EP0654770B1 - Device for early detection of fires - Google Patents

Abstract

The arrangement contains a plurality of detectors which are connected to a central station and of which some are fitted with at least two sensors (1, 2) for monitoring different fire characteristics. One sensor (1) is preferably a thermal sensor and the other sensor (2) is an optical sensor. In addition, the arrangement contains means for processing the signals of the sensors. These means are arranged in a decentralised manner in the detectors and they contain a microcontroller (MCU) for conditioning the sensor signals and for signal processing for the purpose of obtaining danger signals. The danger signals are obtained in a neuronal network (NN). <IMAGE>

Description

The present invention relates to an arrangement for the early detection of fires, with a A plurality of detectors connected to a control center, some with at least two Sensors for monitoring various fire parameters are equipped with Means for processing the signals from the sensors, which are arranged decentrally in the detectors and a microcontroller for processing the sensor signals and for signal processing have with the purpose of obtaining hazard signals, the extraction of the Danger signals occur in a neural network.

Such detectors have several advantages: By relocating signal processing from The central in the detectors is the limitation of the usual communication bandwidth Connections between the control panel and detectors without influence. In addition, the observation length the signals are not subject to restrictions and the possibility of overloading the head office is practically excluded. The high redundancy of the system has also the advantage that in the event of a failure or malfunction of the main processor in the central office Detectors can trigger an alarm themselves.

The use of the neural network has the advantage that the reliability of the detector function is generally improved by a wide range of possibilities Linkage of the different signal signatures, that is the recognition pattern, exists and can also be optimally used in the neural network.

In a fire detector of the type mentioned in EP-A-0 403 659 a further neural network is connected upstream of the neural network for each sensor, which time pattern of the signals of the relevant sensor are sequentially supplied. This other neural networks represent a type of transversal filter and deliver on their Output one signal signature per fire phenomenon.

The invention is intended to further reduce the false alarm rate per detection point and the Reliability of the detectors can be further improved.

This object is achieved according to the invention in that the neural network digital filter bank is connected upstream, which receives the signals of at least one type of sensor are supplied, and which have several signal signatures at their output for the neural network or provides criteria for the fire phenomenon in question.

Through the digital filter bank, which the neural network multiple signal signatures for the fire phenomenon in question, the reliability of the detector further improved because the neural network due to the plurality of signal signatures can be trained so that its functions are fully understandable and clear.

Fig. 1

ein Übersichtsdiagramm der Signalverarbeitung im Melder,

Fig. 2a, b

ein Schema der beiden Signalpfade der Signalverarbeitung; und

Fig. 3

in Diagramm des neuronalen Netzwerks der Signalverarbeitung.

The invention is explained in more detail below with the aid of an exemplary embodiment and the drawings; shows:

Fig. 1

an overview diagram of signal processing in the detector,

2a, b

a schematic of the two signal paths of signal processing; and

Fig. 3

in diagram of the neural network of signal processing.

Fig. 1 shows an overview of the signal processing in the detector, which is divided into five stages S1 to S5 can be. The first stage S1 consists of the sensor hardware and essentially contains a thermal sensor 1 formed by an NTC sensor, one by a light pulse transmitter and an optical sensor 2 formed by a light pulse receiver, a bias network 3 for the thermal sensor 1 and an ASIC 4. The sensor hardware also includes another A / D converter 5 of a microcontroller MCU.

In a known manner, the MCU has a ROM mask that contains the operating system and the sensor software of the detector and thus all processes at the functional level, i.e. the Sensor control, signal processing as well as addressing and communication with the head office controls. The ASIC 4 contains all amplifiers and filters for the signal of the Light pulse receiver, a one-chip temperature sensor, the control electronics for the light pulse transmitter, a crystal oscillator and start-up / power management and line monitoring for the MCU. There is a bidirectional, between the MCU and the ASIC 4 serial data bus and various control lines.

The signals are in the second stage S2 following the A / D converter 5 prepared, trying through different compensations, one if possible to get an exact image of the real measurement parameters. In the third stage S3 Signal signatures or criteria extracted, which are then in a fourth stage S4 neural network NN condensed into a scalar danger signal and one Risk level can be assigned. In the fifth stage S5 is finally in one Verification level 6 the decision about the final danger level is made and together with the functional state or status to the communication interface of the MCU forwarded.

1, the first three stages S1 to S3 are from the signal of the thermal Sensor 1 and separately from the signal of the optical sensor 2, what in the Figure symbolized by two signal paths, a "thermal" and an "optical" path which is then brought together in the fourth stage S4, that is to say in the neural network are. The signal flow of the two paths through levels S1 to S3 is in 2a and 2b, and the neural network NN is shown in Fig. 3 in detail.

In the following, the thermal and then the optical signal path should now be closer The NTC temperature sensor 1 is over the bias network 3 operated pulsed and the NTC voltage is fed to the A / D converter 5. The NTC temperature data are subsequently analyzed in a stage 7, where Interruptions and short circuits are detected. In level 7 there is also the influence of small driving voltage changes to increase the measurement accuracy compensated for the measured value. Any glitches are shown below "anti-EMI" algorithm 8 removed. This limits the signal change from one Measurement to the next to certain values stored in the data memory of the MCU. Normal fire signals pass this algorithm unchanged.

The output signal of the A / D converter is then in a linearization stage 9 using an interpolation table according to the characteristics of the NTC sensor converted into a temperature value. Then in a block 10 the heat dissipation by connecting wires and plastic wall and in a block 11 the Heat capacity of the NTC sensor 1 compensated. The output signals of the blocks 10 and 11 then pass through a digital filter bank 12 and are finally in a level 13 linked with parameters. At the exit of level 13 and thus on At the end of the thermal path there are several, from the NTC signal and thus from temperature-dependent signature signals or criteria S1 to Sm are available.

In the optical signal path, a pulse generator 14 drives which every 100s is almost 100µs long current pulse, an infrared light emitting diode forming the light pulse transmitter 15, which sends a light pulse into the optical scattering space. That of any existing Smoke scattered light is collected by a lens and onto a receiver photodiode 15 'directed. The resulting photocurrent is synchronized with the transmission pulse integrated by an integrator 16. The following, still differential Voltage amplifier 17 offers several selectable gain settings. The coarse detector adjustment is then carried out. A so-called AMB filter 18th eliminates DC components and low-frequency interference from the signal. High frequencies Faults have already been eliminated by integrator 17. At the exit of the AMB filter 18 appears as a single unipolar signal from a voltage amplifier 19 is further strengthened.

The output signal of the amplifier 19 is converted into digital data in the A / D converter 5, with which the software-based signal processing begins (FIG. 1, stage S2). By forming the difference in a stage 20 between a light and a dark measurement the effective signal swing is now determined. This arrives in a block 21 and can be corrected there thanks to the availability of the ASIC temperature so that extensive compensation of the temperature drops of the optoelectronic Components done. As the last and practically continuous adjustment of the signals to one The target size is the software fine adjustment, which is also carried out in block 21. In the next block 22, tracking eliminates those signal components which are caused by very slow environmental influences (for example dustiness) are caused, and which generate a false smoke signal over time and thus change the sensitivity would.

The result of the previous processing steps is a size that the effective, filtered, adjusted, temperature compensated and tracked Represents smoke value and the direct reference for determining the hazard level forms. The last link (block 23) in optical signal processing is from Different parameter-controlled algorithms that determine the temporal behavior assess the size representing the smoke value. At the end of the optical signal processing path then the signature signals Sm + 1 to Sn are available.

The signature signals S1 to Sn of the thermal and the optical path form the Entry level L0 of a layered, neural network NN, which is shown in FIG. 3 is. The representation of the neural network NN in FIG. 1 shows that these input variables are either dependent on the temperature signal (T), or the optical signal (O) or both. The network points next to the entrance level L0 still further levels L1 to L5 with so-called neurons or nodes on. The input variables of an addition weighted with parameters are stored in these and subjected to a maximum and / or minimum linkage. The addition takes place in the with A and the maximum and / or minimum linkage in the with M designated neurons.

The maximum linkage is the nonlinear network function: yi = max (w1 * x1, w2 * x2, ..., wn * xn), [xi = input value, yi = output value] who works on the principle that "everything belongs to the strongest".

The addition is the dot product: yi = Σ wi * xi, [xi = input value, yi = output value].

In principle, all connections are possible between the neurons. In a learning phase During the development of the detector, the network can be used in a learning environment be involved. This will be through the learning effect of the network certain connections prove to be preferred and reinforce and others will atrophy as it were. Alternatively, the network can also be used without a learning phase be structured. In both cases, the weights are used for safety reasons of the network frozen.

Between the input and output levels L0 and L5 of the neural network NN there is a concentration of the respective input variables on a single one Output variable that represents a scalar danger signal. The danger signal will in a quantization stage 24 one of several, for example at least three, assigned to security levels, and this assigned to one of the security levels Signal is the output signal GS of the neural network NN.

Finally, the verification level that follows the neural network takes place 6 verification of the final danger level. The corresponding output signal GSdef is together with the functional state (Fig. 1, "Status") on the Communication interface of the MCU communicated to the control center.

• Die Messung der aktuellen ASIC-Temperatur mit Hilfe eines Einchip-Temperatursensors wurde bereits erwähnt. Diese Messung, die periodisch erfolgt, liefert einen Temperaturwert, mit dem die Temperaturgänge der optoelektronischen Bauteile softwaremässig kompensiert werde, so dass auch bei extremen Temperaturen zuverlässige Rauchdichtemessungen vorgenommen werden können.
• Die Funktionsweise der Signalnachführung wurde ebenfalls bereits erwähnt. Das Rauchdichtesignal wird von sehr niederfrequenten Anteilen befreit, um Einflüsse der Umwelt auszufiltern, die signifikant langsamer sind als Brandphänomene (beispielsweise Verstaubung). Damit wird eine sehr gute Langzeitkonstanz der Rauchempfindlichkeit erreicht.
• Regelmässig wird automatisch ein Selbsttest auf gewisse Fehler durchgeführt, der den Melder einer detaillierten Diagnose unterzieht.

Finally, some particularly advantageous properties and additional functions of the fire detector described should be mentioned:

• The measurement of the current ASIC temperature with the aid of a single-chip temperature sensor has already been mentioned. This measurement, which is carried out periodically, provides a temperature value with which the temperature responses of the optoelectronic components are compensated for by software, so that reliable smoke density measurements can also be carried out at extreme temperatures.
• The functionality of signal tracking has also already been mentioned. The smoke density signal is freed from very low-frequency components in order to filter out environmental influences that are significantly slower than fire phenomena (e.g. dustiness). This ensures a very good long-term consistency in smoke sensitivity.
• A self-test for certain errors is carried out automatically, which subjects the detector to a detailed diagnosis.

If the shift of the signal processing from the central to the detectors and the use of a neural network for signal processing for detectors Multiple sensors is particularly advantageous, so can of course also detectors be formed with only one sensor in the manner described. In addition, is still mentions that the neural network NN has a very special, a fuzzy logic represents related type and can therefore also be replaced by a fuzzy logic could.

• Diese neuronalen Netzwerke wären eine Art von Transversalfilter und hätten ein wesentlich geringeres Gedächtnis als rekursive Filter;
• am Ausgang jedes dieser neuronalen Netzwerke wäre nur je eine Signalsignatur pro Brandphänomen (Rauch, Temperatur) erhältlich, wogegen die vorgeschlagene Lösung S1 bis Sm Signalsignaturen für das Brandphänomen Temperatur und Sm+1 bis Sn Signalsignaturen für das Brandphänomen Rauch zur Verfügung stellt. Diese Mehrzahl von Signalsignaturen ist aber für die sichere Funktion des neuronalen Netzwerks NN (Fig. 3) sehr wichtig, weil man dieses dann so ausbilden kann, dass seine Funktionen voll verständlich und überblickbar sind. Und letzteres ist in einem Sicherheitssystem unbedingt erforderlich.

A very essential feature of the present arrangement is formed by the digital filter bank 12 and the block 23 (FIG. 1), it being possible in particular for the digital filter bank to contain recursive filters. If one were to use a neural network instead of this filter bank and / or block 23 and to supply this time pattern of the sensor signals sequentially, then one would have two major disadvantages compared to the proposed solution:

• These neural networks would be a type of transversal filter and have a much smaller memory than recursive filters;
• only one signal signature per fire phenomenon (smoke, temperature) would be available at the output of each of these neural networks, whereas the proposed solution S1 to Sm provides signal signatures for the temperature fire phenomenon and Sm + 1 to Sn signal signatures for the smoke fire phenomenon. However, this plurality of signal signatures is very important for the safe functioning of the neural network NN (FIG. 3), because it can then be designed in such a way that its functions are fully understandable and clear. And the latter is absolutely essential in a security system.

Claims (16)

1. Arrangement for the early detection of fires, with a number of detectors connected to a control centre, some of which are fitted with at least two sensors (1, 2) for monitoring different fire parameters, and with means for processing the signals of the sensors (1, 2) which are arranged locally in the detectors and have a microcontroller (MCU) for conditioning the sensor signals and for signal processing, with the aim of obtaining alarm signals, wherein the alarm signals are obtained in a neural network (NN), characterised in that a digital filter bank (12) is connected upstream of the neural network (NN), the digital filter bank being fed with the signals of at least one type of the sensors (1), and making available at its output to the neural network several signal signatures or criteria (S1 to Sm) for the respective fire phenomenon.
2. Arrangement according to Claim 1, characterised in that the digital filter bank (12) contains recursive filters.
3. Arrangement according to Claim 1 or 2, characterised in that the neural network (NN) has several levels (L1 to L5) with nodes (A, M), in which the input variables, weighted with parameters, undergo an addition and a maximum and/or minimum linkage.
4. Arrangement according to Claim 3, characterised in that the signal processing has a separate path for each of the two sensors (1, 2), and that the two paths are combined at the input of the neural network (NN).
5. Arrangement according to Claim 4, characterised in that the microcontroller (MCU) has a mask with the operating system and the sensor software of the detector, and a data memory, and that an ASIC (4) that contains the amplifier and filter for the signal of the receiver of the optical sensor (2), a temperature sensor, the drive electronics for the transmitter of the optical sensor, and a quartz oscillator, is allocated to the microcontroller (MCU).
6. Arrangement according to Claim 4, characterised in that the thermal path contains a first stage (S1) with a biassing network (3) for the operation of the thermal sensor (1) and with an A/D converter (5), a second stage (S2) for conditioning the signals for possible compensations, and a third stage (S3) for obtaining signal signatures, which form input variables for the neural network (NN).
7. Arrangement according to Claim 6, characterised in that the second stage (S2) has a block (7) for analyzing the output signals of the A/D converter (5) for possible errors and/or for compensation of the effects of changes in the drive voltage on the measured value and/or a block (8) for removing glitches, a block (9) for converting the measured value into a temperature value and/or a block (10 or 11 respectively) for compensating the heat dissipation and/or the thermal capacity.
8. Arrangement according to Claim 7, characterised in that in the block (8) the signal change from one measurement to the other is limited to certain values to remove glitches.
9. Arrangement according to Claim 6, characterised in that the third stage (S3) contains means for linking the output signals of the said elements, so that various signature signals derived from the temperature signals are available at the end of the thermal path.
10. Arrangement according to Claim 4, characterised in that the optical path contains a first stage (S1) with a pulse generator (14) for driving the transmitter (15) and with an integrator (16) for the signal of the receiver (15') of the optical sensor (2), as well as an A/D converter (5), a second stage (S2) for implementing any compensations, and a third stage (S3) for obtaining signal signatures, which form input variables for the neural network (NN).
11. Arrangement according to Claim 10, characterised in that a voltage amplifier (17) for the coarse adjustment is connected downstream of the integrator (16) and a filter (18) for the selective detection of the received light pulse and suppression of interference signals is connected downstream of said voltage amplifier.
12. Arrangement according to Claim 11, characterised in that a calculation of the signal pulse values is made via the filter (18) before, after and during a light pulse.
13. Arrangement according to Claim 10 or 11, characterised in that the second stage (S2) contains a block (20) for determining the signal deviation, a block (21) for compensation of the temperature outputs of the opto-electronic components and/or for the fine adjustment, and/or a block (22) for compensation of the background signal and for the elimination of signal components composed of slow environmental effects, so that the output signal of the second stage represents an adjusted, temperature-compensated and corrected smoke value.
14. Arrangement according to Claim 10, characterised in that the third stage (S3) contains a block (23) for assessing the time characteristic of the smoke value supplied by the second stage (S2) via a filter arrangement, and that the smoke value signal thus filtered forms a signature signal of the optical path.
15. Arrangement according to Claims 6 and 10, characterised in that a concentration of the input variables takes place in the nodes (A, M) of the neural network (NN), and that a scalar alarm signal is obtainable at the output level (L5) of the network, and is allocated in a quantizing stage (24) to one of several alarm stages.
16. Arrangement according to Claim 15, characterised in that a verification stage (6) for verifying the definitive alarm stage is connected downstream of the neural network (NN).

Images