Field of the invention

The present invention relates to flame detectors and methods for flame detection.

Background

EP3159861 describes how early detection of fires and flames is important in the industrial and domestic environments and describes a flame detector having a fire sensor, capable of detecting a characteristic blackbody-type radiated heat signature emitted by a flaming material and a guard sensor, for detecting an at least further part of the spectrum emitted by the material. The guard sensor serves to assist in rejecting false alarms, in use. The guard band is narrower than the fire band and each band is distinct from the other.

False alarms are disruptive and costly in an industrial environment. Whenever there is an alarm or a false alarm, it is advantageous to have reliable information to diagnose the cause of the alarm or the cause of the false alarm.

It is known (W02012/064115) to use a closed-circuit TV camera in which the camera signal is analysed for the presence of flame indicia, and (EP1851995) images of the illuminated regions can also be viewed by security personnel, e.g. transmitted to the local fire department.

The approximate electromagnetic spectrum produced by burning petrol (by way of example) may be sectioned into three approximate regions. A first region represents both the ultraviolet and visible regions of the spectrum; a second region represents the near-infrared and short-/mid-infrared, which includes a characteristic black body-type heat signature emitted by a flaming material; and a third region represents the mid-/long-infrared which includes the carbon dioxide (hereinafter C02) peak at 4.3 microns. Whilst it is not intended to be bound by theory, when a material becomes hot, for example during combustion, the amount of radiation (blackbody-type radiation) increases, together with a corresponding movement of the wavelength towards the shorter wavelengths. Hereinafter, ultraviolet may be designated and infrared may be designated 'IR\

Not all fuels contain carbon and, as such, when a fuel such as hydrogen burns, there is no€02 peak produced. It is also important to distinguish between a flame producing C02 and C02 produced by, for example, an engine. Real-world fires typically produce a large amount of dirt, soot and smoke. The presence of smoke, soot and other particulates makes fires very challenging to detect, as the smoke created by a 'dirty' flame can block the tell-tale 4.3 micron signal. A further particular disadvantage of narrowband detectors aimed at the 4.3 micron peak is that, in a situation that the fuel is burning in a confined space, carbon monoxide might be created rather than€02, which would lead to a reduced 4.3 micron peak. This can significantly affect the speed of detection. A further disadvantage of these detectors is that, as known by those skilled in the art, 4.3 micron light is blocked by regular glass and, therefore, expensive sapphire windows must be used. Additionally, the 4.3 micron peak can be readily blocked by contaminants, such as water vapour, dirt, ice and snow. Accordingly, such known detectors are often heated and must be cleaned to ensure their correct functioning, which increases the overall cost of the unit and the running cost of the unit and associated infrastructure.

For these reasons, EP3159861 describes increasing the range of wavelengths detected.

Infrared sensors come in a variety of different types. Generally speaking, the wider the effective range of detection of the sensor, the more expensive the sensor. Each sensor has a different response to temperature and its relative degradation over time. Detectors that rely upon interplay of various different sensors will give variable detection with temperature change and their performance will change over time. As such long-term detection of the unit can be compromised.

EP3159861 describes a powerful (in that it is not narrowband) flame detector which, although economically produced, does not compromise on the accuracy of detection.

W02017/065808 relates to a device comprising a flame detector and a camera, and describes images with an overlay representing the field of view of the flame detector.

Summary

In accordance with a first aspect of the invention, an apparatus for flame detection is provided, comprising: at least a first sensor having a first field of view (e.g. a cone of view), a camera having a camera field of view that encompasses the first field of view; and a processor. The processor generates a graphical indicator to be superimposed onto images from the camera to indicate the first field of view of at least the first sensor.

A second sensor having a second field of view may be provided and optionally a third. The graphical indicator may indicate a combined field of view of the sensors. First, second and third sensors preferably detect infrared radiation, or first and second sensors detect infrared and a third sensor detects ultraviolet.

The flame detection apparatus is configured to signal a fire or raise an alarm if there is a match in predetermined characteristics as sensed by the sensor(s) and/or pre-set thresholds are achieved.

The fields of view of the sensors have a common locus of overlap, and the graphical indicator preferably indicates the common locus of overlap.

A series of images are stored in a buffer and are output from the buffer when the apparatus signals a fire or raises an alarm. The camera preferably continues to output images for a predetermined period after the apparatus signals a fire or raises an alarm. The images can be output to a remote server or a cloud computing service.

The camera can to detect an obstruction that may be obstructing the or each sensor.

According to another aspect of the invention of generating a graphical indicator is provided. The method comprises determining a field of view of a first sensor on a field of view of a camera and generating the graphical indicator, indicating, within an image from the camera, the field of view of at least the first sensor.

A second field of view of at least a second sensor may be determined and a common locus of overlap for the first and second sensors may be calculated.

To determine the field of view of the or each sensor, a test light source may be tracked with the camera, e.g. a test light source that includes infrared and visible wavelengths or that comprises a visible trackable target coincident with the test light source.

The camera may have wide-band sensitivity that encompasses, at least in part, light to which the sensor is sensitive, at least in part.

Brief Description of the drawings

Preferred embodiments of the invention will now be disclosed, by way of example only, with reference to the following drawings, in which:

Fig. 1 illustrates an apparatus of a flame detector.

Fig. 2 illustrates the field of view of a camera and sensors of the flame detector. Fig, 3 Illustrates the components of the flame detector.

Fig. 4 illustrates a scene as viewed by the camera.

Fig. 5 illustrates a method of generating a graphical indicator for the camera.

Detailed description

Referring to Fig. 1, a flame detector 10 is shown comprising first, second and third IR sensors 11, 12 and 13.

The sensors have active detection zones manufactured from a material comprising lead sulphide (PbS) or indium gallium arsenide (InGaAs).

The first sensor 11 is for detecting a characteristic blackbody-type radiated heat signature emitted by a flaming material, being arranged to detect radiation having a wavelength range of: from 1.6pm to 2.4pm; from 1.7 p to 2.3pm; or from 1.8pm to 2.2pm.

The second sensor 12 is for detecting an at least further part of the spectrum emitted by the material and which serves to assist in rejecting false alarms, being arranged to detect radiation having a wavelength range of: from 0.6pm to 1.4pm; from 0.7pm to 1.3pm; from 0.8pm to 1.2pm; or from 1pm to 1.2pm.

The third sensor 13 is optional.

Each of the sensors 11, 12 and 13 is capable of detecting infrared. Alternatively sensors 11 and 12 are capable of detecting infrared and sensor 13 is capable of detecting ultra violet.

Sensor 12 may be arranged to detect a part of the radiation of a flame associated with artificial light or sunlight.

Preferably, the first sensor and/or second sensor is/are arranged to operate in the wavelength region of less than about 4pm, less than about 3.2pm, less than about 3pm, or less than about 2.55pm.

Most preferably, the first sensor and/or second sensor are arranged to operate in a wavelength region of from about 0.6pm to about 3.2pm, or from about lpm to about 3.2pm, or from lpm to 2.2pm. Preferably, detection at the second sensor and the first sensor is arranged to be separated by 0.1pm to lpm, or by 0.2pm to 0.8pm, by 0.5pm to 0.7pm, or by 0.6pm. Most preferably, the second sensor detects over a range of about 0,2mih and the first sensor detects over a range of about

0.4mhi.

Advantageously, by acting upon the specific heat signature of a flame across the spectrum, these flame detectors and associated methods are capable of detecting fires from practically all fuels, whilst rejecting false alarms.

Each sensor detects radiation from a cone of view of that sensor. One such cone 24 is shown in Fig. 1.

A camera 15 is mounted in the detector 10. The camera is factory-set to have a field of view 22 that encompasses the cones of view of the three (or two) sensors. This is illustrated in Fig. 2.

Fig. 2 shows the field of view 22 of the camera 15 and the cones of view 24, 25 and 26 of the three sensors 11, 12 and 13 respectively (of which one is optional). These cones of view are typically (but not necessarily) approximately equal in size and are closely aligned to coincide but inevitably have some degree of non-overlap. They have a common locus of overlap 27.

In factory testing, various lights of different wavelengths (or indeed flames) are presented at a certain distance D from the detector, and the responses of the various sensors are measured so as to ascertain the locus of overlap 27. This is the active area of the detector. It is the area in which all (or both) the sensors are reliably active and it is the area in which a flame can reliably be detected (without undue false alarms).

By providing a camera 15 mounted in the detector 10, pointing in the same direction as the sensors 11, 12 and 13, the camera is in a position to "see" the same view as the sensors. The camera captures activity within its field of view at a suitable rate, e.g. 50 to 200 images per second, and saves these in a buffer (e.g. a first-in-first-out buffer, as shown in Fig. 3).

Referring to Fig. 3, the detector 10, the sensors 11, 12 and 13 and the camera 15 are shown. Their outputs all feed into a processor (or microcontroller) 30. Images from the camera 15 are input into buffer 31 (shown as part of the processor 30, but it can be separate processor). The buffer is capable of storing a rolling window of T1 seconds of images (for example 10 seconds to 20 seconds but could be up to a minute). The size of the buffer is a matter of choice.

From the buffer, when triggered by an alarm, the last T1 seconds of images are output by a suitable output transmitter 32 (which is illustrated as a radio transmitter, but can be a closed-circuit TV or simply a memory for access later) or other output 33. These T1 seconds can be output by simply reading out the contents of the buffer while a further T1 seconds are being read in, or the buffer may have parallel output capability, in which case the entire buffer is output in one operation. The latter is advantageous, because it is advantageous to output the contents to a remote location in case the processor and its buffer are consumed by fire. (Alternatively, the processor and buffer can be located at a remote location, remote from the sensors and camera). After the alarm, the camera continues to capture activity, which it saves in the buffer 31 and outputs (via output transmitter 32 or otherwise) for a second window of T2 seconds. T2 may be a period of several minutes, e.g. 10 minutes to an hour. After T2, the controller reverts to normal operation.

The images can be compressed prior to storing in the buffer (whereby storage in the buffer is more efficient) or on reading out from the buffer (whereby compression is not performed unnecessarily and there is no delay storing into the buffer, which can be important if what is being recorded is an explosion). If uncompressed images are stored in the buffer, they can be compressed at the time of sending. Thus, the camera 15 allows for image capture before and after an activation, e.g. alarm or fault.

Thus, a first window, Tl, prior to the alarm event and a second window, T2, after the event are captured and can be relayed to a remote location where they can be analysed by fire personnel or by technical personnel.

Thus, the camera 15 allows for image capture before and after an activation, e.g. alarm or fault.

Images captured can be remotely reported via output 33 and managed and/or manipulated. Thus, for example, output 33 may be an Internet Protocol interface allowing reporting to a remote server or a cloud computing service. At the remote service, diagnostics can be conducted, in particular in relation to activity within the locus 27 of operation of the detector, but also outside that locus.

Thus, for example, where generation of an alarm requires a predetermined modulation of a signal, the signal with the cone of operation of that sensor (or within the locus 27 of more than one sensor) can be analysed to confirm that it showed the predetermined modulation and therefore that the alarm was correctly triggered.

Thus, for example, if the line of sight of a sensor has been compromised, e.g. by obstruction, this can be identified. Examples include identifying an increase or decrease in radiation (indicative of reflection or obstruction). Such an increase or decrease may have little or no modulation or the wrong modulation (too fast or too slow). Advantageously, the detectors are capable of being used behind standard glass, such as to separate them from dirty environments, whilst still maintaining their function. As such, this provides both practicality and an economic advantage over narrowband C02 detectors. If the glass is soiled in front of one or all of the sensors, such conditions can be diagnosed.

Images captured by the camera outside the locus 27 can be useful to fire and security agencies for diagnosing the cause of an alarm or the cause of a fire.

Where an alarm is triggered by a first modulation of a first signal from a first sensor and a second modulation of a second signal from a second sensor, the signals with the cones of operation of those sensors can be analysed against those modulations. This can be extended to a third sensor and third modulation (which may be negative in the sense of suppressing an alarm if positively sensed).

Where it is discovered that the alarm was a false alarm, an adjustment of the alarm criteria, e.g. an adjustment of the triggering modulation, can be sent back over the IP interface 33 to the controller 30 to adjust the trigger conditions for a future alarm.

To determine whether there has been a false alarm, it is most useful to view the images prior to the event to ascertain what has triggered the sensors, and it is particularly useful to know exactly what the sensors themselves "see". To assist in this analysis, an indication is provided, within the captured scene, in the form of one or more graphical indicators of the cones of view 24, 25 and 26 of the sensors and/or their locus of overlap 27. This is described with reference to Fig. 4.

Fig. 4 shows the scene as viewed by the camera, with a superimposed graphical indicator 40, for presentation on a screen. Indicator 40 is generated by processor 30 (in a manner described below) and superimposed on images from the camera that are input into the buffer 31 or output from the buffer 31.

Referring to Fig. 5, the example will be described in which indicator 40 represents the locus of overlap 27 of the three sensors.

On entering a testing mode 51 in the factory, the cone of the first sensor 1 is tested, 52. This is done by presenting, within the field of view 22 of the camera 15, a test light source that includes light within the range of sensitivity of the first sensor 11 - i.e. from about 1.6pm to about 2.4pm; from about 1.7pm to about 2.3pm; or from about 1.8pm to about 2.2pm.

The x and y co-ordinates of the test light source are tracked by the camera. This can be achieved in various ways. For example, the test light source can be a wideband light source that includes IR and visible wavelengths (e.g. an electric bulb). Alternatively, if the camera is not sensitive to the test light source, the test light source can have a suitable visible trackable target co-incident with the test light source (e.g. the source passes through a hole in disc having trackable black-and-white or contrasting fiducial markers). Alternatively, the camera may have wide-band sensitivity and may be sensitive to the IR range of the test light source.

The test light source is presented across a number of points in the field of view 22, e.g. by performing a raster scan that includes the field of view 22. The processor notes the x and y co-ordinates at which the sensor 11 senses the presence of the test light source. In this way, the processor identifies the cone of view 24 of the sensor 11. (More exactly, the sensor identifies the cross-section of that cone at distance D, from which the cone is derivable if required.)

The operation is repeated in step 53 to test sensor 12 using a light source having a wavelength that corresponds to the wavelength of sensor 12 (0.6pm to 1.4pm; 0.7pm to 1.3pm; 0.8pm to 1.2pm; or lpm to 1.2pm). Thus cone of view 25 is derived. The operation is repeated in step 54 with a suitable light source to derive cone of view 26. It will be appreciated that steps 52, 53 and 54 can be performed simultaneously, in a single scan, by suitably presenting a light source that can be tracked by all three sensors and the camera.

In step 55, the processor 30 calculates the locus of operation 27 of the three sensors. This can be calculated as an AND operation of the cones of view of the individual sensors, or as an OR operation, or as a combination of an AND of two of them and an OR of the other (or an OR of two with an AND of the other). These will be explained.

Having calculated the locus of operation, the graphical indicator 40 is generated in step 56 representing that locus visually. The scene captured by the camera 15 and the generated graphical indicator 40 are sent to a screen for presentation. The screen may be a camera display, a mobile phone screen, or any monitor, display or screen that is connected to the camera. In an example, the edge of the locus is marked with a line or a dotted line. In an example, the area outside the locus is shaded darker.

By inserting the graphical indicator 40 into the scene captured by the camera 15, a test engineer is able, at a later time, to view exactly what was in the cones of operation of the sensors that caused the sensors to trigger an alarm (or to fail to suppress an alarm).

For example, optical filters of crystalline materials with different optical characteristics can be placed in front of the sensors 11, 12, 13, which only allow radiation of a particular range of wavelengths through. One may have a wavelength region of 0.6 to 1.4 microns and a second have a wavelength region of about 1.6 to 2.4 microns. The wavelength region of 0.6 to 1.4 microns defines a 'guard band' and such a sensor can be considered a guard sensor. The wavelength region of 1.6 to 2.4 microns defines a 'fire band' and such a sensor can be considered a fire sensor.

One sensor may be of a type that is opaque to visible light, preventing substantially all light below 0.8 microns from reaching the sensor.

In use, electromagnetic radiation emitted from a flaming material is incident upon the detector 10. Owing to the optical filtering, radiation which is outside of the desired transmission wavelengths is substantially hindered and prevented from being transmitted to the sensors. At the sensors, the intensity of radiation signals in the desired wavelength ranges can be easily detected. Software in the processor 30 compares the characteristics of the detected signals with those of a flame. For example the signal amplitude, modulation regularity and proportional differences at specific wavelengths which are considered intrinsic to practically all flame types.

The detector 10 will signal a 'fire' if there is a match in various characteristics and if pre-set thresholds are achieved, and activate a corresponding LED and switching relay in the fire monitoring control equipment. Additionally, signal filtering may occur in processor 30, in combination or separately to analysis of the frequency of modulation of the source of radiation and/or analysis of the ratios of wavelengths being detected. In particular, for a 'fire' to be signalled, an intensity of radiation F detected in the 'fire band' must be greater than an intensity of radiation G detected in the 'guard band' and both sensors must each be receiving a signal (F > 0 and G > 0).

In order to improve performance with respect to false alarms, flame flicker analysis can be included during processing, which has the effect of discounting regularly modulated radiation which is typically emitted by simple hot objects, as compared to a flaming material. False alarm rejection can be further enhanced by considering the flame flicker produced during the burning of a material. For example, a natural fire will always have some turbulence created by differences within the fuel and airflows. Through looking at these phenomena, it is possible to create a detector which rejects virtually all false alarms. In particular, this is partly achieved through considering only those signals having a frequency of between, say, 1 and 25 Hertz.

When a technician views the images after an alarm event, the technician can immediately see what has occurred within the locus 27. Only signals that have the necessary characteristics

(flicker/modulation) within that locus should trigger an alarm, i.e. where all three sensors detect the predefined condition. One or more of the sensors may be present for false alarm prevention - e.g. for detection of signals that are not indicative of a flame - e.g. visible light of high frequency in the absence of IR. In such a case, it is useful to know what was within the cone of view of that sensor that perhaps should have prevented an alarm. In this case, the locus 27 may include the cone of view of that (or those) sensors.

A so-called sensory gap is provided between the fire band and the guard band, in which no or practically no detection occurs. Preferably, the guard sensor is arranged to detect the intensity of radiation from a part of the spectrum associated with artificial light or sunlight.

Preferably, the fire sensor and/or guard sensor are arranged to operate in the wavelength region of less than about 4mih, less than about 3.2pm, less than about 3pm, or less than about 2.55pm. Most preferably, the fire sensor and/or guard sensor are arranged to operate in a wavelength region of from about 0.6pm to about 3.2pm, or from about lpm to about 3.2pm, or from about lpm to about 2.2pm.

Preferably, detection at the distinct guard band and fire band is arranged to be separated by a sensory gap of about 0.1pm to about lpm, of about 0.2pm to about 0.8pm, of about 0.5pm to about 0.7pm, or of about 0.6pm. Most preferably, the guard band detects over a range of about 0.2pm and the fire band detects over a range of about 0.4pm.

Preferably, the fire sensor is arranged to detect radiation having a wavelength range of: from about 1.6pm to about 2.4pm; from about 1.7pm to about 2.3pm; or from about 1.8pm to about 2.2pm; or other distinct ranges within any of those extremes.

Preferably, the guard sensor is arranged to detect radiation having a wavelength range of: from about 0.6pm to about 1.4pm; from about 0.7pm to about 1.3pm; from about 0.8pm to about 1.2pm; or from about lpm to about 1.2pm; or other distinct ranges within any of those extremes.

Radiation of regular modulation can be filtered out or rejected. A flame is preferably detected only upon detecting radiation of irregular modulation at about 1 Hz to about 30Hz or about 1 Hz to about 25Hz.

Preferably, the fire sensor and the guard sensor comprise a common sensor comprising PbS or InGaAs. Most preferably, the sensors are arranged to have different optical filters.

Preferably, the fire sensor and the guard sensor are different sensors; however, they could be the same sensor arranged to act firstly as a fire sensor and secondly as a guard sensor or vice versa. Preferably, the method comprises detecting in a wavelength region of the spectrum of less than about 4pm, less than about 3.2pm, less than about 3pm or less than about 2.55pm.

Preferably, the method comprises prising detecting in a wavelength region of from about 0.6pm to about 3.2pm, from about lpm to about 3.2pm, or from about lpm to about 2.2pm. Advantageously, a 'dirty' flame does not block the heat signature produced by the flaming material and, therefore, does not prejudice detection.

A phenomenon such as sunlight tends to move slowly across the face of the detector and this will tend to hit one sensor and then another. This may cause a fire alarm activation if the light is modulated externally by, for example, a tree moving in the wind, however, the third sensor and the way the signals are compared reduces this occurrence of false alarms. In particular, there may be two decisions made - a comparison between the first guard band and the fire band, and a comparison between a second guard band and the fire band. False alarm rejection can be further enhanced by considering the flame flicker produced during the burning of a material.