representing a measure for the total assessed air quality as would be experienced by a human being.

3. System according to claim 1, specifically destined to detect the presence of predetermined substances, which emit an unknown combination of gases, volatile substances or physical particles.

4. System according to claim 3, specifically destined to detect the presence of drugs, whereby the reference panel consists of one or more trained animals such as dogs.

5. System according to claim 3, specifically destined to detect the presence of explosives, whereby the reference panel consists of one or more trained animals such as dogs.

6. Method for calibrating a system according to one of the pre- ceding claims, characterized in that, the result of the observation by the system carried out in a specific space is compared with the result of an observation in the same space by the reference panel, whereafter in case there is a difference in both results, the para- meters of the neural network are adapted according to a predeter¬ mined algorithm, which procedure is repeated until the appearing difference is within a predetermined margin.

7. System according to any of the preceding claims, whereby the neural network is installed in a portable housing together with a suitable power supply unit and whereby the assembly of the various sensors is installed within a portable sensor head which through a suitable cable is connected to the neural network

8. Apparatus for controlling the operation of a ventilation system comprising means for varying one or more parameters of the ventilation system dependant on the assessed concentration of one or more gases and/or physical components and/or perceived air qual¬ ity in the space in which the ventilation system has to be operat¬ ive and/or in the air guiding ducts of the ventilation system, characterized in that ventilation system is coupled to an assess¬ ment system according to claim 1 or 2 whereby the sensors are installed at fixed positions within said space and the output of the neural network is connected to said parameter varying means such that during operation the assessment system supplies signals, representing the assessed air quality, to the parameter varying means, which in response thereto varies the parameters of the ven¬ tilation system such that the assessed air quality will be enhanced or maintained within predetermined limits.

*****