Catalytic combustion type combustible gas sensor and method thereof

Abstract

The invention discloses a catalytic combustion type combustible gas sensor and a method thereof, and the method comprises the steps: continuously monitoring a baseline of analog signal voltage, timely capturing a dynamic response sequence, and carrying out the dynamic fingerprint feature vector extraction of the response sequence, so as to accurately extract multi-dimensional features including steady-state voltage; based on the extracted dynamic feature vector, fine classification is carried out on the health state of the sensor, and a state probability vector with confidence is output; on this basis, a self-adaptive concentration resolving mechanism based on state probability vector weighted fusion is introduced, the steady-state voltage is corrected, and the uncertainty score is synchronously quantified and resolved, so that the output of the corrected concentration and the corresponding alarm are realized. In this way, the continuity and reliability of the concentration measurement result are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent alarm, and more particularly, to a catalytic combustion combustible gas sensor and method thereof. BACKGROUND

[0002] Combustible gas is ubiquitous in industrial production, civil facilities, and mine environments, and its leakage can lead to serious safety accidents such as fire and explosion, and even endanger life safety. Therefore, it is crucial to continuously, accurately and reliably monitor combustible gas in the environment. However, as the sensor is used for a long time, the internal catalyst will gradually age and be poisoned, leading to performance degradation, which not only affects the sensitivity and response speed of the sensor, but also can cause the measurement result to be distorted, thereby causing unpredictable risks. Therefore, an intelligent sensor that can self-diagnose at each power-on self-test or regular work is an inevitable demand for the industry development.

[0003] Existing catalytic combustion combustible gas sensors still face many challenges in practical application. First, the traditional detection method generally simplifies the output signal of the sensor as a single steady-state direct current voltage. This static single-variable processing method ignores the complex physical and chemical reaction process and its dynamic changes between the gas and the sensor catalyst. If only the steady-state voltage is used for judgment, it may underestimate the gas concentration, especially in the case of sensor aging or poisoning, which can lead to serious consequences, because important dynamic information such as changes in response time is completely ignored, so that the real health status of the sensor cannot be accurately diagnosed, and real-time self-diagnosis cannot be achieved without using standard gas for manual calibration.

[0004] Therefore, an optimized catalytic combustion combustible gas sensor is expected. SUMMARY

[0005] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a catalytic combustion combustible gas sensor and method thereof, which continuously monitors the baseline of the analog signal voltage and timely captures the dynamic response sequence, and extracts the kinetic fingerprint feature vector from the response sequence to accurately extract multi-dimensional features including steady-state voltage; then based on the extracted kinetic feature vector, the sensor health status is finely classified, and a state probability vector with confidence is output; on this basis, an adaptive concentration solving mechanism based on state probability vector weighted fusion is introduced, the steady-state voltage is corrected, and the uncertainty score is quantitatively solved at the same time, so as to realize the output of the corrected concentration and the corresponding alarm. In this way, the continuity and reliability of the concentration measurement result are significantly improved.

[0006] According to one aspect of the present application, there is provided a method of catalytic combustion combustible gas sensor, comprising: continuously monitoring a baseline value of the analog signal voltage and determining whether to trigger dynamic response capture; capturing a response sequence in response to the dynamic response capture; performing kinetic fingerprint feature vector extraction on the response sequence to obtain a kinetic feature vector and a steady-state voltage; performing sensor health state classification based on the kinetic feature vector to obtain a sensor health state and a state probability vector; performing state-adaptive concentration calculation on the steady-state voltage based on the sensor health state and the state probability vector to obtain a corrected concentration and a calculation uncertainty score; performing decision output and alarm based on the corrected concentration and the calculation uncertainty score to obtain a final concentration display value and a gas overrun alarm.

[0007] According to another aspect of the present application, there is provided a catalytic combustion combustible gas sensor, comprising: a dynamic response capture monitoring module for continuously monitoring a baseline value of the analog signal voltage and determining whether to trigger dynamic response capture; a response sequence capturing module for capturing a response sequence in response to the dynamic response capture; a kinetic fingerprint feature vector extraction module for performing kinetic fingerprint feature vector extraction on the response sequence to obtain a kinetic feature vector and a steady-state voltage; a sensor health state classification module for performing sensor health state classification based on the kinetic feature vector to obtain a sensor health state and a state probability vector; a state-adaptive concentration calculation module for performing state-adaptive concentration calculation on the steady-state voltage based on the sensor health state and the state probability vector to obtain a corrected concentration and a calculation uncertainty score; a decision output and alarm module for performing decision output and alarm based on the corrected concentration and the calculation uncertainty score to obtain a final concentration display value and a gas overrun alarm.

[0008] Compared with the prior art, the catalytic combustion combustible gas sensor and the method thereof provided by the application can accurately extract multi-dimensional features including steady-state voltage by continuously monitoring the baseline of an analog signal voltage and timely capturing a dynamic response sequence, and extracting a kinetic fingerprint feature vector from the response sequence; then, based on the extracted kinetic feature vector, the sensor health state is finely classified, and a state probability vector with confidence is output; on this basis, an adaptive concentration calculation mechanism based on state probability vector weighted fusion is introduced to correct the steady-state voltage and simultaneously quantify and calculate the uncertainty score, so that the output of the corrected concentration and the corresponding alarm are realized. In this way, the continuity and reliability of the concentration measurement result are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0009] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application taken in conjunction with the accompanying drawings. The drawings provided in the present application are used to provide further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 Flow chart of the method of the catalytic combustion combustible gas sensor according to the embodiments of the present application; Figure 2 Data flow schematic diagram of the method of the catalytic combustion combustible gas sensor according to the embodiments of the present application; Figure 3 Block diagram of the catalytic combustion combustible gas sensor according to the embodiments of the present application. DETAILED DESCRIPTION

[0011] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0012] As shown in the present application and claims, unless the context clearly indicates otherwise, "one", "a", "an", and / or "the" do not refer to the singular, but can also include the plural. Generally, the terms "comprise" and "include" only indicate the inclusion of the steps and elements explicitly identified, and these steps and elements do not constitute an exclusive list, and the method or device can also include other steps or elements.

[0013] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are merely illustrative and different aspects of the system and method can use different modules.

[0014] Flowcharts are used in the present application to illustrate the operations performed by the system according to embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in sequence. Instead, various steps can be processed in reverse order or simultaneously, as desired. Other operations can also be added to or removed from these processes, or one or more steps can be removed from these processes.

[0015] In the following, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It is apparent that the described embodiments are only a part of the embodiments of the present application, and not all embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.

[0016] In the technical solutions of the present application, a method for a catalytic combustion combustible gas sensor is proposed. Figure 1 Flowchart of the method for the catalytic combustion combustible gas sensor according to embodiments of the present application. Figure 2 System architecture diagram of the method for the catalytic combustion combustible gas sensor according to embodiments of the present application. As shown in Figure 1 and Figure 2 As shown, the method for the catalytic combustion combustible gas sensor according to embodiments of the present application includes the steps of: S1, continuously monitoring the baseline value of the analog signal voltage and determining whether to trigger dynamic response capture; S2, in response to dynamic response capture, collecting a response sequence; S3, performing kinetic fingerprint feature vector extraction on the response sequence to obtain a kinetic feature vector and a steady-state voltage; S4, performing sensor health state classification based on the kinetic feature vector to obtain a sensor health state and a state probability vector; S5, based on the sensor health state and the state probability vector, performing state-adaptive concentration calculation on the steady-state voltage to obtain a corrected concentration and a calculation uncertainty score; S6, based on the corrected concentration and the calculation uncertainty score, performing decision output and alarm to obtain a final concentration display value and a gas overrun alarm.

[0017] In particular, the S1 continuously monitors the baseline value of the analog signal voltage and determines whether to trigger a dynamic response capture. It should be understood that in the field of combustible gas monitoring, the baseline value of the analog signal of the sensor is not constant, and it is easily affected by environmental temperature, humidity fluctuations, and the aging and drift of the sensor itself. If the baseline value is not continuously monitored and dynamically adjusted, the sensor may also produce slight signal fluctuations in the absence of gas leakage. These fluctuations, once misjudged as gas response, will lead to false alarms, thereby reducing the credibility of the system and the trust of users in the system; on the contrary, if the baseline drift causes the real gas response signal to be submerged, it may cause a false negative, which may cause serious safety hazards. Therefore, accurately and in real time identifying the difference between the signal change caused by the real gas and the background noise or baseline drift is a key prerequisite for ensuring the effective capture of subsequent response sequences, avoiding resource waste, and ensuring system safety.

[0018] wherein the baseline value of the analog signal voltage refers to the stable output voltage reference point of the catalytic combustion type combustible gas sensor in the absence of target gas. This value is not fixed, but is dynamically changed by the environment and the state of the sensor itself; the trigger threshold is a preset voltage difference parameter used to determine how much the analog signal voltage rises relative to the baseline value to be considered a potential gas response, in order to distinguish between real events and background noise.

[0019] In specific implementation, first, the time series of the analog signal voltage is dynamically baseline voltage self-adaptive estimation to obtain the baseline voltage estimation value. This means that the analog signal voltage output by the sensor is not considered fixed, but is treated as a sequence that changes over time. By using advanced adaptive estimation algorithm, the system can track and estimate the baseline voltage in real time and accurately. The estimation value can intelligently adapt to the slow drift or long-term performance change of the sensor in the absence of gas environment, thereby providing a stable and accurate real-time reference point for subsequent event judgment, ensuring that the system can always distinguish between normal background fluctuations and potential gas events; Secondly, it is determined whether the value of the continuous rising counter of the real-time analog signal voltage greater than the sum of the baseline voltage estimation value and the trigger threshold is greater than a preset duration period, and if so, it is confirmed that the dynamic response capture is triggered, wherein the preset duration period is 3. Specifically, the system continuously acquires the real-time analog signal voltage and compares it with the sum of the dynamic baseline voltage estimation value and a preset trigger threshold. Only when the real-time analog signal voltage is greater than the sum of the baseline voltage estimation value and the trigger threshold, the internal continuous rising counter starts to increment; in order to effectively filter out the false trigger caused by transient noise or occasional small fluctuations, the system further requires that the value of the continuous rising counter must be greater than a preset duration period; wherein the preset duration period is 3. Only when the signal continuously meets the rising condition and reaches or exceeds the preset period, the system will finally confirm that the dynamic response capture is triggered, thereby starting the accurate acquisition of the sensor response sequence. This multiple confirmation mechanism significantly improves the accuracy and robustness of event trigger judgment.

[0020] In particular, the S2 acquires the response sequence in response to the dynamic response capture. It should be understood that when the catalytic combustion combustible gas sensor contacts the combustible gas, the output voltage does not instantaneously reach a steady state, but undergoes a dynamic change process from the baseline, gradually rising, and finally tending to be stable, i.e. the rising process. The voltage time sequence of this rising process contains rich catalytic reaction kinetic information, which is crucial for accurately determining the true health status of the sensor (such as healthy, aging or poisoning). For example, a healthy sensor has high catalyst activity, rapid reaction, steep rising curve and short response time; while an aging or poisoned sensor has low catalyst activity, slow reaction, flat rising curve and significantly longer response time. Simply reading the final steady-state value, as in the prior art, will lose these valuable dynamic information, making it impossible to distinguish between real low-concentration gas and low output after sensor failure, thereby seriously affecting the reliability of the detection result. Therefore, immediately after the dynamic response capture is triggered, the complete response sequence is acquired, which is the fundamental prerequisite for obtaining these key kinetic fingerprint information, realizing intelligent diagnosis of the sensor and accurate concentration calculation.

[0021] In implementation, first, the baseline voltage estimate at the time of dynamic response capture triggering is obtained as the trigger-time baseline. In the foregoing step, the system has already obtained the real-time baseline voltage value through dynamic baseline voltage adaptive estimation. When the triggering condition is met and dynamic response capture is confirmed, the system accurately records the baseline voltage estimate at that time; this value will be used as a reference for subsequent processing to eliminate the influence of environmental background signals on the original response sequence and ensure that the focus of subsequent analysis is on the real signal change caused by the gas. Here, the trigger-time baseline specifically refers to the background baseline value of the sensor analog signal voltage estimated by the system at the time point when dynamic response capture is triggered. It is the reference zero point for the difference calculation of the subsequent original response sequence; Next, the original response sequence is obtained. That is, after dynamic response capture is triggered, the sensor begins to output its real-time analog signal voltage in response to the gas. The system will continuously collect these original analog signal voltage values at a preset sampling frequency and duration, forming a time sequence, i.e., the original response sequence, which records the complete voltage change trajectory of the sensor from the start to the steady state under gas stimulation; Further, the difference between each original response in the original response sequence and the trigger-time baseline is calculated to obtain the response sequence. Specifically, for each sampling point in the original response sequence, the system subtracts the baseline voltage estimate determined at the time of dynamic response capture triggering. Through this difference calculation, the baseline drift component in the original signal is effectively removed, resulting in a response sequence that is zero-based and purely reflects the change in gas concentration. Specifically, the response sequence refers to the continuous data set of the analog signal voltage output by the sensor over time after dynamic response capture triggering. It contains complete process information of the sensor from the start of the response to the steady state and is the direct input for subsequent dynamic feature extraction and analysis.

[0022] In particular, the S3 extracts the dynamic fingerprint vector from the response sequence to obtain the dynamic feature vector and the steady-state voltage. It should be understood that when the catalytic combustion combustible gas sensor contacts the target gas, its output voltage does not jump instantaneously, but undergoes a process of continuous rise and finally reaches a steady state. During the rising process, the steepness of the rise, the time required to reach the steady state, and the intermediate form contain rich catalytic reaction dynamic information. These dynamic information is the key fingerprint for diagnosing the health status of the sensor. Specifically, the catalyst activity of a healthy sensor is high, and the reaction is rapid, which is manifested as a steep rising curve and a short response time; while the aged or poisoned sensor, due to the decrease of catalyst activity, the reaction will become slow, which is manifested as a flat rising curve and a significantly longer response time. If only the final steady-state value is read, as in the defects existing in the prior art, these valuable dynamic characteristics will be lost, resulting in the inability to accurately distinguish whether the sensor is in a healthy, aged or poisoned state, and also unable to effectively solve the confusion between real low concentration and invalid low output. Therefore, capturing the voltage time sequence of the entire rising process and extracting representative dynamic features from it to form a unique dynamic fingerprint is the fundamental way to realize sensor self-diagnosis, improve detection reliability and accuracy.

[0023] In specific implementation, first, the steady-state response feature is extracted from the response sequence to obtain the steady-state voltage. It should be understood that in the working principle of the catalytic combustion combustible gas sensor, when the combustible gas contacts the sensitive element of the sensor and a catalytic combustion reaction occurs, heat will be generated, which will change the resistance of the sensitive element, and finally reflected in the change of the analog signal voltage. After a period of reaction, when the gas concentration reaches equilibrium and the sensor output tends to be stable, the obtained voltage value is the steady-state voltage. This steady-state voltage directly corresponds to the concentration of combustible gas in the environment and is the basis for quantitative analysis. Although the present application emphasizes the use of dynamic information for sensor health status diagnosis, the steady-state voltage as the most intuitive and direct quantitative indicator of gas concentration, its accurate extraction is still crucial for accurate concentration calculation. If the steady-state voltage extraction is not accurate, it will directly lead to the deviation of the final concentration display value, thereby affecting the reliability and safety of the entire gas detection system.

[0024] In particular implementation, in order to extract the steady-state voltage, the system needs to identify the end part of the response sequence that has reached the steady state. A commonly used method is to first determine the duration of the response sequence or the total number of sampling points. Then, the last time window (e.g., the last 10% or 20% of sampling points) of the response sequence is selected, and it is considered that the signal fluctuation in this time period is very small and can represent the steady state. For all voltage values in the selected steady-state window, the system calculates their average, median or other statistical processing to obtain a representative steady-state voltage. This method can effectively smooth the small noise that the sensor may have in the steady-state stage, and provide a stable and reliable concentration corresponding value; Next, the time dynamic feature extraction is performed on the response sequence to obtain the 90% response time. It can be understood that when the catalytic combustion type combustible gas sensor is exposed to the target gas, its output voltage does not reach the final steady state instantaneously, but experiences a dynamic process of gradually rising from the baseline to the steady state. The speed of this rising process, i.e., the response time, contains rich catalytic reaction kinetics information and is the core basis for judging the health status of the sensor. For example, a healthy sensor has high catalyst activity and rapid reaction, and its response time is usually short; when the sensor is aged or poisoned due to long-term use, the catalyst activity will be significantly reduced, resulting in slow reaction and significantly longer response time. If only the steady-state voltage is concerned and the response time is ignored, it will not be possible to distinguish whether the sensor is in a normal working state or a performance degradation state, which directly affects the timeliness and accuracy of gas leakage warning. Therefore, accurately extracting the 90% response time (T90) as a time dynamic feature is an indispensable link to realize intelligent diagnosis of the sensor and ensure safe and reliable operation of the system.

[0025] In particular implementation, the time required for the signal to reach 90% of its steady-state value needs to be determined from the response sequence that has been collected and corrected for baseline. Specifically, based on the obtained steady-state voltage, the system calculates 90% of its value, i.e., the target voltage. This target voltage is the key threshold for determining the 90% response time; during the traversal process, the system identifies the first time point in the response sequence that reaches or exceeds the target voltage; and the final 90% response time is calculated by subtracting the starting time of the response from the time point that exceeds the target voltage. This value represents the time required for the sensor to respond from the start to reach most of the steady state, and is a direct quantitative indicator of its response speed; Further, the rate dynamic feature of the response sequence is extracted to obtain the maximum slope. It can be understood that when the catalytic combustion combustible gas sensor contacts the target gas, the output voltage thereof will experience a rapid rising process from the baseline. The rate of this rising process, especially the slope of the steepest part thereof, directly reflects the activity of the catalyst and the reaction efficiency between the gas and the sensitive element. For example, the catalyst activity of a healthy sensor is high, and the response to the gas is rapid and intense, so the rising slope of the response curve thereof is usually large; when the sensor is aged or poisoned due to long-term use, the catalyst activity decreases, leading to slow reaction, and the rising slope of the response curve thereof will be significantly reduced. Relying only on a single or limited dynamic feature such as the steady-state voltage or the 90% response time may not be sufficient to fully capture the subtle changes in sensor performance degradation, especially when the sensor performance starts to degrade but has not yet seriously affected the steady-state output. Therefore, extracting the maximum slope of the response sequence can accurately quantify the fastest change rate in the sensor response process, providing a sensitive and independent indicator for diagnosing the health status of the sensor, thereby enriching the dimension of the kinetic fingerprint and enhancing the accuracy and reliability of the self-diagnosis of the sensor.

[0026] In specific implementation, the rate analysis is performed on the collected and baseline-corrected response sequence to determine the point at which the voltage changes fastest during the rising process. Specifically, in order to obtain the rate dynamic feature of the response sequence, the system needs to perform numerical differentiation on the sequence to calculate the instantaneous slope at different time points. For discrete sampled data, the instantaneous slope can be approximated by calculating the ratio of the voltage change between adjacent sampling points to the time interval; after obtaining all the instantaneous slope values, the system will traverse the slope sequence and find the maximum positive slope value from it, which represents the moment at which the sensor output voltage rises fastest during the entire response process, accurately capturing the most intense moment of the catalytic reaction; Then, based on the 90% response time, the response sequence is subjected to response morphology feature extraction to obtain the early-stage integral area. After contacting the combustible gas, the output voltage rising process of the catalytic combustion combustible gas sensor not only reflects the response speed (such as the 90% response time) and the maximum change rate (such as the maximum slope), but also reflects the overall morphology of the response curve. Different sensor health states, even in some cases may have similar 90% response times, but the cumulative signal amount or rising morphology of the initial stage of the response curve may still have significant differences. For example, a slightly aged sensor may have a 90% response time close to that of a healthy sensor, but its initial stage response may be more gentle, resulting in different cumulative signal amounts (integral areas) in the early stage. Traditionally, if only relying on the steady-state value, the 90% response time or the maximum slope, these subtle but important morphological differences may not be captured, thereby affecting the accuracy of the sensor health state classification. By extracting the early-stage integral area of the response sequence, the signal accumulation amount in the initial stage of the response (from the start of the response to the 90% steady-state value) can be quantified, thereby providing a complementary and recognizable morphology feature. In this way, the health state of the sensor can be more accurately diagnosed. Specifically, the 90% response time obtained previously is used as the upper limit of integration, and the baseline-corrected response sequence is subjected to integration operation to obtain the early-stage integral area; Further, the steady-state voltage, the 90% response time, the maximum slope and the early-stage integral area are subjected to kinetic feature vector aggregation to obtain the kinetic feature vector. It can be understood that only the steady-state voltage may not be able to distinguish between true low concentration and failed low output; only the 90% response time may not be able to distinguish the initial stage morphology difference in the response process. The health state of the sensor needs to consider all the key dynamic characteristics exhibited by the sensor in the gas response process. Aggregating these complementary features that describe the dynamic behavior of the sensor from different angles can form a more comprehensive and accurate kinetic fingerprint. This multi-dimensional vector can more effectively reflect the changes of the core parameters such as the catalyst activity and the reaction kinetics inside the sensor, thereby significantly improving the accuracy and robustness of the subsequent machine learning model (such as SVM, decision tree) for classifying the health state of the sensor. Specifically, the numerical features obtained by independently calculating the foregoing sub-steps are combined into a unified vector in a predetermined order to obtain the early-stage integral area.

[0027] In particular, the S4 performs sensor health state classification based on the kinetic feature vector to obtain a sensor health state and a state probability vector. It should be understood that catalytic combustible gas sensors are susceptible to aging, poisoning and other factors during long-term operation, and performance degradation will seriously endanger the reliability of the gas monitoring system if not diagnosed in time and accurately. Traditional diagnostic methods cannot accurately evaluate the sensor health state in real time without interrupting the monitoring and using standard gas. In the technical solution of the present application, the kinetic feature vector is used to comprehensively capture the transient kinetic characteristics of the sensor gas response, thereby realizing intelligent identification and classification of the health state and providing reliable prior information for subsequent concentration calculation.

[0028] In specific implementation, the polymerized kinetic feature vector is input into a pre-trained machine learning classification model. Based on the feature vector, the model outputs the sensor health state of the sensor, i.e. the most likely health state category (such as "healthy", "mild aging", etc.). At the same time, it will generate a state probability vector to quantify the confidence of the sensor belonging to each possible health state. This probability output mechanism provides more information than a single label, avoids the rigid switching problem of traditional classifiers in the critical state region, and provides continuous and robust prior information for subsequent state adaptive concentration calculation.

[0029] In particular, the S5 performs a state-adaptive concentration calculation on the steady-state voltage based on the sensor health state and the state probability vector to obtain a corrected concentration and a calculation uncertainty score. It should be understood that the existing state-adaptive concentration calculation mechanism has a core bottleneck of rigid switching of processing logic and one-sided utilization of information. The traditional mechanism often selects a single calculation model from a pre-set model library based on the most likely state label given by the classifier. This single optimal model selection strategy ignores the probabilistic nature of the sensor health state classification. For example, when the dynamics of the sensor make the classification confidence of the health state and the slight aging state extremely close, any slight signal fluctuation can cause the classification result to flip, thereby triggering the calculation model to switch from one to another, and ultimately causing the output concentration value to produce a stepwise mutation that does not conform to the physical law, seriously damaging the continuity and reliability of the measurement result. More deeply, the mechanism completely discards the complete state probability distribution information output by the classifier. A health state with a confidence of 95% and a health state with a confidence of 55% are treated equally in the mechanism, but the uncertainty behind them is completely different. This waste of information makes the concentration calculation result particularly vulnerable when facing classification uncertainty. The root cause is the failure to recognize and utilize the inherent correlation between the selection weight of the calculation model and the complete probability distribution of the sensor state. An ideal calculation mechanism should regard the final concentration as an expected value based on all possible health states, rather than relying on only the most likely discrete state. To overcome the above technical defects, the present application introduces a multi-model probability fusion and uncertainty quantification mechanism to completely eliminate the discontinuity of the measurement value caused by the hard switching of the state classification, ensure the smoothness and stability of the output signal, and significantly improve the accuracy and robustness of the concentration calculation result when there is uncertainty in the sensor state.

[0030] In specific implementation, first, the steady-state voltage is input in parallel to all N health state calculation models in the concentration model library to obtain a candidate concentration vector. It should be understood that, in order to avoid rigid single model selection, the present application takes all potential possibilities into account. Specifically, the steady-state voltage is input in parallel to all N calculation models corresponding to different health states stored in the model library, each model is independently calculated, thereby generating a candidate concentration vector containing all possible concentration results. That is, it creatively converts the calculation problem from a discrete choice of this or that to a continuous spectrum problem containing multiple possibilities, which provides basic data support for subsequent smooth fusion. Specifically, the process is expressed by formula as follows: , wherein, represents the candidate concentration vector; is the candidate concentration value calculated by the i th model; is the concentration estimation model corresponding to the i-th health state in the model library; is the input steady-state response voltage; N is the preset total number of health states; Then, based on the state probability vector, a classification probability-based weighted fusion estimation is performed on the candidate concentration vector to obtain the corrected concentration. That is, the state probability vector output by the upstream classifier is used as a weight to perform a weighted summation on the candidate concentration vector generated in the previous step, thereby obtaining a unique final concentration value that integrates all state possibilities. This borrows the idea of model fusion and uses the classification probability as a confidence level to perform a weighted average on the physical quantity. In this way, when the sensor state is smoothly transitioning between different categories, the probability weight will also change smoothly, thereby ensuring that the final output concentration value can achieve seamless and continuous transition, completely solving the step change problem of the original mechanism; at the same time, by fully utilizing the complete probability distribution information, the robustness and reliability of the concentration estimation result when the classifier has uncertainty are greatly enhanced. Specifically, this process is expressed by the formula as follows: , wherein, represents the corrected concentration; is the probability that the sensor is in the i-th health state; is the state probability vector composed of all . Further, the corrected concentration is quantified for uncertainty to obtain an estimation uncertainty score. It should be understood that a complete measurement result should not only include a numerical value, but also include its reliability evaluation. Specifically, the state probability vector is calculated using the normalized Shannon entropy in information theory to obtain a score quantifying the uncertainty of the current estimation result. In this way, a new data quality dimension is creatively added to the traditional sensor measurement value, and an uncertainty score between 0 and 1 is finally output. A score close to 0 indicates that the state classification is very clear and the estimation result is highly reliable; a score close to 1 indicates that the state classification is highly ambiguous and the estimation result, although the current optimal estimate, has a very high inherent uncertainty. This score provides unprecedented rich decision-making basis for the downstream decision-making and maintenance system, realizing the transformation of the sensor from a simple data source to an intelligent information node. Specifically, this process is expressed by the formula as follows: , wherein, represents the estimation uncertainty score; represents the calculation of the normalized Shannon entropy on the probability vector .

[0031] It is worth mentioning that by introducing a solution method based on multi-model probability fusion, the discontinuity of the measurement value caused by the hard switching of state classification is completely eliminated, ensuring the smoothness and stability of the output signal. Specifically, this method makes full use of the complete probability distribution information provided by the upstream classifier, significantly improving the accuracy and robustness of the concentration solution result when there is uncertainty in the sensor state. More importantly, by creatively introducing an uncertainty quantification index, the sensor has the ability to self-evaluate the quality of its output data. This not only provides the end user with double information containing numerical value and reliability, but also provides a solid data foundation for the implementation of predictive maintenance and intelligent alarm strategies, the ultimate goal of which is to upgrade the sensor from a passive measurement tool to an intelligent sensing node that can actively evaluate its own state and information quality, thereby fundamentally improving the reliability and intelligence level of the entire gas safety monitoring system.

[0032] In particular, the S6, based on the corrected concentration and the solution uncertainty score, makes a decision output and alarm to obtain the final concentration display value and gas overrun alarm. It should be understood that in the actual application of combustible gas monitoring, it is far from enough to provide only a concentration value. The performance of the sensor will decay over time, and the accuracy of the measurement result may fluctuate due to the uncertainty of the sensor health state. Traditional alarm systems often only make judgments based on a single concentration threshold, ignoring the reliability or confidence of the measurement result, which may result in false alarms or missed alarms when the sensor state is ambiguous. In the technical solution of the present application, by introducing the solution uncertainty score, the sensor has the ability to self-evaluate the quality of its output data. This not only provides the end user with double information containing numerical value and reliability, greatly enhancing the transparency and reliability of the measurement result, but more importantly, it provides a solid data foundation for the implementation of more refined and intelligent decision output and predictive maintenance and intelligent alarm strategies. By considering the corrected concentration and its corresponding solution uncertainty, the system can upgrade from a passive measurement tool to an intelligent sensing node that can actively evaluate its own state and information quality, thereby fundamentally improving the reliability and intelligence level of the entire gas safety monitoring system, ensuring that the most accurate and reliable judgment and response can be made at critical moments.

[0033] In specific implementation, first, the corrected concentration is taken as the final concentration display value, which is presented to the operator in real time through the user interface (such as a digital display screen, a SCADA system interface, etc.). This display value is the result after multi-model probability fusion and state adaptive correction, which can maintain better continuity and accuracy when the sensor health state changes compared to traditional methods; Then, the system continuously compares the final concentration display value with preset gas overrun alarm thresholds. Generally, these thresholds include a low alarm threshold and a high alarm threshold. Once the final concentration display value exceeds any of the preset alarm thresholds, the system triggers a corresponding gas overrun alarm; Further, the solution uncertainty score is used to enhance the intelligence of decision-making and the robustness of alarm strategy. This score provides a quantitative index between 0 and 1, reflecting the ambiguity of the current concentration solution result. Specifically, when the solution uncertainty score is low (close to 0), it indicates that the sensor health state classification is clear, and the solution result is highly reliable. At this time, the system can strictly follow the preset concentration threshold for alarm, and may directly trigger a regular level alarm (such as audible and visual alarm, sending a short message notification, etc.) when the alarm threshold is reached; when the solution uncertainty score is high (close to 1), it indicates that the sensor health state classification is highly ambiguous, and although the solution result is the current optimal estimate, its inherent uncertainty is extremely high. In this case, the system can adopt a more cautious strategy: lower the alarm threshold to some extent, so that it is easier to trigger an alarm at the same concentration, to improve safety; upgrade the alarm level: even if the concentration just exceeds the regular alarm threshold, if accompanied by high uncertainty, the system can trigger a higher level alarm (for example, in addition to audible and visual alarm, it also forces higher authority managers to be notified, starts emergency video monitoring, etc.); trigger maintenance or calibration suggestions: a high uncertainty score can be directly used as the basis for predictive maintenance, prompting the operator or maintenance system to check, calibrate or replace the sensor, even if the current concentration has not reached a dangerous level; provide additional information: when displaying the final concentration, the current solution uncertainty score is also displayed or informed to the operator in other ways, so that they have more comprehensive information support when making manual judgments.

[0034] In summary, the method of the catalytic combustion combustible gas sensor according to the embodiments of the present application is illustrated, which continuously monitors the baseline of the analog signal voltage and timely captures the dynamic response sequence, and extracts the kinetic fingerprint feature vector from the response sequence, to accurately extract multi-dimensional features including steady-state voltage; further, based on the extracted kinetic feature vector, the sensor health state is finely classified, and a state probability vector with confidence is output; on this basis, an adaptive concentration solution mechanism based on state probability vector weighted fusion is introduced, the steady-state voltage is corrected, and the solution uncertainty score is quantified simultaneously, so as to realize the output of the corrected concentration and the corresponding alarm. In this way, the continuity and reliability of the concentration measurement result are significantly improved.

[0035] Further, a catalytic combustion combustible gas sensor is also provided.

[0036] Figure 3This is a block diagram of a catalytic combustion combustible gas sensor according to an embodiment of this application. Figure 3 As shown, the catalytic combustion combustible gas sensor 300 according to an embodiment of this application includes: a dynamic response capture monitoring module 310, used to continuously monitor the baseline value of the analog signal voltage and determine whether dynamic response capture is triggered; a response sequence acquisition module 320, used to acquire a response sequence in response to dynamic response capture; a kinetic fingerprint feature vector extraction module 330, used to extract a kinetic fingerprint feature vector from the response sequence to obtain a kinetic feature vector and a steady-state voltage; a sensor health status classification module 340, used to classify the sensor health status based on the kinetic feature vector to obtain a sensor health status and a state probability vector; a state-adaptive concentration calculation module 350, used to perform state-adaptive concentration calculation on the steady-state voltage based on the sensor health status and the state probability vector to obtain a corrected concentration and a calculation uncertainty score; and a decision output and alarm module 360, used to perform decision output and alarm based on the corrected concentration and the calculation uncertainty score to obtain a final concentration display value and a gas over-limit alarm.

[0037] As described above, the catalytic combustion combustible gas sensor 300 according to the embodiments of this application can be implemented in various wireless terminals, such as servers with algorithms for catalytic combustion combustible gas sensors. In one possible implementation, the catalytic combustion combustible gas sensor 300 according to the embodiments of this application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the catalytic combustion combustible gas sensor 300 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the catalytic combustion combustible gas sensor 300 can also be one of many hardware modules of the wireless terminal.

[0038] Alternatively, in another example, the catalytic combustion combustible gas sensor 300 and the wireless terminal can also be separate devices, and the catalytic combustion combustible gas sensor 300 can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0039] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for a catalytic combustion type combustible gas sensor, characterized in that, include: Continuously monitor the baseline value of the analog signal voltage and determine whether dynamic response capture is triggered; In response to dynamic response capture, the response sequence is collected; Dynamic fingerprint feature vector extraction is performed on the response sequence to obtain dynamic feature vector and steady-state voltage; Sensor health status is classified based on dynamic feature vectors to obtain sensor health status and state probability vectors; Based on the sensor's health status and state probability vector, a state-adaptive concentration solution is performed on the steady-state voltage to obtain the corrected concentration and the solution uncertainty score. The decision output and alarm are based on the corrected concentration and the solution uncertainty score to obtain the final concentration display value and gas over-limit alarm.

2. The method for a catalytic combustion combustible gas sensor according to claim 1, characterized in that, Continuously monitor the baseline value of the analog signal voltage and determine whether dynamic response capture is triggered, including: Dynamic baseline voltage adaptive estimation is performed on the time series of analog signal voltage to obtain the baseline voltage estimate; Determine whether the value of the continuously rising counter, which indicates that the real-time analog signal voltage is greater than the sum of the baseline voltage estimate and the trigger threshold, is greater than the preset duration period. If so, confirm the triggering of dynamic response capture, where the preset duration period is 3.

3. The method for a catalytic combustion combustible gas sensor according to claim 2, characterized in that, In response to dynamic response capture, the response sequence is acquired, including: The baseline voltage estimate at the time of dynamic response capture trigger is used as the baseline at the trigger moment; Obtain the original response sequence; The difference between each original response in the original response sequence and the baseline at the trigger time is calculated to obtain the response sequence.

4. The method for a catalytic combustion combustible gas sensor according to claim 1, characterized in that, The dynamic fingerprint feature vector of the response sequence is extracted to obtain the dynamic feature vector and steady-state voltage, including: Steady-state response features are extracted from the response sequence to obtain the steady-state voltage; Temporal dynamic features are extracted from the response sequence to obtain 90% of the response time; Rate dynamic feature extraction is performed on the response sequence to obtain the maximum slope; Based on the 90% response time, response morphology features are extracted from the response sequence to obtain the initial integral area; The dynamic eigenvectors are obtained by aggregating the steady-state voltage, 90% response time, maximum slope, and early integral area.

5. The method for a catalytic combustion combustible gas sensor according to claim 1, characterized in that, Based on the sensor's health state and state probability vector, a state-adaptive concentration calculation is performed on the steady-state voltage to obtain the corrected concentration, including: The steady-state voltage is input in parallel into all N healthy state solution models in the concentration model library to obtain candidate concentration vectors; Based on the state probability vector, the candidate concentration vector is weighted and fused according to the classification probability to obtain the corrected concentration. Uncertainty quantification is performed on the corrected concentration to obtain the solution uncertainty score.

6. The method for a catalytic combustion combustible gas sensor according to claim 5, characterized in that, Based on the state probability vector, a weighted fusion calculation based on classification probability is performed on the candidate concentration vector to obtain the corrected concentration. This includes: based on the state probability vector, a weighted fusion calculation based on classification probability is performed on the candidate concentration vector using the following formula: in, This represents the corrected concentration; It is the probability that the sensor is in the i-th health state; It is made by all The resulting state probability vector.

7. The method for a catalytic combustion combustible gas sensor according to claim 5, characterized in that, Uncertainty quantification of the corrected concentration to obtain a solution uncertainty score includes: quantifying the uncertainty of the corrected concentration using the following formula: in, This represents the score for solving the uncertainty. Represents the probability vector Calculate the normalized Shannon entropy.

8. A catalytic combustion type combustible gas sensor, characterized in that, include: The dynamic response capture monitoring module is used to continuously monitor the baseline value of the analog signal voltage and determine whether to trigger dynamic response capture. The response sequence acquisition module is used to acquire response sequences in response to dynamic response capture. The dynamic fingerprint feature vector extraction module is used to extract dynamic fingerprint feature vectors from the response sequence to obtain dynamic feature vectors and steady-state voltage; The sensor health status classification module is used to classify the sensor health status based on the dynamic feature vector to obtain the sensor health status and state probability vector. The state-adaptive concentration calculation module is used to perform state-adaptive concentration calculation on steady-state voltage based on sensor health status and state probability vector to obtain corrected concentration and calculation uncertainty score. The decision output and alarm module is used to make decision outputs and alarms based on the corrected concentration and the calculated uncertainty score to obtain the final concentration display value and gas over-limit alarm.

Images