Hydrogen leakage diagnosis system and method based on multi-modal decision

Abstract

The invention discloses a hydrogen leakage diagnosis system and method based on multi-modal decision. The system comprises an environment perception and compensation unit, a multi-modal decision center and a cloud collaboration engine. The method comprises the following steps: S1, signal acquisition and preprocessing; s2, adopting an environment coupling compensation algorithm; s3, triggering a multi-mode decision; and S4, carrying out cloud collaborative calibration. Complex interference such as temperature and humidity fluctuation and barometric disturbance and time-varying drifting of the sensor are overcome, and the risk of false alarm and missing alarm is remarkably restrained.

Description

Technical Field

[0001] This invention relates to the field of gas detection technology, and in particular to a hydrogen leak diagnosis system and method based on multimodal decision-making. Background Technology

[0002] Hydrogen, as a key carrier of clean energy, is colorless, odorless, flammable, and explosive, making highly reliable sensors essential for safe applications. Among current mainstream hydrogen detection technologies, optical sensors, while possessing advantages in resisting electromagnetic interference, suffer from high false alarm rates in industrial scenarios due to significant temperature drift and cross-sensitivity to mechanical stress. Electrochemical sensors, while exhibiting excellent selectivity, are limited by their electrochemical reaction mechanisms, generally suffering from short lifespans and slow response at low temperatures. Semiconductor sensors based on the work function principle, while theoretically promising in terms of sensitivity, face three common bottlenecks in practical applications: First, environmental parameters such as temperature, humidity, and air pressure fluctuations directly modulate the work function of electrons on the surface of the sensitive material, causing measurement drift of up to ±10%, with false alarm rates surging by 40% especially under conditions of humidity >80%. Second, the hydrogen embrittlement effect of palladium-based catalysts leads to insufficient long-term stability, with sensitivity decreasing by over 30% after 5000 adsorption cycles. Third, while existing solutions attempt to improve dynamic range through multi-mode switching, they lack integrated environmental adaptive strategies, failing to cope with real-time interference in complex industrial scenarios.

[0003] The root cause of these problems lies in the fact that existing technologies focus excessively on hardware structure improvements while neglecting the value of system-level algorithm collaboration. Particularly in the field of work function sensors, the long-standing absence of cloud-based drift correction mechanisms forces manual calibration up to once a month, leading to a surge in maintenance costs. As the hydrogen energy industry expands, it urgently needs to overcome three major pain points: poor environmental adaptability, short lifespan, and insufficient full-range accuracy. Summary of the Invention

[0004] To address the issues of temperature drift interference, insufficient long-term stability, and false alarm rate in high-humidity environments faced by existing work function hydrogen sensors in complex industrial scenarios, this invention proposes a hydrogen leak diagnosis system and method based on multimodal decision-making, thereby resolving the aforementioned problems.

[0005] This application discloses a hydrogen leak diagnosis system based on multimodal decision-making, including an environmental perception and compensation unit, a multimodal decision-making center, and a cloud-based collaborative engine;

[0006] The environmental sensing and compensation unit includes:

[0007] Work function hydrogen sensor, used to acquire raw hydrogen concentration signals in real time;

[0008] Temperature, humidity and barometric pressure sensors are used to collect environmental interference parameters in real time.

[0009] An edge computing node configured to execute a signal noise reduction and environmental coupling compensation algorithm;

[0010] The multi-modal decision hub is deployed on an edge computing node, and through four-level series logic of variance analysis, spectrum detection, concentration gradient evaluation, and temperature and humidity threshold checking, dynamically switches the sensor working mode;

[0011] The cloud collaborative engine connects the edge computing node through a LoRa network, constructs a regional background concentration model based on geographic tags, and generates a dynamic compensation coefficient to perform real-time correction.

[0012] The application also discloses a hydrogen leakage diagnosis method based on multi-modal decision, which is realized based on the above hydrogen leakage diagnosis system and includes the following steps:

[0013] S1, signal acquisition and preprocessing, using a sensor to collect an original hydrogen concentration signal and environmental interference parameters, and outputting a stable time sequence signal after noise reduction;

[0014] S2, using an environmental coupling compensation algorithm to select a compensation algorithm according to a real-time humidity value;

[0015] S3, multi-modal decision triggering, based on the compensated hydrogen concentration sequence to execute a four-level series logic of a work function hydrogen sensor;

[0016] S4, cloud collaborative calibration, aggregating node data in a region, constructing a regional background concentration model, and generating a dynamic compensation coefficient to correct the work function output.

[0017] Preferably, the S1 includes the following steps:

[0018] An original hydrogen concentration signal S raw (t) is collected by a work function hydrogen sensor, a temperature T is collected by a temperature sensor, a humidity H is collected by a humidity sensor, and a gas pressure P is collected by a gas pressure sensor.

[0019] And a filter algorithm is used to reduce noise of the original hydrogen concentration signal, and a stable time sequence signal V out is output.

[0020] Preferably, the filter algorithm uses a smoothing filter algorithm:

[0021]

[0022] Wherein, t is time, i is the sample point number, and s represents unit second.

[0023] Preferably, the coupling model of the environmental coupling compensation algorithm in the S2 is:

[0024]

[0025] Wherein, VH are coupling coefficients. ΔT is temperature change, ΔH is humidity change, ΔP is air pressure change, ε noise is noise.

[0026] Preferably, the compensation algorithm is as follows:

[0027] When the humidity is ≤ 95% RH, a linear compensation model is used to calculate the real work function change:

[0028]

[0029] where ΔΦ H is the real work function change, k1 is the intrinsic gain of the work function hydrogen sensor, k2, k3, and k4 are dynamic weight coefficients.

[0030] When the humidity is > 95% RH, a nonlinear compensation model is used to calculate the real work function change:

[0031]

[0032] where,

[0033] f(H) = k2(1 + 0.05(H - 95));

[0034] g(T) = k3e 0.02(T-25) ;

[0035] f(H) is the temperature-compensated humidity correction term, and g(T) is the humidity-compensated temperature correction term.

[0036] Preferably, the S3 includes the following steps:

[0037] Calculate the sliding window variance σ 2 , if σ 2 < 0.1% FS, activate the static sampling mode;

[0038] If σ 2 ≥ 0.1% FS, analyze the original voltage spectrum S(50Hz), if S(50Hz) > 3dB, activate the differential mode;

[0039] If S(50Hz) ≤ 3dB, calculate the concentration gradient G, if G > 100ppm / s, activate the auxiliary sensor;

[0040] If G ≤ 100ppm / s, detect the temperature and humidity, and when the temperature and humidity exceed the threshold, activate the fusion verification mode.

[0041] Preferably, the S4 includes the following steps:

[0042] Automatically trigger the zero-point calibration process at a low activity period every day;

[0043] Constructing the regional background concentration model C of the nodes within the aggregated geographical radius of 500 meters base ;

[0044] If a single node deviates from C for more than 48 hours base , a dynamic compensation coefficient is generated and the function output is corrected.

[0045] Preferably, the zero-point calibration process is as follows:

[0046] The edge computing node cuts off the hydrogen sampling channel, collects the background output value V and real-time environmental parameters, and uploads them to the cloud collaborative engine via the LoRa network. base

[0047] Preferably, the regional background concentration model is:

[0048]

[0049] Where x, y are geographical coordinates, C i(t) represents the hydrogen concentration value reported by the mth node at time t, w m is the weight, and d m is the distance.

[0050] Advantages of the invention:

[0051] 1. The present application continuously optimizes the reference value by real-time sensing of environmental parameters and sensor state changes, effectively overcoming complex disturbances such as temperature and humidity fluctuations, air pressure disturbances, and sensor time-varying drift.

[0052] 2. The present application significantly suppresses false alarms and missed alarms, and improves monitoring accuracy, system reliability and environmental adaptability, giving the hydrogen leakage monitoring system strong robustness and excellent performance in dealing with variable working conditions in industrial sites and continuous stable operation. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 is the flowchart of the hydrogen leakage diagnosis method based on multi-modal decision-making of the embodiment of the present application;

[0054] Figure 2 is the flowchart of the environmental coupling compensation algorithm of the embodiment of the present application;

[0055] Figure 3 is the flowchart of the multi-modal decision-making triggering process of the embodiment of the present application;

[0056] Figure 4 is the flowchart of the cloud collaborative calibration process of the embodiment of the present application. DETAILED DESCRIPTION

[0057] ​For the purpose, technical solutions and advantages of the present application to be more clearly and obviously understood, the present application is further described in detail below with reference to the accompanying drawings and examples.

[0058] One embodiment of the present application discloses a hydrogen leakage diagnosis system based on multi-modal decision-making, including an environment perception and compensation unit, a multi-modal decision-making hub and a cloud collaborative engine.

[0059] The environment perception and compensation unit takes the work function hydrogen sensor (MOSFET work function sensor in this embodiment) as the core detection terminal, and collects the original hydrogen concentration signal in real time. At the same time, it integrates temperature sensor, humidity sensor and barometric pressure sensor to collect environmental interference parameters in real time. Through edge computing node, signal noise reduction and environmental coupling compensation algorithm are run to suppress the original drift caused by temperature fluctuation and the positive correlation error caused by humidity interference, so as to provide accurate hydrogen concentration signal for subsequent decision-making.

[0060] The multi-modal decision-making hub is deployed in the edge computing node. Through four-level serial logic of variance analysis, spectrum detection, concentration gradient evaluation and temperature and humidity threshold verification, the sensor static sampling mode, differential mode and sensor auxiliary mode are dynamically activated to realize adaptive balance of power consumption and response speed.

[0061] The cloud collaborative engine connects the edge computing node through LoRa network, automatically aggregates the background data within a geographical radius of 500 meters to construct a regional background concentration model, generates a dynamic compensation coefficient and issues it to the work function hydrogen sensor for real-time correction.

[0062] Another embodiment of the present application discloses a hydrogen leakage diagnosis method based on multi-modal decision-making, which is realized based on the above hydrogen leakage diagnosis system. The flow is as shown in Figure 1 The flow is as shown in

[0063] S1, signal acquisition and pretreatment, using sensor to collect original hydrogen concentration signal and environmental interference parameters, and outputting stable time sequence signal after noise reduction.

[0064] At the key points of the monitored object, the work function hydrogen sensor is deployed to form a sensor array, and the temperature sensor, humidity sensor and barometric pressure sensor are integrated. The edge computing node collects the original sensor signal through SPI interface at a frequency of 1 kHz to obtain the original hydrogen concentration sensor signal S raw (t) and environmental interference parameters (temperature T, humidity H, barometric pressure P). The original hydrogen concentration signal is denoised by using filtering algorithm (such as smoothing filtering algorithm, Kalman filtering algorithm) to output stable time sequence signal V out In this embodiment, smoothing filtering algorithm is used for pretreatment:

[0065]

[0066] wherein t is time, i is the sampling point number, and s represents a unit of second.

[0067] The sliding window is 5 sampling points, i.e. 0.5 s, or the Kalman filter is switched according to the electromagnetic interference intensity, and a stable timing signal V after noise reduction is output out .

[0068] S2, an environmental coupling compensation algorithm is adopted, and a compensation algorithm is selected according to a real-time humidity value, and a specific flow is as shown in Figure 2 .

[0069] The stable timing signal V output by the work function hydrogen sensor out essentially reflects the change (ΔΦ H ) in surface electron work function caused by hydrogen adsorption, but the signal is disturbed by three environments of temperature T, humidity H and air pressure P, and a disturbance coupling model is as follows:

[0070]

[0071] wherein V H is a real hydrogen signal, and α, β and γ are coupling coefficients. ΔT is a temperature change, ΔH is a humidity change, ΔP is an air pressure change, and ε noise is noise.

[0072] When the humidity is ≤95% RH (relative humidity), a linear compensation model is used to calculate the real work function change amount:

[0073]

[0074] wherein ΔΦ H is the real work function change amount (directly corresponding to the hydrogen concentration), k1 is the intrinsic gain of the work function hydrogen sensor, which is determined by the factory setting of the sensor, k2, k3 and k4 are dynamic weight coefficients, which are determined by an adaptive algorithm, and the initial value is taken as 1. In the embodiment, the dynamic weight coefficients are updated every hour by a stochastic gradient descent algorithm:

[0075]

[0076] wherein η is a learning rate, which is 0.01 by default in the embodiment, is the dynamic weight coefficient of the next hour, is the dynamic weight coefficient of the previous hour, and k i is the current dynamic coefficient.

[0077] When the humidity is >95% RH, a nonlinear compensation model is switched to calculate the real work function change amount:

[0078]

[0079] wherein,

[0080] f(H) = k2(1 + 0.05(H - 95));

[0081] g(T) = k3e 0.02(T-25) ;

[0082] f(H) is a humidity correction term compensated by temperature, and g(T) is a temperature correction term compensated by humidity.

[0083] S3, multi-modal decision trigger, execute the work function hydrogen sensor four-stage series logic based on the compensated hydrogen concentration sequence. The specific process is shown in Figure 3

[0084] The work function hydrogen sensor needs to balance sensitivity, power consumption and anti-interference in different scenarios. The work function sensor should have four working modes: ① low concentration steady state, prefer power optimization (μW level static sampling mode); ② electromagnetic interference environment, sacrifice power for noise immunity (differential mode); ③ sudden leakage, enable auxiliary sensor to ensure response speed (sensor auxiliary mode); ④ high temperature and humidity, multi-sensor cross verification to prevent false alarm.

[0085] Decision trigger logic: first, evaluate the environmental steady state (variance analysis), input the compensated hydrogen concentration sequence C(t), t∈[t-60s,t], then calculate the sliding window variance:

[0086]

[0087] wherein, j is the serial number of data points in the sliding window, the value range is from 1 to N, N is the total number of data points contained in the sliding window, C i is the hydrogen concentration value of the i-th data point in the sliding window, and μ is the arithmetic mean of all concentration data in the sliding window.

[0088] If σ 2 <0.1%·FS (FS is the accuracy unit, representing the percentage of the measurement range, 0.1%·FS is 0.1% of the full scale), the sensor activates the static sampling mode.

[0089] If σ 2 ≥0.1%·FS, electromagnetic interference detection is performed. The input work function sensor original voltage V raw(t) (1kHz sampling, 0.5s window), weighted by Hanning window as:

[0090] V win(t) = V raw(t) ·[0.5-0.5cos(2πt / T)];

[0091] wherein, V win(t) ​Vraw is the voltage value of the work function sensor after being weighted by the Hanning window raw(t) Vraw is the voltage value of the work function sensor after being weighted by the Hanning window

[0092] The spectrum FET is:

[0093]

[0094] If S(50Hz)>3dB, activate the differential mode.

[0095] If S(50Hz)<3dB, proceed to the leakage risk judgment stage. Input the current compensated hydrogen concentration sequence C(t) and historical value C(t-Δt), calculate the concentration gradient G(ppm / s):

[0096]

[0097] If G>100ppm / s, activate the auxiliary sensor to improve the response speed.

[0098] If G<100ppm / s, detect the temperature and humidity, input the real-time temperature T and humidity H, when the temperature is too high or too low (in this embodiment, T [20℃,60℃]), the humidity (in this embodiment, H>0% RH) is too high, activate the fusion verification mode, otherwise maintain the current mode.

[0099] S4, cloud collaborative calibration, aggregate regional node data, construct regional background concentration model, and generate dynamic compensation coefficient correction work function output. The specific process is as shown in Figure 4

[0100] To realize the stability improvement of the work function hydrogen sensor in the whole life cycle, the long-term drift problem is solved through spatio-temporal correlation analysis and adaptive compensation mechanism. The system automatically triggers the zero point calibration process every day in the preset low activity period (default 02:00-04:00): the edge computing node cuts off the hydrogen sampling channel, collects the background output value V base (raw voltage output without hydrogen environment) and real-time environmental parameters, and uploads them to the cloud collaborative engine via LoRa / network.

[0101] The cloud collaborative engine aggregates the node data within a radius of 500 meters based on the geographic label (x,y), and constructs a regional background concentration model:

[0102]

[0103] Wherein, x,y are geographic coordinates, C i(t) m represents the hydrogen concentration value reported by the mth node at time t, w m is the weight, d m is the distance, and the weight w​m decreases with the square of the distance d m .

[0104] When a single node deviates from C base for more than 48 hours threshold (±5%·FS in this embodiment), it is determined that the sensor drift is abnormal, and the cloud collaborative engine generates a dynamic compensation coefficient:

[0105]

[0106] Where V real is the original voltage value of the work function sensor of the drift node.

[0107] And inject edge computing nodes through the downlink , real-time correction of work function output:

[0108]

[0109] Where V calib is the voltage value of the work function sensor after cloud collaborative calibration.

[0110] The embodiments of the present application aim at the core defects of the existing commercial work function hydrogen sensor in complex industrial scenes, such as temperature drift interference, insufficient long-term stability, and high false alarm rate in high humidity environment, etc. A hydrogen leakage diagnosis system and method based on multi-modal decision are constructed. Without changing the hardware structure of the commercial sensor, the dynamic weight coefficient of the environmental coupling compensation algorithm is used to offset the temperature / humidity / pressure interference in real time, combined with the reinforcement learning decision mechanism triggered by multi-modal decision, and relying on the spatio-temporal correlation analysis of the cloud collaborative engine to realize drift correction, finally forming a three-level architecture of "edge perception-intelligent decision-cloud verification", which significantly improves the detection accuracy, reduces the false alarm rate and prolongs the service life. With pure algorithm and system architecture innovation, it breaks through the traditional hardware preparation barrier, and constructs a three-dimensional protection system covering environmental compensation model, multi-modal logic rule and cloud-edge collaborative network.

[0111] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A hydrogen leak diagnostic system based on multimodal decision-making, characterized in that, This includes an environmental perception and compensation unit, a multimodal decision-making center, and a cloud-based collaborative engine; The environmental sensing and compensation unit includes: Work function hydrogen sensor, used to acquire raw hydrogen concentration signals in real time; Temperature, humidity and barometric pressure sensors are used to collect environmental interference parameters in real time. Edge computing nodes are configured to perform signal denoising and environmental coupling compensation algorithms. The multimodal decision-making center is deployed on an edge computing node and dynamically switches the sensor's working mode through a four-level serial logic of variance analysis, spectrum detection, concentration gradient assessment, and temperature and humidity threshold verification. The cloud-based collaborative engine connects to edge computing nodes via a LoRa network, constructs a regional background concentration model based on geographic tags, and generates dynamic compensation coefficients to perform real-time correction.

2. A hydrogen leakage diagnosis method based on multimodal decision-making, characterized in that, The hydrogen leak diagnostic system based on claim 1 includes the following steps: S1. Signal acquisition and preprocessing: The original hydrogen concentration signal and environmental interference parameters are acquired by a sensor, and a stable time-series signal is output after noise reduction. S2. An environmental coupling compensation algorithm is adopted, and the compensation algorithm is selected based on the real-time humidity value. S3, Multimodal decision triggering, based on the compensated hydrogen concentration sequence to execute the work function of the hydrogen sensor's four-level serial logic; S4, cloud-based collaborative calibration, aggregates node data within the region, constructs a regional background concentration model, and generates a dynamic compensation coefficient correction work function output.

3. The hydrogen leakage diagnosis method based on multimodal decision-making according to claim 2, characterized in that, S1 includes the following steps: The raw hydrogen concentration signal S is acquired using a work function hydrogen sensor. raw (t), temperature T is collected using a temperature sensor, humidity H is collected using a humidity sensor, and air pressure P is collected using an air pressure sensor; The original hydrogen concentration signal is denoised using a filtering algorithm, and a stable time-series signal V is output. out .

4. The hydrogen leakage diagnosis method based on multimodal decision-making according to claim 3, characterized in that, The filtering algorithm used is a smoothing filtering algorithm: Where t is time, i is the sampling point number, and s represents the unit of seconds.

5. The hydrogen leakage diagnosis method based on multimodal decision-making according to claim 4, characterized in that, The coupling model of the environmental coupling compensation algorithm in S2 is as follows: Among them, V H The signal represents real hydrogen, with α, β, and γ being coupling coefficients. ΔT represents temperature change, ΔH represents humidity change, ΔP represents pressure change, and ε... noise It is noise.

6. The hydrogen leakage diagnosis method based on multimodal decision-making according to claim 5, characterized in that, The compensation algorithm is as follows: When the humidity is ≤95%RH, the actual change in work function is calculated using a linear compensation model: Where, ΔΦ H k1 represents the actual change in the work function, k2 represents the intrinsic gain of the hydrogen sensor, and k3 and k4 represent the dynamic weighting coefficients. When the humidity is greater than 95% RH, switch to the nonlinear compensation model to calculate the actual change in the work function: in, f(H) = k2(1 + 0.05(H - 95)); g(T)=k3e 0.02(T-25) ; f(H) is the humidity correction term for temperature compensation, and g(T) is the temperature correction term for humidity compensation.

7. The hydrogen leakage diagnosis method based on multimodal decision-making according to claim 6, characterized in that, S3 includes the following steps: Calculate the variance σ of the sliding window 2 If σ 2 <0.1%·FS, activate static sampling mode; If σ 2 ≥0.1%·FS, analyze the original voltage spectrum S(50Hz), if S(50Hz)>3dB, activate differential mode; If S(50Hz)≤3dB, calculate the concentration gradient G; if G>100ppm / s, activate the auxiliary sensor. If G≤100ppm / s, temperature and humidity are detected. When the temperature and humidity exceed the threshold, the fusion verification mode is activated.

8. The hydrogen leakage diagnosis method based on multimodal decision-making according to claim 7, characterized in that, S4 includes the following steps: The zero-point calibration process is automatically triggered daily during periods of low activity. A regional background concentration model C is constructed by aggregating node data within a 500-meter geographic radius. base ; If a single node deviates from C for 48 consecutive hours base When the threshold is exceeded, a dynamic compensation coefficient is generated and the work function output is corrected.

9. The hydrogen leakage diagnosis method based on multimodal decision-making according to claim 8, characterized in that, The zero-point calibration process is as follows: The edge computing node cuts off the hydrogen sampling channel and collects the background output value V. base Real-time environmental parameters are uploaded to the cloud-based collaborative engine via the LoRa network.

10. The hydrogen leakage diagnosis method based on multimodal decision-making according to claim 9, characterized in that, The regional background concentration model is as follows: Where x, y are geographic coordinates, C i(t) w represents the hydrogen concentration value reported by the m-th node at time t. m For weights, d m For distance.

Images