Temperature modulated semiconductor gas sensor with analysis of a step response

Abstract

A computer-implemented hazard detection method using at least one processing unit in signal communication with at least one storage device and a gas-sensitive sensor element, wherein the sensor element is configured to be in environmental contact with an external atmosphere, comprises (i) applying a temperature sequence comprising multiple target temperature levels for a predetermined duration to the sensor element, by optionally using a heater, (ii) obtaining at least one electrical property measurement of the sensor element, (iii) processing the at least one electrical property measurement using the at least one processing unit to produce a data set, and optionally saving the data set to the at least one storage device, wherein the data set comprises an electrical property value list of at least one electrical property value, wherein the at least one electrical property value is representative of the at least one electrical property measurement at one of the multiple target temperature levels or during a transition between one pair of target temperature levels, and a change time value list with at least one change time value, wherein the at least one change time value is representative of the change time of the at least one electrical property measurement at one of the multiple target temperature levels or during a transition between one pair of target temperature levels, and / or electrical property change value list with at least one electrical property change value, wherein the at least one electrical property change value is representative of the change of the electrical property change at one of the multiple target temperature levels or during a transition between one pair of target temperature levels, (iv) determining a result based on the data set and a prediction data set stored on the at least one storage device, wherein the result comprises fake abnormal, normal or pre-alarm, and (v) determining whether a hazard is present based on the result.

Description

[0001] Eisenfiihr Speiser

[0002] Munich, 14 October 2025

[0003] Our Ref.: MA10048-02WO MVO / fgo / ssm

[0004] Appiicant / Proprietor: Minimax Viking Patent Management GmbH office reference: Subsequent to EP 24206445.9

[0005] Minimax Viking Patent Management GmbH IndustriestraBe 10 / 12, 23840 Bad Oldesloe, DE

[0006] Temperature modulated semiconductor gas sensor with analysis of a step response

[0007] The present invention relates to a computer-implemented hazard detection method. The invention relates further to a hazard detection system. Furthermore, the invention relates to a computer-implemented method for producing a prediction data set. Even further, the invention also relates to a machine learning computing system.

[0008] 5 Background of the Invention

[0009] MOX gas detectors have become a vital technology for detecting and measuring gas concentrations in various environments. These detectors operate based on the principle that certain metal oxides change their electrical resistance or other properties in the presence of specific gases. This characteristic allows them to detect and quantify gases such as0 carbon monoxide (CO), methane (CH4), nitrogen dioxide (NO2), ammonia (NH3), and volatile organic compounds (VOCs).

[0010] International patent application published as WO 2009 / 135524 A1 describes a danger alarm comprising a gas-sensitive semiconductor sensor device, an optical smoke detection device and an evaluation unit. The evaluation unit is adapted to carry out an evaluation of5 a first output signal of the gas-sensitive semiconductor sensor device and of a second output signal of the optical smoke detection device. The gas-sensitive semiconductor sensor device comprises a metal oxide semiconductor gas sensor and / or a gas-sensitive field effect transistor. Further provided is a method for recognizing a danger situation using the danger alarm. The evaluation optionally comprises a combining and / or comparing of the two output signals. Furthermore, the evaluation is carried out relative to a temporal progression of the two output signals.

[0011] German patent application published as DE 10 2012 110 095 A1 discloses a method for detecting at least one oxidizable and I or reducible gas in an atmosphere surrounding a sensor surface of a semiconductor gas sensor, in particular a metal oxide semiconductor gas sensor, which in addition to this at least one gas to be detected also contains a further gas or gas mixture as a main component the atmosphere. In the disclosed method, the following steps are provided: data recording of a sensor measured variable of the semiconductor gas sensor at several different sensor temperatures in a predetermined temperature range, the semiconductor gas sensor of the atmosphere with the at least one (to be detected) oxidizable and I or reducible gas as well as the main component of the atmosphere forming further gas or gas mixture, for example air. Normalization of the recorded data by means of a temperature-dependent normalization variable provided. Comparison of the data standardized in this way with at least one data set of standardized data of the sensor measured variable at corresponding sensor temperatures for an atmosphere with at least one oxidizable and I or reducible reference gas of known concentration and the further gas or gas mixture forming the main component of the atmosphere. Detection of the at least one gas to be detected on the basis of the comparison results as the reference gas or at least one of the reference gases. It is further described a corresponding gas sensor device with at least one semiconductor gas sensor.

[0012] Despite the widespread use and proven efficacy, existing MOX gas detectors face several challenges and limitations. Traditional MOX sensors often struggle with issues such as selectivity. For example, differentiating between multiple gases present in the environment can be difficult, leading to cross-sensitivity and inaccurate readings. A further problem relates to the sensitivity. While MOX detectors are generally sensitive, there is a continuous demand for even higher sensitivity to detect lower concentrations of gases, which is crucial for early warning systems and precise environmental monitoring. Moreover, the speed at which a detector can identify and measure the presence of a gas, i.e. the response time, is critical. Faster response times are necessary for applications like industrial safety and emergency response. Further, long-term stability and resistance to environmental factors such as humidity, temperature variations, and contaminants are essential for reliable operation over extended periods. It is therefore an objective of the present invention to overcome those drawbacks. Moreover, while known MOX gas detectors are reliable in known scenarios, they are rather unreliable in unknown scenarios. It is therefore a further objection of the application to provide a hazard detection system that is reliable in unknown scenarios.

[0013] These challenges highlight the need for advancements in MOX gas detector technology. Recent research has focused on developing novel metal oxide materials, improving sensor design, and integrating advanced signal processing techniques to enhance the performance of MOX detectors.

[0014] The present invention aims to address these issues by introducing innovative approaches to improve the selectivity, sensitivity, response time and stability of hazard detection.

[0015] Summary of the Invention

[0016] The present invention pertains to devices and methods for detecting and measuring concentrations of various gases using metal oxide semiconductor technology for hazard detection, such as fire detection. Hazard detection systems are widely used in environmental monitoring, industrial safety, and indoor air quality assessment due to their sensitivity, selectivity, and rapid response to gaseous compounds. The invention addresses improvements in the design, sensitivity, functionality, and efficiency of hazard detection systems, enabling more accurate and reliable gas detection in various applications, in particular for hazard detection.

[0017] According to a first aspect of the invention a computer-implemented hazard detection method using at least one processing unit in signal communication with at least one storage device and a gas-sensitive sensor element, wherein the sensor element is configured to be in environmental contact with an external atmosphere, comprises (i) applying a temperature sequence comprising multiple target temperature levels for a predetermined duration to the sensor element, by optionally using a heater, (ii) obtaining at least one electrical property measurement of the sensor element, (iii) processing the at least one electrical property measurement using the at least one processing unit to produce a data set, and optionally saving the data set to the at least one storage device, wherein the data set comprises an electrical property value list of at least one electrical property value, wherein the at least one electrical property value is representative of the at least one electrical property measurement at one of the multiple target temperature levels or during a transition between one pair of target temperature levels, and a change time value list with at least one change time value, wherein the at least one change time value is representative of the change time of the at least one electrical property measurement at one of the multiple target temperature levels or during a transition between one pair of target temperature levels, and / or electrical property change value list with at least one electrical property change value, wherein the at least one electrical property change value is representative of the change of the electrical property at one of the multiple target temperature levels or during a transition between one pair of target temperature levels, (iv) determining a result based on the data set and a prediction data set stored on the at least one storage device, wherein the result comprises fake abnormal, normal or pre-alarm, and (v) determining whether a hazard is present based on the result. This hazard detection method benefits improved sensitivity by leveraging known situations stored as the prediction data set and comparing it with the data set comprising measured data. Optionally, the hazard detection method uses a heater to apply the temperature sequence to the sensor element. Within the context of the present invention, the term “during a transition between a pair of target temperature levels” means in specific one or more points in time during the transition. In other words, an electrical property change value is determined at a temperature between a pair of target temperature levels. For example, when the processing unit sets the temperature to a higher level, there is a de facto temperature transition at the sensor element until the set temperature level is reached. The present invention is able to obtain an electrical property change value also at a point in time during such a transition between a pair of target temperature levels. Within the context of the present invention, the term “fake abnormal” relates to a detection of an event, but is determined based on the data and the prediction data to be a non-hazard event. “Fake abnormal” may also relate to manipulated data. “Fake abnormal” is for example a true (or false...) negative. “Normal” relates to a state where no event is detected, e.g. the result is within the boundaries of a standard setting or baseline. “Pre-alarm” is opposed to “fake abnormal” and relates to a detection of an event which instead is determined based on the data and the prediction data to be a hazard event. A “fake” result may also relate to a manipulated state, wherein for example the data set comprises artificial electrical property values.

[0018] In some embodiments of the computer-implemented hazard detection method the at least one electrical property value and / or the at least one change time value and / or the at least one electrical property change value is / are modelled from the corresponding measurement using a model for modelling transient transitions, preferably a PT1 -model or a PT2-model, wherein the Levenberg-Marquardt algorithm can be used for determining the electrical property values. PT1 is also known as 1st order lag element whereas PT2 is also known as 2nd order lag element. By modelling the transient transitions using preferable models the accuracy of the detection method can be further improved. In some embodiments of the computer-implemented hazard detection method step (iii) further comprises baseline correcting the electrical property value list and / or the change time value list and / or the electrical property change value list by detecting changes of values in each list and determining a baseline for each list based on the changes of values. As measured data can comprise a drift, a baseline correction as proposed helps in improving the accuracy and sensitivity of the detector further. As measured data can comprise a drift, a baseline correction as proposed helps in improving the accuracy and sensitivity of the detector further. For an improved model determination, outliers in the measurements may be ignored. After the baseline is determined and the electrical property values are corrected data analysis methods well known in the art such as linear regression or least squared method are used for hazardous detection.

[0019] In some embodiments of the computer-implemented hazard detection method the prealarm relates to a hazard, preferably a smoke hazard ora fire hazard. The employed hazard detection method is specifically suited for detecting smoke or fire hazard. However, the method may also be utilized for detecting other harmful gases or substances in the atmosphere that are not primarily released during a fire incident, such as methane, propane, butane, ammonia, nitrogen oxide, sulfur dioxide, carbon monoxide or volatile organic compounds. In general, while the computer-implemented method is used for hazard detection, the method may also be used for non-hazard detection, such as for detection of substances or particles in the atmosphere.

[0020] In some embodiments of the computer-implemented hazard detection method the sensor element comprises a sensing conductor. Further, the sensor element is a semiconductor.

[0021] In some embodiments of the computer-implemented hazard detection method the external atmosphere is an atmosphere inside a separate environment, preferably a building. Preferably the method is utilized in interiors, such as inside buildings, cellars, bunkers, cars, planes or boats. However, the method may also be utilized outside of interiors.

[0022] In some embodiments of the computer-implemented hazard detection method the sensor element comprises a metal-oxide semiconductor. Preferably, the metal-oxide semiconductor comprises silicon.

[0023] In some embodiments of the computer-implemented hazard detection method the at least one electrical property measurement is obtained by applying an electrical signal with either constant voltage or constant current to the sensor element and analyzing either current or voltage of the electrical signal using the at least one processing unit. A change of resistance of the sensor element based on the composition of the external atmosphere adjacent the sensor element may be noticeable by the at least one processing unit via a changing voltage or current. Accordingly, the at least one processing unit may be configured to detect and preferably record such changing voltage, current or resistance of the sensor element.

[0024] In some embodiments of the computer-implemented hazard detection method the temperature sequence comprises a step function of multiple step-wise constant, preferably adjacent, temperature levels. By using temperature sequences the sensitivity of the detection method can be improved. At or between different temperature levels the sensitivity of the sensor element for hazard detection changes. A heating of the sensor element to different temperature levels can thus improve the sensitivity of the hazard detection method.

[0025] In some embodiments of the computer-implemented hazard detection method the change time is the time required for the time course to transition from a first electrical property value attributed to a first target temperature level, to a second electrical property value attributed to a second target temperature level.

[0026] In some embodiments of the computer-implemented hazard detection method the change time is a rise and / or fall time. Preferably, the rise and / or fall time may be a response time.

[0027] In some embodiments of the computer-implemented hazard detection method the method further comprises the steps of (vi) collecting a group of, preferably consecutive, data sets, (vii) selecting at least one value index of the electrical property value list and / or the change time value list and / or the electrical property change value list, (viii) generating a value series comprising values of the electrical property value list and / or the change time value list and / or the electrical property change value list of each of the group of data sets, corresponding to the selected value index, (ix) determining a change and / or a rate of change of the values of the value series, (x) classifying of each value index based on comparing the change and / or rate of change of the values of the value series to at least one predetermined change threshold and / or rate of change threshold, modifying the data set to disregard or reduce weighting of individual values based on the classification of each value index. The employment of said steps further enhances the sensor sensitivity. By recording and analyzing a group of data sets in preferably similar or non-similar environments, the sensor can be trained for either environments. Further, the recorded data can be used for noise reduction to improve the hazard detection method. According to another aspect of the invention a hazard detection system, preferably a fire detection system, comprises at least one processing unit in signal communication with at least one storage device and a gas-sensitive sensor element, wherein the hazard detection system is configured to execute the computer-implemented method according to another aspect of the invention.

[0028] In some embodiments of the hazard detection system the at least one processing unit is further configured to communicate the result of the computer-implemented method to a fire alarm system. The at least one processing unit may comprise a network element to communicate a result of the hazard detection method, preferably to a fire alarm system. For example, the at least one processing unit may be designed as a system-on-a-chip comprising a computing unit and / or memory and / or an antenna and / or a network interface. The at least one processing unit may be further configured to be controlled by an external processing unit.

[0029] In some embodiments of the hazard detection system the sensor element comprises a semi-conductor, preferably a metal-oxide semiconductor. The metal oxide semiconductor may preferably comprise silicon.

[0030] In some embodiments of the hazard detection system the hazard detection system is configured to be in environmental contact with an external atmosphere, preferably an atmosphere inside a separate environment, such as a building.

[0031] According to another aspect of the invention a computer-implemented method for producing a prediction data set, preferably a prediction table, comprises (a) providing individually external atmospheres representative of at least one known hazard and / or at least one known non-hazard, (b) producing a group of training data sets by applying method steps (i) to (iii) of any of the methods according to another aspect of the invention with each of the provided external atmospheres, (c) training a machine-learning algorithm with the group of training data sets and a group of indicators representative of the hazard state of the provided external atmospheres, and (d) generating a prediction data set using the machinelearning algorithm, for determining a result that is either fake abnormal, normal or pre-alarm based on a data set, wherein the data set is preferably obtained by any method according to another aspect of the invention. By combining known hazard and non-hazard scenarios with measurement data, training data sets that refer to hazard or non-hazard situations are generated for training of the algorithm. Using this method the machine learning algorithm can be trained to distinguish between hazard and non-hazard situations when measurement data is inputted to the algorithm.

[0032] In some embodiments of the method for producing a prediction data set the method further comprises the step of (e) saving the generated prediction data set to the at least one storage device.

[0033] In some embodiments of the method for producing a prediction data set the prediction data set is produced by applying method steps (i) to (iii) multiple times of the computer-implemented method according to another aspect of the invention individually with external atmospheres specific to different hazards and / or non-hazards to train the machine-learning algorithm.

[0034] In some embodiments of the method for producing a prediction data set the hazard is a smoke hazard or a fire hazard.

[0035] According to another aspect of the invention, a machine learning computing system is configured to execute the computer-implemented method for producing a prediction data set. The machine learning computing system can comprise supervised learning and / or semisupervised learning and / or reinforcement learning and / or unsupervised learning. For example, the machine learning system can comprise linear regression, logistic regression, decision trees, random forests, support vector machines, k-nearest neighbors or neural networks. Furthermore, the machine learning system can also comprise k-means clustering, hierarchical clustering, principal component analysis, anomaly detection or association rules. Further, the machine learning computing system can comprise Q-learning, deep Q- networks or policy gradient methods.

[0036] A machine learning computing system according an aspect of the invention comprises at least one processing unit for executing the computer-implemented method, a memory for storing and reading the prediction data set and at least one interface to access the data sets produced by a hazard detection system according to another aspect of the invention.

[0037] The present disclosure is not limited to the disclosed methods and systems. Rather, the skilled person understands that sub-combinations of the disclosed systems and methods are also comprised by the disclosure. Unless physically or logically impossible, elements of the described methods and systems may be left out and the left over sub-combinations are also comprised by the current disclosure. Brief Description of the Drawings

[0038] In the following, embodiments of the present invention are illustrated and explained by the figures. Herein shows

[0039] Fig. 1 illustrates a method for hazard detection according to an embodiment of the invention,

[0040] Fig. 2a shows a block diagram of a hazard detection system according to an embodiment of the invention,

[0041] Fig. 2b shows a block diagram of a hazard detection system according to another embodiment of the invention,

[0042] Fig. 3 illustrates a method for hazard detection according to another embodiment of the invention,

[0043] Fig. 4 depicts a method for producing a prediction data set according to an embodiment of the invention,

[0044] Fig. 5 depicts a machine learning computing system according to an aspect of the present invention.

[0045] Detailed Description of the drawings

[0046] Fig. 1 illustrates a method for hazard detection according to an aspect of the invention comprising steps S1 to S5. The depicted method is preferably computer-implemented and uses a hazard detection system 200 as shown in Fig. 2a according to another aspect of the invention.

[0047] As depicted in Fig. 2a, the hazard detection system 200 comprises at least one processing unit 110. The processing unit 110 is in signal communication with at least one storage device 120. Further, the hazard detection system 200 comprises a gas-sensitive sensor element 130 and the processing unit 110 is in signal communication with the gas-sensitive sensor element 130. The gas-sensitive sensor element 130 is configured to be in environmental contact with an external atmosphere A. The computer-implemented method for hazard detection as shown in Fig. 1 comprises the step S1 , wherein a temperature sequence comprising multiple target temperature levels is applied to the sensor element 130 for a predetermined duration. The temperature sequence can be applied by using a heater 140 (not shown). Further, the temperature sequence can also be applied solely via exposing the sensor element 130 to the external atmosphere such that the sensor element is heated by the external atmosphere A.

[0048] In step S2, at least one electrical property measurement of the sensor element 130 is obtained. The measurement is obtained from the sensor element 130 and is communicated to the processing unit 110 for evaluation. The measurement corresponds to an electrical property of the sensor element 130, such as for example resistance, voltage or current. The electrical property of the sensor element 130 can change depending on the external atmosphere A, the currently applied temperature or other impact factors. A measurement of the electrical property may be corresponding for a certain exposed external atmosphere and a determined temperature of the sensor element 130. The measurement of the electrical property can be taken during the temperature sequence when temperature sequence applied is changing from a first temperature to a second temperature and / or at the corresponding temperature levels.

[0049] During step S3, the at least one electrical property measurement is processed using the at least one processing unit 110 to produce a data set. The data set is optionally saved to the at least one storage device 120. The data set comprises an electrical property value list with at least one electrical property value. The at least one electrical property value is representative of the at least one electrical property measurement at one of the multiple target temperature levels. Alternatively, the at least one electrical property value is representative of the at least one electrical property measurement during a transition between a pair of target temperature levels. Accordingly, while a temperature sequence is applied electrical property measurements are taken at or between the target temperature levels of the temperature sequence.

[0050] The data set can further comprise a change time value list with at least one change time value. The change time value is representative of the change time of the at least one electrical property measurement at one of the multiple target temperature levels. Further, the change time value can be representative of the change time of the at least one electrical property measurement during a transition between one pair of target temperature levels. Furthermore, the data set can comprise an electrical property change value list with at least one electrical property change value. The at least one electrical property change value is representative of the change of the electrical property value at one of the multiple target temperature levels and / or during a transition between one pair of target temperature levels.

[0051] Step S4 comprises the determination of a result based on the data set and a prediction data set. The prediction data set is preferably stored on the at least one storage device 120. However, the prediction data set can also be stored somewhere else. For example, if the at least one processing unit 110 comprises a memory, the prediction data set can also be stored there. Further, the prediction data set can also be stored at the at least one sensor element 130 if the at least one sensor element comprises a memory. The result comprises a state that can be fake, fake abnormal, normal or pre-alarm. During step S4 multiple results can be determined in parallel or consecutively.

[0052] In step S5 it is determined whether a hazard is present based on the result. If multiple results are present in a certain time sequence, the determination whether a hazard is present may be based on the multiple results. For example, if the results change from normal to pre-alarm, a hazard may be present.

[0053] Fig. 2b depicts another embodiment of a hazard detection system as shown in Fig. 2a. The hazard detection system according to Fig. 2b comprises additionally to the hazard detection system as shown in Fig. 2a a fire alarm system 230. At least one processing unit 210 may be similar to the at least one processing unit 1 10 of Fig. 2a. Furthermore, the at least one processing unit 210 may be configured to communicate 220 the result of the computer implemented method depicted in Fig. 1 to the fire alarm system 230. To utilize this, the at least one processing unit 210 may be able to access cellular resources or comprise a cellular model. Furthermore, the at least one processing unit 210 may comprise a network element such as a network interface that communicates the result to the fire alarm system. The communication may be wireless or wired.

[0054] Fig. 3 illustrates an alternative embodiment of a method for hazard detection according to an embodiment of the present invention. Steps S1 to S5 are the same as earlier described for Fig. 1 . Further, steps S6 to S11 are part of the method. Step S6 further comprises the collecting of a group of consecutive data sets. Step S7 further comprises selecting at least one value index of the electrical property value list and / or the change time value list and / or the electrical property change value list. Step S8 further comprises generating a value series comprising values of the electrical property value list and / or the change time value list and / or the electrical property change value list of each of the group of data sets, corresponding to the selected value index. Step S9 further comprises determining a change and / or a rate of change of the values of the value series.

[0055] In step S10 a classification of each value index based on comparing the change and / or rate of change of the values of the value series to at least one predetermined change threshold and / or rate of change threshold is done. Step S1 1 comprises modifying the data set to disregard or reduce weighting of individual values based on the classification of each value index.

[0056] Further steps S6 to S11 are executed to collect a group of data sets and to compare the group of data sets. The determination of a change and / or a rate of change of the values of the value series as defined in step S9 identifies characteristics of change rates for different data sets.

[0057] Fig. 4 illustrates a computer-implemented method for producing a prediction data set. As shown in Fig. 4, the method comprises step P1 that is providing individually external atmospheres representative of at least one known hazard and / or at least one known nonhazard. Further, according to step P2 a group of training data sets is produced by applying method steps (i) to (iii) of any of the former described method according to the invention with each of the provided external atmospheres. Step P3 comprises the training of a machine-learning algorithm with the group of training data sets and a group of indicators representative of the hazard state of the provided external atmospheres. Step P4 comprises generating a prediction data set using the machine-learning algorithm, for determining a result that is either fake abnormal, normal or pre-alarm based on a data set. The data set is preferably obtained by the computer-implemented method for hazard detection according to the invention.

[0058] Fig. 5 illustrates a schematic of a machine learning system 300 according to another aspect of the invention. The machine learning system 300 comprises a processing unit 310, a storage device 320 and an interface 350. The processing unit 310 executes the computer- implemented method for producing a prediction data set according to another aspect of the invention. In particular, the machine learning algorithm is executed by the processing unit 310. Interface 350 is provided to gather data sets produced by a computer-implemented hazard detection method according to another aspect of the invention. The interface may be a wired or wireless connection to communicate data sets to the machine-learning sys- tern 300. The data sets can be stored on the storage device 320. Further, results as determined by the machine learning algorithm may be stored on the storage device 320 and communicated to other devices as disclosed using interface 350.

[0059] List of reference signs

[0060] A external Atmosphere

[0061] P1 step

[0062] P2 step

[0063] P3 step

[0064] P4 step

[0065] 51 step

[0066] 52 step

[0067] 53 step

[0068] 54 step

[0069] 55 step

[0070] 56 step

[0071] 57 step

[0072] 58 step

[0073] 59 step

[0074] 510 step

[0075] 511 step

[0076] 110 Processing unit

[0077] 120 storage device

[0078] 130 sensor element

[0079] 140 heater

[0080] 200 System

[0081] 210 processing unit

[0082] 220 communication

[0083] 230 fire alarm system

[0084] 300 machine learning system

[0085] 310 processing unit

[0086] 320 storage device 350 interface

Claims

Claims1 . A computer-implemented hazard detection method using at least one processing unit(1 10) in signal communication with at least one storage device (120) and a gas-sensitive sensor element (130), wherein the sensor element (130) is configured to be in environmental contact with an external atmosphere, the method comprising:(i) applying a temperature sequence comprising multiple target temperature levels for a predetermined duration to the sensor element (130), by optionally using a heater (140),(ii) obtaining at least one electrical property measurement of the sensor element (130),(111) processing the at least one electrical property measurement using the at least one processing unit (110) to produce a data set, and optionally saving the data set to the at least one storage device (120), wherein the data set comprises: o an electrical property value list of at least one electrical property value, wherein the at least one electrical property value is representative of the at least one electrical property measurement at one of the multiple target temperature levels or during a transition between one pair of target temperature levels, and o a change time value list with at least one change time value, wherein the at least one change time value is representative of the change time of the at least one electrical property measurement at one of the multiple target temperature levels or during a transition between one pair of target temperature levels, and / or o an electrical property change value list with at least one electrical property change value, wherein the at least one electrical property change value is representative of the change of the electrical property value at one of the multiple target temperature levels or during a transition between one pair of target temperature levels,(iv) determining a result based on the data set and a prediction data set stored on the at least one storage device (120), wherein the result comprises fake abnormal, normal or pre-alarm, and(v) determining whether a hazard is present based on the result.

2. The computer-implemented method of claim 1 , wherein step (iii) of the method further comprises baseline correcting the electrical property value list and / or the change timevalue list and / or the electrical property change value list by detecting changes of values in each list and determining a baseline for each list based on the changes of values, preferably.

3. The computer-implemented method of any previous claim, wherein the at least one electrical property value and / or the at least one change time value and / or the at least one electrical property change value is / are modelled from the corresponding measurement using a model for modelling transient transitions, preferably a PT1 -model or a PT2-model.

4. The computer-implemented method of any previous claim, wherein the pre-alarm relates to a hazard, preferably a smoke hazard or a fire hazard.

5. The computer-implemented method of any previous claims, wherein the sensor element (130) comprises a sensing conductor and / or a metal-oxide semiconductor.

6. The computer-implemented method of any previous claims, wherein the external atmosphere is an atmosphere inside a separate environment, preferably a building.

7. The computer-implemented method of any previous claims, wherein the at least one electrical property measurement is obtained by applying an electrical signal with either constant voltage or constant current to the sensor element (130) and analyzing either current or voltage of the electrical signal using the at least one processing unit (1 10).

8. The computer-implemented method of any previous claims, wherein the temperature sequence comprises a step function of multiple step-wise constant, preferably adjacent, temperature levels and / or wherein the change time is the time required for the time course to transition from a first electrical property value attributed to a first target temperature level, to a second electrical property value attributed to a second target temperature level, wherein the change time preferentially is a rise and / or fall time.

9. The computer-implemented method of any previous claims, wherein the method further comprises:(vi) collecting a group of, preferably consecutive, data sets,(vii) selecting at least one value index of the electrical property value list and / or the change time value list and / or the electrical property change value list,(viii) generating a value series comprising values of the electrical property value list and / or the change time value list and / or the electrical property change value list of each of the group of data sets, corresponding to the selected value index,(ix) determining a change and / or a rate of change of the values of the value series,(x) classifying of each value index based on comparing the change and / or rate of change of the values of the value series to at least one predetermined change threshold and / or rate of change threshold,(xi) modifying the data set to disregard or reduce weighting of individual values based on the classification of each value index.

10. A hazard detection system (200), preferably a fire detection system, comprising at least one processing unit (210) in signal communication with at least one storage device (120) and a gas-sensitive sensor element (130), wherein the hazard detection system (200) is configured to execute the computer-implemented method of any of claims 1 to 9, wherein the at least one processing unit (210) preferentially is further configured to communicate (220) the result of the computer-implemented method to a fire alarm system (230).11 . The hazard detection system (200) according to claim 10, wherein the sensor element (130) comprises a semi-conductor, preferably a metal-oxide semiconductor, and wherein the hazard detection system (200) is configured to be in environmental contact with an external atmosphere, preferably an atmosphere inside a separate environment, such as a building.

12. A computer-implemented method for producing a prediction data set, preferably a prediction table, the method comprising:(a) providing individually external atmospheres representative of at least one known hazard and / or at least one known non-hazard, the hazard being a smoke hazard or a fire hazard,(b) producing a group of training data sets by applying method steps (i) to (iii) of any of the methods of claims 1 to 9 with each of the provided external atmospheres,(c) training a machine-learning algorithm with the group of training data sets and a group of indicators representative of the hazard state of the provided external atmospheres, and(d) generating a prediction data set using the machine-learning algorithm, for determining a result that is either fake abnormal, normal or pre-alarm based on a data set, wherein the data set is preferably obtained by any method pursuant to claims 1 to 9.

13. The computer-implemented method of claim 12, further comprising the following step:(e) saving the generated prediction data set to the at least one storage device (120).

14. The computer-implemented method of claims 12 or 13, wherein the prediction data set is produced by applying method steps (i) to (iii) multiple times individually with external atmospheres specific to different hazards and / or non-hazards to train the machine-learning algorithm.

15. A machine learning computing system (300) configured to execute the computer-implemented method according to any of claims 12 to 14, wherein the machine learning computing system (300) preferentially comprises at least one processing unit (310), at least one storage device (320) and at least one interface (350) to access data sets produced by a method according to claims 1 to 9.

Images