A miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation

Abstract

The application discloses a miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation and relates to the technical field of gas sensors.The miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation comprises a microfluidic chip base body, a microfluid channel arranged in the microfluidic chip base body, the microfluid channel having an air inlet and an air outlet, and a micro-nano composite sensitive array integrated in the microfluid channel.The miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation changes the temperature and humidity compensation from traditional signal post-processing to active environment decoupling before response, controls the error absolute value of gas concentration detection within a set temperature range in degrees Celsius and a relative humidity range, and compensates for the response without additional delay.The feature extraction and early warning level determination are completed by a pulse neural network in an event-driven manner without the need of uploading data to an upper computer for processing, so that independent decision-making and instant response of the sensing end are realized.

Description

Technical Field

[0001] This invention relates to the field of gas sensor technology, specifically to a miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation. Background Technology

[0002] Volatile organic compounds (VOCs) are a significant class of environmental pollutants, widely present not only in industrial emissions and indoor decoration but also in certain characteristic VOCs such as toluene, formaldehyde, and acetone, which are closely related to human health, serving as biomarkers for diseases like lung cancer. Therefore, achieving high-precision and rapid detection of trace VOCs has significant application value in environmental monitoring, public safety, and early medical diagnosis.

[0003] Currently, gas detection methods mainly include titration, ultraviolet absorption colorimetry, and gas chromatography-mass spectrometry. However, traditional methods often rely on large instruments and equipment, are complex to operate, have long detection cycles, and are costly, making it difficult to meet the needs of real-time on-site detection. In recent years, gas sensors based on electrochemical or semiconductor principles have gradually attracted attention. For example, CN104965003A describes a trace VOCs gas detection system based on a four-wire sensor. This system uses a four-wire gas sensor in conjunction with a digital temperature and humidity sensor, and through coordinated control between a host computer and a slave computer, it achieves the detection of VOCs gas and temperature and humidity correction, with advantages such as high detection accuracy and short response time.

[0004] While the aforementioned existing technologies have made some progress in VOCs detection, their system structure still has significant shortcomings. First, the system includes a VOCs mixed gas bladder, a nitrogen cylinder, multiple valves, multiple pumps, and an independent detection chamber, resulting in a large overall size and complex structure, which hinders portability and rapid on-site deployment. Second, although it incorporates temperature and humidity sensors for data correction, this correction method primarily relies on offline processing by a host computer, making real-time dynamic compensation at the sensor level impossible. This limits detection stability when environmental conditions change rapidly and makes it difficult to provide immediate warnings of hazardous gas concentrations.

[0005] Therefore, how to further miniaturize, reduce power consumption, and achieve real-time temperature and humidity compensation for VOCs gas sensor systems while ensuring detection accuracy, and how to build intelligent detection terminals with local early warning capabilities, have become urgent technical problems to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to provide a miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation, so as to solve the problems mentioned in the background art.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation, comprising:

[0008] Microfluidic chip substrate;

[0009] Microchannels are disposed inside the microfluidic chip substrate, and the microchannels have an air inlet and an air outlet;

[0010] A micro-nano composite sensing array integrated within the microchannel, the micro-nano composite sensing array comprising micro-nano sensors of at least two different types of sensing materials for outputting response signals;

[0011] A digital temperature and humidity sensor integrated within the microchannel is used to output real-time temperature and humidity data.

[0012] A micro heating array is disposed on the microfluidic chip substrate, the micro heating array is located below the micro-nano composite sensing array, and forms thermal coupling with the micro-nano composite sensing array;

[0013] A control module electrically connected to the micro heating array, the micro-nano composite sensitive array, and the digital temperature and humidity sensor, the control module comprising a microcontroller, a storage unit, and a wireless communication unit;

[0014] The storage unit contains a temperature and humidity-thermal field coupling proxy model. The microcontroller is configured to call the temperature and humidity-thermal field coupling proxy model to calculate thermal field modulation parameters based on the real-time temperature data and the real-time humidity data, and to control the micro heating array to perform differentiated heating on each micro-nano sensor based on the thermal field modulation parameters.

[0015] The microcontroller is also configured to acquire the response signal of the micro-nano composite sensitive array, input the response signal into the spiking neural network decision engine, and the spiking neural network decision engine outputs a graded early warning signal based on the input signal.

[0016] The microcontroller drives the wireless communication unit to send warning information based on the graded warning signal.

[0017] Preferably, the micro-nano composite sensing array includes multiple sensing units arranged in an array configuration, wherein the multiple sensing units include at least two of the following: metal oxide semiconductor sensing units, conductive polymer sensing units, and catalytic combustion sensing units.

[0018] Each sensitive unit has a three-dimensional nanodendritic structure, which is grown on the electrode surface inside the microchannel by electrodeposition.

[0019] The width of the microchannel is 100μm to 300μm, and a detection cavity is formed inside the microchannel, with a volume of 20μL to 50μL.

[0020] The digital temperature and humidity sensor and the micro-nano composite sensitive array are encapsulated in the same detection cavity, and the sampling frequency of the digital temperature and humidity sensor is synchronized with the response signal acquisition frequency of the micro-nano composite sensitive array.

[0021] Preferably, the micro heating array is an independently controllable heating point array, which includes at least 16 heating points, each heating point corresponding to a sensitive unit or a group of sensitive units;

[0022] The microcontroller controls the heating temperature of the micro heating array to switch as needed within a set temperature range, where the lower limit of the set temperature range is 150°C and the upper limit is 400°C.

[0023] The thermal field modulation parameters include the target temperature value, heating duration and heating sequence of each heating point. The microcontroller updates the thermal field modulation parameters at a frequency of 50Hz or higher and controls the micro heating array to execute.

[0024] Preferably, the temperature and humidity-thermal field coupling proxy model is a deep kernel learning proxy model, which is obtained through offline training. The training data includes the optimal thermal field parameters under multiple temperature and humidity combinations.

[0025] The deep kernel learning proxy model is embedded in the storage unit. When the microcontroller runs, it calls the deep kernel learning proxy model to perform forward inference. The input is real-time temperature data and real-time humidity data, and the output is thermal field modulation parameters.

[0026] The number of parameters in the deep kernel learning agent model does not exceed 2000 floating-point numbers.

[0027] Preferably, the spiking neural network decision engine includes an input layer, a hidden layer, and an output layer;

[0028] The input layer corresponds to each sensitive unit of the micro-nano composite sensitive array and is used to receive the response signal;

[0029] The hidden layer includes leaky integrated firing neurons, which are connected to each other via synaptic plasticity.

[0030] The output layer includes three neurons, corresponding to the first-level warning, the second-level warning, and the third-level warning, respectively.

[0031] The microcontroller is configured to run the spiking neural network decision engine, which operates in an event-driven manner and triggers pulse calculation only when the change in the response signal exceeds a set threshold.

[0032] Preferably, the microcontroller is further configured to perform time window integration on the pulse frequency of the output layer neurons of the spiking neural network decision engine, with each time window lasting for 200 ms.

[0033] The microcontroller determines whether the pulse frequency of the output layer neurons exceeds the frequency threshold of the corresponding warning level within three consecutive time windows. When the frequency threshold is exceeded in three consecutive time windows, the corresponding warning level is triggered.

[0034] The microcontroller drives the local audible and visual alarm to output the corresponding level of audible and visual signal based on the confirmed warning level, and sends the warning information to the external terminal through the wireless communication unit.

[0035] Preferably, a micro diaphragm pump is also integrated on the microfluidic chip substrate. The micro diaphragm pump is connected to the air inlet of the microchannel and is used to actively draw external gas into the microchannel.

[0036] The outlet of the microchannel is connected to the atmosphere, forming a gas flow path;

[0037] The microcontroller is configured to control the start-up, shutdown, and flow rate of the micro diaphragm pump, the flow rate of which is matched with the sampling frequency of the micro-nano composite sensing array.

[0038] Preferably, the control module further includes a local audible and visual alarm, which is electrically connected to the microcontroller;

[0039] The local sound and light alarm includes a multi-color light-emitting diode and a buzzer. The microcontroller controls the multi-color light-emitting diode to display the corresponding color according to the graded warning signal, and controls the buzzer to emit a prompt sound of the corresponding frequency.

[0040] The wireless communication unit is a low-power Bluetooth communication unit or a long-range radio communication unit.

[0041] Preferably, the microcontroller is configured to acquire real-time temperature and humidity data from the digital temperature and humidity sensor before acquiring the response signal of the micro-nano composite sensitive array, calculate thermal field modulation parameters based on the real-time temperature and humidity data and calling the temperature-humidity-thermal field coupling proxy model, control the micro heating array to execute the thermal field modulation parameters, and then acquire the response signal of the micro-nano composite sensitive array.

[0042] A miniaturized VOCs gas detection method based on temperature and humidity compensation, applied to the aforementioned system, includes the following steps:

[0043] Step 1: Obtain real-time temperature and humidity data;

[0044] Step 2: Input the real-time temperature data and the real-time humidity data into the temperature-humidity-thermal field coupling proxy model stored in the storage unit, and calculate the thermal field modulation parameters;

[0045] Step 3: Control the micro heating array to perform differentiated heating on the micro-nano composite sensitive array according to the thermal field modulation parameters;

[0046] Step 4: Acquire the response signal output by the micro-nano composite sensing array;

[0047] Step 5: Input the response signal into the spiking neural network decision engine, and the spiking neural network decision engine outputs a graded early warning signal;

[0048] Step Six: Drive the wireless communication unit to send warning information according to the graded warning signal, and drive the local sound and light alarm to output the corresponding level of sound and light signal.

[0049] This invention provides a miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation. It has the following beneficial effects:

[0050] This invention proposes a miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation. It integrates a microfluidic chip, a micro-nano composite sensing array, a micro-heating array, a digital temperature and humidity sensor, and a control module into a single unit, achieving handheld portable deployment and long-term continuous operation. By encapsulating the digital temperature and humidity sensor and the micro-nano composite sensing array in the same detection cavity and maintaining synchronized sampling frequencies, the time misalignment between temperature and humidity measurement and gas response signals is eliminated. Furthermore, before acquiring the response signal, the microcontroller first obtains real-time temperature and humidity data, calls a deep kernel learning proxy model stored in the memory unit to calculate thermal field modulation parameters, and controls the micro-heating array to provide differentiated heating to each sensing unit. This ensures that the operating temperature of each sensing unit actively avoids the sensitive range under the current temperature and humidity conditions. Temperature and humidity compensation is transformed from traditional post-processing of the signal to proactive environmental decoupling before the response. Within the set temperature range and relative humidity range, the absolute value of the gas concentration detection error can be controlled, and the compensation response has no additional delay.

[0051] This system integrates a spiking neural network decision engine and a hierarchical early warning mechanism at the sensor end. The microcontroller directly inputs the multidimensional response signal output from the micro-nano composite sensing array into the spiking neural network, which then performs feature extraction and early warning level determination in an event-driven manner. This eliminates the need to upload data to a host computer for processing, enabling independent decision-making and real-time response at the sensor end. The spiking neural network employs leaky integration firing neurons and synaptic plasticity mechanisms, combined with continuous time window integral decision logic, effectively filtering out instantaneous interference caused by environmental fluctuations, thus improving early warning accuracy. The local audible and visual alarm and the wireless communication unit work together, simultaneously outputting on-site warning signals and remotely transmitted information when an early warning is triggered, forming a complete edge-end intelligent sensing and early warning closed loop. Compared to traditional detection systems that rely on host computers for post-correction of temperature and humidity and lack built-in early warning capabilities, this invention achieves improvements in miniaturization integration, environmental adaptability, real-time decision-making, and early warning reliability. Attached Figure Description

[0052] Figure 1 This is a data flow diagram of a miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation according to the present invention.

[0053] Figure 2 This is a flowchart illustrating the early warning judgment logic of a miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation according to the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Please see Figure 1 and Figure 2 This invention provides a technical solution: a miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation. The microfluidic chip substrate is made of polydimethylsiloxane material through a multilayer soft photolithography process. The substrate is etched to form a microchannel with a width of 200μm and a depth of 150μm. One end of the microchannel is an air inlet and the other end is an air outlet. The air inlet is directly connected to the air outlet of a miniature diaphragm pump integrated on the same substrate. The miniature diaphragm pump has a size of 5mm×5mm×2mm and is used to actively draw external gas into the microchannel.

[0056] The microfluidic chip substrate adopts a multilayer composite structure of glass and polydimethylsiloxane. The bottom substrate, where the micro heating array is located, is made of borosilicate glass or single-crystal silicon. These materials can withstand temperatures above 400°C without deformation. Polydimethylsiloxane is only used to prepare the upper flow channel layer of the microchannels and detection chamber. A silicon oxide thermal insulation film layer is set between the micro heating array and the upper polydimethylsiloxane flow channel layer. The thickness of the thermal insulation film layer is 5μm to 20μm, which can effectively block the high temperature generated by the heating array from being conducted to the polydimethylsiloxane layer, preventing the polydimethylsiloxane from melting or deforming due to high temperature. At the same time, it ensures that the thermal coupling efficiency between the micro heating array and the micro-nano composite sensing array is not affected, and ensures that the overall structure of the microfluidic chip substrate remains intact and functions normally when the heating temperature is stably controlled within the range of 150°C to 400°C.

[0057] A detection chamber with a volume of 30 μL is formed in the middle section of the microchannel. This chamber integrates a micro-nano composite sensing array, which consists of nine sensing units arranged in a 3×3 array. These nine units include three metal oxide semiconductor sensing units, three conductive polymer sensing units, and three catalytic combustion sensing units. Each sensing unit has a three-dimensional nanodendritic structure grown on the electrode surface via electrode deposition. The specific surface area of ​​this three-dimensional nanodendritic structure is 50 m². 2 / g.

[0058] A digital temperature and humidity sensor is also integrated within the same detection chamber. This digital temperature and humidity sensor is packaged on the same plane as the nine sensing units, and its sampling frequency is set to 100Hz, which is synchronized with the frequency of the response signal acquisition of the sensing units.

[0059] Below the detection cavity, a micro heating array is embedded on the microfluidic chip substrate. This micro heating array is an independently controllable 4×4 heating point matrix with a total of 16 heating points. Each heating point corresponds to a sensitive unit or a group of adjacent sensitive units. The heating temperature of each heating point can be independently adjusted under the control of the microcontroller, and the heating temperature can be switched as needed within the range of 150℃ to 400℃.

[0060] The microfluidic chip substrate also integrates a control module, which includes a microcontroller, a storage unit, and a wireless communication unit. The microcontroller adopts the ARM Cortex-M4F architecture, and the storage unit contains a deep kernel learning agent model obtained through offline training. The model has 1800 floating-point parameters, and the inputs are temperature and humidity values. The outputs are thermal field modulation parameters consisting of the target temperature values, heating duration, and heating sequence of 16 heating points.

[0061] After the system starts, the microcontroller first reads the real-time temperature and humidity data output by the digital temperature and humidity sensor, inputs the real-time temperature and humidity data into the deep kernel learning surrogate model for forward inference, obtains the thermal field modulation parameters under the current operating conditions, and then controls 16 heating points to perform differentiated heating on their respective sensitive units according to the thermal field modulation parameters, so that the operating temperature of each sensitive unit actively avoids the temperature and humidity sensitive range.

[0062] Temperature and humidity-thermal field coupling refers to the fact that environmental temperature and humidity parameters directly change the surface adsorption state, carrier migration rate, and gas response characteristics of the micro-nano composite sensitive array. At the same time, the thermal field parameters output by the micro-heating array will inversely regulate the local temperature and humidity environment in which the sensitive array is located. The two form a physical field correlation that influences and relates each other. Active environmental decoupling refers to actively counteracting the interference of environmental temperature and humidity on the detection performance of the sensitive unit by using differentiated thermal field modulation of the micro-heating array before the sensitive array generates a gas response signal. This separates the environmental factors of temperature and humidity from the gas concentration response signal, avoiding the direct superposition of temperature and humidity fluctuations into the gas detection results, and achieving the goal of the detection signal reflecting only the change in VOCs gas concentration.

[0063] After the differential heating is completed, the microcontroller collects nine response signals output by nine sensitive units and sends these nine response signals as inputs to the spiking neural network decision engine running inside the microcontroller. The spiking neural network decision engine includes an input layer, a hidden layer, and an output layer. The input layer contains nine neurons corresponding to the nine response signals. The hidden layer contains 128 leaky integration firing neurons, which are connected to each other through synaptic plasticity. The output layer contains three neurons corresponding to the first-level warning, the second-level warning, and the third-level warning, respectively. The spiking neural network decision engine runs in an event-driven manner, triggering pulse calculation only when any signal change in the nine response signals exceeds a set threshold.

[0064] The pulse frequency of the output layer neurons is integrated by the microcontroller over a time window. Each time window lasts for 200ms. The microcontroller determines whether the pulse frequency of any output layer neuron exceeds the frequency threshold corresponding to the warning level within three consecutive time windows. If the frequency exceeds the threshold in all three consecutive time windows, the corresponding warning level is triggered.

[0065] Based on the confirmed alarm level, the microcontroller drives the multi-color LEDs in the local audible and visual alarm to display the corresponding color and controls the buzzer to emit a corresponding frequency of alert sound. On the other hand, it sends alarm information containing the alarm level, real-time gas concentration value, and temperature and humidity data to an external terminal through the wireless communication unit.

[0066] The micro-nano composite sensing array consists of nine sensing units arranged in a 3×3 array within a detection cavity inside a microfluidic channel. Specifically, the nine sensing units include three metal oxide semiconductor sensing units, three conductive polymer sensing units, and three catalytic combustion sensing units. The metal oxide semiconductor sensing units use tin oxide nanofibers as the sensing material, the conductive polymer sensing units use a polyaniline and carbon nanotube composite film as the sensing material, and the catalytic combustion sensing units use platinum-supported alumina as the sensing material.

[0067] Three-dimensional nanodendritic structures are grown on the electrode surface of each sensitive unit by electrodeposition. The electrodeposition process adopts a three-electrode system, with the working electrode potential controlled between -0.8V and -1.2V, and the deposition time between 300s and 600s. The deposition solution is a mixed solution of chloroauric acid and polyvinylpyrrolidone. The resulting three-dimensional nanodendritic structure exhibits a porous morphology, with the dendrite trunk diameter ranging from 80nm to 120nm and the interdendritic pore diameter ranging from 50nm to 100nm. This three-dimensional nanodendritic structure can be used directly as the substrate for the sensitive material or as a component of the sensitive material itself.

[0068] The microchannel adopts a composite structure of polydimethylsiloxane and glass. The width of the microchannel is 200μm and the depth is 150μm. The detection cavity set in the middle section of the microchannel is formed by local expansion of the microchannel. The length of the detection cavity is 3mm, the width is 2mm and the depth is 200μm. The calculated volume of the detection cavity is 30μL.

[0069] The digital temperature and humidity sensor uses an I²C interface digital temperature and humidity sensor chip. This chip and nine sensitive units are packaged in the same detection cavity. The packaging method is a flip-chip bonding process, which makes the humidity sensing layer of the temperature and humidity sensor chip and the surface of the sensitive units on the same horizontal plane with a distance of no more than 500μm. The sampling frequency of the digital temperature and humidity sensor is set to 100Hz by the synchronous clock signal output by the microcontroller. At each sampling moment, the microcontroller simultaneously triggers the acquisition of temperature and humidity data from the digital temperature and humidity sensor and the acquisition of response signals from the nine sensitive units, so that the temperature and humidity data and response signals are strictly aligned in time.

[0070] The micro heating array uses platinum metal resistance heating wires arranged in a serpentine pattern on the bottom layer of the microfluidic chip substrate. The micro heating array consists of 16 independently controllable heating points arranged in a 4×4 matrix. Each heating point occupies a plane size of 0.8mm×0.8mm, and the center-to-center distance between adjacent heating points is 1mm. The platinum resistance wire corresponding to each heating point has a width of 20μm, a thickness of 5μm, and a resistance value of 30Ω to 50Ω.

[0071] The power supply for each heating point is achieved by the microcontroller through 16 independently controlled metal-oxide-semiconductor field-effect transistor switches. Each switch is equipped with a pulse width modulation output. The microcontroller controls the heating power of each heating point by adjusting the duty cycle of the pulse width modulation signal.

[0072] The microcontroller has a built-in closed-loop control algorithm for heating temperature. Each heating point is equipped with a thin-film thermocouple temperature sensor. The thin-film thermocouple temperature sensor collects the actual temperature of the heating point in real time and feeds it back to the microcontroller. The microcontroller compares the actual temperature with the target temperature value of the heating point in the thermal field modulation parameters. It dynamically adjusts the pulse width modulation duty cycle through a proportional-integral-derivative algorithm to keep the actual temperature within ±5℃ of the target temperature value.

[0073] The PID closed-loop control of the miniature heating array adopts conventional temperature closed-loop regulation logic in this field. The microcontroller collects the feedback temperature of the thin-film thermocouples at each heating point in real time, compares the feedback temperature with the target temperature, and dynamically adjusts the heating power according to the temperature deviation. The PID parameters are tuned through offline batch testing. Multiple sets of temperature regulation experiments are completed under different temperature and humidity conditions. The parameter combination that makes the heating temperature stable, responds quickly and has no overshoot is selected as the final tuning parameters. After tuning, the parameters are stored in the storage unit. During operation, the microcontroller directly calls the tuned parameters to execute closed-loop control, ensuring that the actual temperature of each heating point is stable within ±5℃ of the target temperature.

[0074] The thermal field modulation parameters are obtained from the output of the temperature and humidity-thermal field coupling proxy model. These parameters include the target temperature value, heating duration and heating sequence of each of the 16 heating points. The target temperature value can be selected as needed within the range of the lower limit of 150°C to the upper limit of 400°C. The target temperature values ​​of different heating points at the same time can be different from each other.

[0075] The microcontroller performs the following operations cyclically at a frequency of 60Hz: read real-time temperature and humidity data, call the temperature-humidity-thermal field coupling proxy model to calculate new thermal field modulation parameters, compare the new thermal field modulation parameters with the currently executed thermal field modulation parameters, and if the target temperature value of any of the 16 heating points changes by more than 10℃ or the heating sequence changes, the microcontroller immediately updates the pulse width modulation output and heating sequence arrangement of that heating point; if the target temperature value changes of all heating points does not exceed 10℃ and the heating sequence does not change, the microcontroller maintains the current heating control state unchanged.

[0076] After the microcontroller completes the update of the thermal field modulation parameters, it continuously controls the heating state of each heating point according to the updated parameters in the next 60Hz cycle, so that the working temperature of each sensitive unit in the micro-nano composite sensitive array is always consistent with the optimal working temperature under the current temperature and humidity conditions.

[0077] The temperature and humidity-thermal field coupled surrogate model is constructed using a deep kernel learning framework, which consists of a deep neural network and a Gaussian process regression part cascaded together. The deep neural network part contains three fully connected layers. The first layer has an input dimension of 2 corresponding to temperature and humidity values, and an output dimension of 3 (12). The second layer has an output dimension of 64, and the third layer has an output dimension of 128. Each layer uses a modified linear unit as the activation function.

[0078] The Gaussian process regression part takes the feature vector output by the deep neural network as input and uses the radial basis function as the covariance function. The length scale parameter of the radial basis function is optimized and determined during the training phase by maximizing the marginal log-likelihood function.

[0079] The offline training process of the deep kernel learning agent model is as follows: The microfluidic chip substrate is placed in a temperature and humidity test chamber. The temperature of the test chamber is set at nine gradients with an interval of 5℃ in the range of 10℃ to 5℃, and the relative humidity is set at eight gradients with an interval of 10% in the range of 20% to 90%, forming 72 temperature and humidity combination conditions.

[0080] Under each operating condition, the microcontroller controls the micro heating array to sequentially set the temperature combinations of 16 heating points in an iterative manner. The temperature of each heating point is set with 11 gradients at 25°C intervals within the range of 150°C to 400°C. The total number of temperature combinations of the 16 heating points reaches 11 to the power of 16. The Bayesian optimization method is used to select 300 temperature combinations from this combination space for actual testing. During the test, the response signal of the micro-nano composite sensitive array to standard concentration VOCs gas is recorded. Taking the maximization of the signal-to-noise ratio of the response signal as the objective function, the optimal thermal field modulation parameters under the temperature and humidity conditions are determined through Bayesian optimization iteration.

[0081] A training dataset was constructed using 72 temperature and humidity conditions and their corresponding optimal thermal field modulation parameters. The deep kernel learning surrogate model was trained using the stochastic gradient descent method. The loss function was the mean square error between the predicted thermal field modulation parameters and the optimal thermal field modulation parameters. Training was stopped when the loss function value stabilized below 0.01.

[0082] After training, the structural parameters and weight parameters of the deep kernel learning agent model are stored in binary format in a storage unit, which occupies 1800 floating-point numbers, each of which occupies four bytes, for a total of 7200 bytes.

[0083] The deep kernel learning proxy model uses a combination of deep feature extraction and Gaussian process regression to achieve forward inference. After offline training, the model is stored in a memory unit. During runtime, the microcontroller normalizes the collected real-time temperature and humidity data and inputs it into the model. The model first performs nonlinear extraction of temperature and humidity features through a multi-layer fully connected structure, and then performs regression calculation of thermal field parameters based on the extracted feature vectors. It directly outputs the target temperature, heating duration, and time series parameters of each heating point. The entire inference process does not require manual intervention, and the number of model parameters is controlled within 2000 floating-point numbers. The calculation can be completed quickly on the microcontroller. Those skilled in the art can complete the training and deployment of the model based on the existing deep kernel learning framework and the conventional implementation of Gaussian process regression, combined with the training data conditions, model structure, and parameter constraints disclosed in this application.

[0084] When the microcontroller is running, the real-time temperature and humidity data are normalized to the range of zero to one and then input into the deep kernel learning surrogate model. The model forward propagation calculates a 16-dimensional output vector, which is then converted into the target temperature values ​​of the 16 heating points after inverse normalization. The model output also includes heating duration and heating sequence parameters. The heating duration is set to 3 to 10 seconds depending on the stability of the temperature and humidity conditions. The heating sequence parameters are the heating order and overlap relationship of the 16 heating points. When the rate of change of temperature and humidity conditions exceeds 5°C per second or 10% relative humidity per second, the heating sequence is set to staggered heating mode, that is, each heating point is started sequentially with a time interval of 0.5 seconds between the start of adjacent heating points. When the rate of change of temperature and humidity conditions is lower than the above thresholds, the heating sequence is set to synchronous heating mode, that is, all 16 heating points are started simultaneously.

[0085] In subsequent operation, the microcontroller reads real-time temperature and humidity data every 10 seconds. If the temperature changes by more than 3°C or the relative humidity changes by more than 5% compared to the previous operation when the deep kernel learning agent model was called, the deep kernel learning agent model is called again to calculate the new thermal field modulation parameters and update the heating control strategy. If the change in operating conditions does not exceed the above thresholds, the current thermal field modulation parameters are used. This reduces computational overhead while ensuring the accuracy of thermal field modulation.

[0086] The spiking neural network decision engine runs in software within the microcontroller. Its input layer consists of nine neurons, each corresponding to a sensitive unit in the micro-nano composite sensitive array. The input signal received by the input neuron is the voltage response value output by the sensitive unit. This voltage response value is converted into an input current within the input neuron through a current-to-voltage conversion circuit model.

[0087] The hidden layer contains 128 leaky integration firing neurons. These neurons are connected by synapses to form a network topology. The synaptic connections are forward fully connected, meaning that each input neuron is connected to all hidden layer neurons. There are no lateral connections between hidden layer neurons. Each synapse corresponds to an adjustable synaptic weight value, which is stored in a fixed-point format in the storage unit. Each synaptic weight value occupies two bytes. There are a total of 1152 synapses between the 128 hidden layer neurons and the nine input layer neurons, and a total of 384 synapses between the hidden layer neurons and the output layer. The total storage space for synaptic weights is 3072 bytes.

[0088] The membrane potential dynamics model of each leaking integrated firing neuron is iteratively calculated in the microcontroller in a discrete-time manner, with the iteration time step set to 10ms. The membrane potential update formula is executed within each time step. When the accumulated membrane potential exceeds the ignition threshold, the neuron outputs a pulse and resets the membrane potential to the resting potential. The leak term causes the membrane potential to decay proportionally within each time step, with the decay time constant set to 20ms.

[0089] The spiking neural network decision engine operates in an event-driven manner. Specifically, the microcontroller continuously monitors the changes in the input signals of the nine input neurons in the main loop. When the change in any input signal exceeds a set threshold, a pulse calculation is triggered. The set threshold is 2% of the full scale of the input signal. The pulse calculation process involves iteratively calculating the membrane potential and pulse output of the hidden and output layers starting from the input layer according to the time step, for a total of 10 time steps, i.e., a 100ms time window. When there is no change in the input signal, the spiking neural network decision engine pauses the calculation, and the microcontroller enters a low-power standby mode.

[0090] The event-driven event described in this application is defined as the change in the response signal of any sensitive unit in the micro-nano composite sensitive array exceeding a set threshold. This threshold is set according to the noise level and detection sensitivity of the sensitive array, specifically 2% of the full-scale output of the sensitive array, ensuring that calculation is triggered only when there is a change in the effective gas signal, and the system remains in low-power standby when there is no signal change. The event-driven operation mode is as follows: the microcontroller continuously monitors the response signal of the micro-nano composite sensitive array. When the change in the response signal of any channel exceeds the preset threshold, it is determined to be a valid detection event. At this time, the spiking neural network decision engine starts the pulse calculation process to complete feature extraction and early warning judgment. If the response signal does not change significantly or the change does not reach the threshold, it is determined that there is no valid event, the spiking neural network pauses the calculation, and the microcontroller enters a low-power operation mode, which can effectively filter signal fluctuations caused by environmental noise and ensure the accuracy of early warning judgment.

[0091] The synaptic weights of the spiking neural network decision engine were obtained through offline training using a surrogate gradient method. The training dataset was collected in a temperature and humidity test chamber, with conditions ranging from 10°C to 50°C and relative humidity from 20% to 90%. Under each operating condition, response signals of the micro-nano composite sensitive array to five VOCs gases and their mixtures were collected, and the response signals and their corresponding warning level labels were used as training samples. During backpropagation, the surrogate gradient method replaced the non-differentiable pulse output of the leak-integrated firing neuron with a surrogate gradient function. The surrogate gradient function adopted a rectangular function form, with a constant gradient value within a 2mV interval near the ignition threshold of the membrane potential, and a zero gradient value in other intervals. After training, the synaptic weight values ​​were quantized and stored in the storage unit, and the microcontroller no longer performed online updates to the synaptic weights during operation.

[0092] The leaky integral ignition neuron in the spiking neural network decision engine adopts the conventional dynamic membrane potential update logic in the art. The neuron's membrane potential accumulates and increases with the input signal. After reaching the ignition threshold, it outputs a pulse and resets. If the threshold is not reached, the membrane potential decays naturally over time, and the decay rate matches the neuron's intrinsic time constant. Synaptic plasticity adopts the pulse time-dependent plasticity implementation method commonly used in the art. It adaptively adjusts the synaptic connection strength according to the temporal correlation of neuron pulse firing, without the need for online weight updates. The surrogate gradient training method adopts the conventional alternative gradient calculation method of spiking neural networks in the art. The weight optimization is completed during the offline training phase of the model. After training, the weights are fixed and stored. During runtime, only forward pulse calculation is performed.

[0093] The output layer contains three neurons corresponding to the first-level warning, second-level warning, and third-level warning, respectively. The pulse frequency of each output layer neuron represents the probability of occurrence of the warning level. After each pulse calculation, the microcontroller counts the number of pulses of the three output layer neurons in the most recent 100ms time window as the pulse frequency value. This pulse frequency value is used for subsequent warning determination.

[0094] After completing the pulse calculation of the spiking neural network decision engine, the microcontroller obtains the number of pulses of each of the three neurons in the output layer within the most recent 100ms time window. This pulse count is achieved through a sliding window counter. The microcontroller maintains three circular buffers of length 10, each buffer corresponding to one output layer neuron. Each buffer element stores the number of pulses output by the neuron within a 10ms time step. Every 10ms, the microcontroller writes the pulse count of the current time step into the corresponding buffer and overwrites the oldest element. At the same time, it sums the 10 elements to obtain the total number of pulses in the most recent 100ms as the pulse frequency value of the output layer neuron.

[0095] The microcontroller has three pre-stored frequency thresholds corresponding to the warning levels. The first-level warning frequency threshold is set to 5 pulses per 100ms, the second-level warning frequency threshold is set to 15 pulses per 100ms, and the third-level warning frequency threshold is set to 30 pulses per 100ms. Each frequency threshold is determined by taking the average value after multiple tests of the standard gas concentration near the danger threshold.

[0096] The microcontroller maintains a continuous compliance counter for each warning level. The initial value of the counter is zero. At the end of each 200ms time window, a decision logic is executed. The decision logic is as follows: the microcontroller obtains the pulse frequency values ​​of the three output layer neurons at the end of the current time window, compares the pulse frequency of the output layer neurons corresponding to each warning level with the frequency threshold of that warning level, and if the pulse frequency corresponding to a certain warning level reaches or exceeds the frequency threshold of that warning level, the continuous compliance counter for that warning level is incremented by one; otherwise, the counter is cleared to zero. After three consecutive 200ms time windows, i.e., a cumulative decision cycle of 600ms, the microcontroller checks the value of the continuous compliance counter for each warning level. If the value of the continuous compliance counter for a certain warning level reaches 3, the warning for that level is confirmed to be triggered.

[0097] After confirming the triggering of the alarm, the microcontroller drives the local audible and visual alarm through the general-purpose input / output interface. The local audible and visual alarm includes red, green, and blue LEDs and a piezoelectric buzzer. The microcontroller outputs a corresponding pulse width modulation signal to control the color of the LEDs according to the alarm level. Green light is output for level one alarm, yellow light for level two alarm, and red light for level three alarm. At the same time, the microcontroller outputs square wave signals of different frequencies to drive the buzzer. The square wave frequency is 1000Hz for level one alarm, 2000Hz for level two alarm, and 4000Hz for level three alarm. The buzzer drive duty cycle is fixed at 50%.

[0098] The wireless communication unit uses a Bluetooth Low Energy protocol stack. After confirming the triggering of the warning, the microcontroller packages the warning level, the response signal values ​​of the nine channels of the current micro-nano composite sensitive array, the current real-time temperature data, and the real-time humidity data into a data frame. The data frame format is as follows: two bytes for the frame header, one byte for the warning level, two bytes for each of the nine response signals, two bytes for the temperature data, two bytes for the humidity data, and one byte for the checksum, totaling 26 bytes. The microcontroller sends the data frame to the Bluetooth Low Energy chip through the serial peripheral interface. The Bluetooth Low Energy chip broadcasts the data frame in broadcast mode with a broadcast interval of 1 second, and continues to send until the warning is lifted.

[0099] The warning is lifted when the pulse frequency of all output layer neurons is lower than 80% of the corresponding warning level frequency threshold within six consecutive 200ms time windows. The microcontroller then determines that the warning is lifted, stops the output of the audible and visual alarm signal, and stops broadcasting the warning information.

[0100] The microfluidic chip substrate adopts a multilayer polydimethylsiloxane and glass composite structure. After forming microchannels inside the substrate through soft photolithography, a micro diaphragm pump is integrated in the same plane of the substrate. The micro diaphragm pump and the air inlet of the microchannel are connected by a connecting channel with a width of 200μm and a length of 500μm.

[0101] The miniature diaphragm pump adopts a piezoelectric driven structure. Its pump body consists of three layers: the bottom layer is a glass substrate, the middle layer is a pump membrane formed by a polydimethylsiloxane film, and the top layer is a polydimethylsiloxane cover plate with inlet and outlet air channels. A pump cavity is formed between the pump membrane and the glass substrate. The pump cavity has a diameter of 3 mm and a depth of 150 μm. A circular piezoelectric ceramic sheet with a diameter of 4 mm and a thickness of 200 μm is bonded on top of the pump membrane.

[0102] The positive and negative electrodes of the piezoelectric ceramic sheet are connected to the pulse width modulation output port of the microcontroller via flexible circuit board leads. The microcontroller outputs a square wave signal with a frequency of 200Hz and a duty cycle of 50% to drive the piezoelectric ceramic sheet to vibrate. The piezoelectric ceramic sheet drives the pump diaphragm to move up and down reciprocally. When the pump diaphragm moves downward, the pump cavity volume increases and the pressure decreases, drawing external gas into the pump cavity through the air inlet channel. When the pump diaphragm moves upward, the pump cavity volume decreases and the pressure increases, forcing the gas in the pump cavity into the air inlet of the microchannel through the connecting channel.

[0103] A one-way valve is installed at the inlet of the air intake channel. The one-way valve adopts a polydimethylsiloxane cantilever beam structure with a length of 300μm, a width of 200μm, and a thickness of 20μm. When the pump membrane moves downward, the cantilever beam opens upward to allow gas to enter, and when the pump membrane moves upward, the cantilever beam closes to prevent gas backflow.

[0104] The outlet of the microchannel is directly connected to the atmosphere, with an outlet diameter of 300μm. No valves are installed at the outlet, forming an open gas flow path.

[0105] The microcontroller controls the micro diaphragm pump as follows: The microcontroller maintains a sampling state machine, which includes standby, cleaning and sampling states. In standby state, the micro diaphragm pump stops working. In cleaning state, the microcontroller drives the micro diaphragm pump to run at full speed. In sampling state, the microcontroller drives the micro diaphragm pump to run intermittently at integer multiples of the sampling frequency.

[0106] Specifically, when the microcontroller starts a gas detection process, it first enters the cleaning state, driving the micro diaphragm pump to work continuously for 3 seconds to expel residual gas from the microchannel and detection chamber. After cleaning, it enters the sampling state. The microcontroller drives the micro diaphragm pump according to a rhythm that matches the sampling frequency of the micro-nano composite sensitive array. The sampling frequency of the micro-nano composite sensitive array is set to 100Hz, that is, a response signal is collected once every 10ms. The microcontroller divides every 10ms into two 5ms periods. The micro diaphragm pump is driven to run in the first 5ms and stopped in the last 5ms. Fresh gas is drawn into the detection chamber during the operation of the micro diaphragm pump. During the stop period, the gas stays briefly in the detection chamber for the sensitive unit to adsorb and react.

[0107] The micro diaphragm pump operates at a frequency of 200 Hz and a duty cycle of 50%, achieving a flow rate of approximately 5 mL per minute. This flow rate, combined with the 30 μL volume of the detection chamber, ensures that the volume of gas entering the detection chamber every 5 ms cycle is 4 times the volume of the detection chamber, guaranteeing that the gas in the detection chamber is completely replaced before each sampling.

[0108] The microcontroller controls the flow rate of the micro diaphragm pump by adjusting the amplitude of the driving voltage. The driving voltage amplitude ranges from 5V to 24V. The higher the voltage amplitude, the greater the diaphragm amplitude and the higher the flow rate. The microcontroller dynamically adjusts the driving voltage amplitude according to the difference between the current ambient temperature and the standard temperature of 10℃. When the ambient temperature is below 10℃, the driving voltage amplitude is set to 24V. When the ambient temperature is between 10℃ and 30℃, the driving voltage amplitude is set to 12V. When the ambient temperature is above 30℃, the driving voltage amplitude is set to 8V. This compensates for the influence of temperature changes on gas viscosity and keeps the gas volume entering the detection chamber stable under different temperature conditions.

[0109] When the micro diaphragm pump is running, the microcontroller monitors the working status of the pump diaphragm by detecting the feedback current of the piezoelectric ceramic plate. The feedback current is obtained through a sampling resistor connected in series in the piezoelectric ceramic plate drive circuit. The sampling resistor has a resistance of 1Ω. The microcontroller reads the voltage drop across the sampling resistor through an analog-to-digital converter. When the voltage drop deviates from the normal operating range of ±10%, it is determined that the pump diaphragm is malfunctioning. The microcontroller issues a fault indication and reports the fault information through the wireless communication unit.

[0110] The local audible and visual alarm is integrated with the microcontroller on the same circuit board, which measures 15mm × 15mm and is connected to the microfluidic chip substrate via a flexible circuit board. The local audible and visual alarm includes a tri-color light-emitting diode and a piezoelectric buzzer.

[0111] The tri-color LED uses a common cathode package and integrates a red, green, and blue LED chip. The anode of each LED chip is connected to one of the three general-purpose input / output ports of the microcontroller through a current-limiting resistor with a resistance of 150Ω. When the microcontroller outputs a high level to a certain general-purpose input / output port, the corresponding color LED chip lights up.

[0112] The microcontroller outputs corresponding control signals based on the confirmed warning level: when a Level 1 warning is confirmed, the microcontroller outputs a high level only to the general-purpose input / output port corresponding to the green LED, causing the tri-color LED to emit green light; when a Level 2 warning is confirmed, the microcontroller outputs a high level to the general-purpose input / output ports corresponding to both the red and green LEDs, causing the tri-color LED to emit yellow light; when a Level 3 warning is confirmed, the microcontroller outputs a high level only to the general-purpose input / output port corresponding to the red LED, causing the tri-color LED to emit red light.

[0113] The piezoelectric buzzer uses a passive piezoelectric ceramic buzzer with a diameter of 12mm and a thickness of 0.2mm. The positive and negative terminals of the buzzer are connected to the pulse width modulation output port of the microcontroller through wires. The microcontroller drives the buzzer to vibrate and produce sound by outputting square wave signals of different frequencies.

[0114] The microcontroller has three pre-stored frequency values ​​corresponding to three warning levels. The driving square wave frequency corresponding to the first warning level is 1000Hz, the driving square wave frequency corresponding to the second warning level is 2000Hz, and the driving square wave frequency corresponding to the third warning level is 4000Hz. The duty cycle of the driving square wave for all warning levels is fixed at 50%.

[0115] After confirming the alarm is triggered, the microcontroller continuously outputs a square wave signal of the corresponding frequency to drive the buzzer, and at the same time controls the tri-color LED to keep lit. After the alarm is cleared, the square wave output stops and all general-purpose input / output ports are set to low level.

[0116] The wireless communication unit employs either a low-power Bluetooth communication unit or a long-range radio communication unit, with the choice of integration into the control module depending on the application scenario. The low-power Bluetooth communication unit uses a Bluetooth 5.0 protocol stack chip, specifically the nRF52832. This chip connects to the microcontroller via a serial peripheral interface. The microcontroller packages the data to be transmitted and sends it to the Bluetooth chip through the serial peripheral interface. The Bluetooth chip transmits the data in broadcast mode, with a broadcast interval set to 1000ms. Three main broadcast channels are selected, and the transmit power is set to 0dBm. The data transmission distance can reach 50m in open environments.

[0117] The long-range radio communication unit adopts LoRa modulation, with the chip model SX1278, operating frequency set at 470MHz, spreading factor set at 12, bandwidth set at 125kHz, coding rate set at 4 / 5, transmit power set at 20dBm, and data transmission distance up to 3000m in open environments.

[0118] After confirming the triggering of the warning, the microcontroller packages the warning level, nine response signal values, real-time temperature data, and real-time humidity data into a 26-byte data frame and sends it to the wireless communication unit through the serial peripheral interface. The wireless communication unit sends the data frame immediately upon receiving it. After the warning is lifted, the microcontroller sends a warning lift flag data frame. The wireless communication unit automatically enters sleep mode after not receiving a new data frame for three consecutive times, with a sleep current of 2μA.

[0119] The microcontroller determines whether the data has been successfully transmitted by detecting the transmission completion interrupt pin of the wireless communication unit. If three consecutive transmission failures occur, the microcontroller temporarily stores the data frame in a circular buffer in the storage unit and retransmits it after communication is restored. The circular buffer has a capacity of 100 data frames. When the capacity is exceeded, the oldest data frame is discarded according to the first-in-first-out principle.

[0120] The microcontroller runs a main control loop that executes periodically at a frequency of 100Hz, with each cycle lasting 10ms.

[0121] At the start of the main control loop, the microcontroller first uses I... 2 The C-bus interface sends a read command to the digital temperature and humidity sensor. After receiving the command, the digital temperature and humidity sensor returns the real-time temperature and humidity data in the current detection chamber within 200μs. The temperature data is represented in 16-bit binary two's complement form with a resolution of 0.01℃, and the humidity data is represented in 16-bit binary two's complement form with a resolution of 0.05% relative humidity.

[0122] The microcontroller stores the received real-time temperature and humidity data into its internal register. Then, it checks the deviation between the current temperature and humidity conditions corresponding to the current thermal modulation parameters stored in the internal register and the latest acquired temperature and humidity data. Specifically, it calculates the absolute value of the difference between the latest temperature value and the temperature value corresponding to the current thermal modulation parameters, and the absolute value of the difference between the latest humidity value and the humidity value corresponding to the current thermal modulation parameters. If the absolute value of the temperature difference exceeds 3°C or the absolute value of the humidity difference exceeds 5% relative humidity, it determines that the temperature and humidity conditions have changed significantly and the thermal modulation parameters need to be updated. If the absolute value of the temperature difference does not exceed 3°C and the absolute value of the humidity difference does not exceed 5% relative humidity, the current thermal modulation parameters remain unchanged.

[0123] When it is determined that the thermal field modulation parameters need to be updated, the microcontroller uses the latest real-time temperature and humidity data as input parameters and calls the deep kernel learning agent model stored in the storage unit. The deep kernel learning agent model calculates and outputs the target temperature value, heating duration and heating timing parameters of 16 heating points through forward propagation. The microcontroller writes the calculated new thermal field modulation parameters into the heating control register and overwrites the original parameters.

[0124] After updating the thermal field modulation parameters, the microcontroller starts the micro heating array execution program. The execution program determines the start-up order and overlap relationship of the 16 heating points according to the heating timing parameters in the new thermal field modulation parameters. When the heating timing parameters are set to synchronous heating mode, the microcontroller simultaneously writes the duty cycle corresponding to the target temperature value to the pulse width modulation channel corresponding to the 16 heating points. Each heating point independently performs closed-loop temperature control until the target temperature value is reached. When the heating timing parameters are set to staggered heating mode, the microcontroller starts each heating point in sequence according to the order specified in the heating timing parameters. After each heating point is started, the next heating point is started after 0.5 seconds.

[0125] After starting the micro heating array, the microcontroller sets a waiting time according to the heating duration parameter in the thermal field modulation parameters. During the waiting time, the microcontroller continuously monitors the actual temperature feedback of each heating point to ensure that the actual temperature of all heating points reaches the target temperature value within ±5℃.

[0126] After the waiting time ends, the microcontroller determines that the heating array has reached a stable working state and then begins to collect the response signal of the micro-nano composite sensing array. During the acquisition process, the output voltage values ​​of the nine sensing units are read sequentially through the analog-to-digital converter. The sampling rate of the analog-to-digital converter is set to 1000Hz per channel. Each channel is sampled 16 times continuously, and the arithmetic mean is taken as the response signal value of that channel to eliminate noise interference.

[0127] After the microcontroller completes the acquisition of response signals from nine channels, it stores the acquired response signal values ​​into the data buffer and returns to the start of the main control loop to begin processing the next 10ms cycle.

[0128] If the temperature and humidity conditions at the beginning of a certain cycle determine that no update of the thermal field modulation parameters is required, the microcontroller skips the calling steps of the deep kernel learning proxy model and the reconfiguration steps of the micro heating array, directly uses the currently executing thermal field modulation parameters, and waits for the remaining time set by the heating duration parameter before acquiring the response signal.

[0129] Throughout the process, the microcontroller completes a closed loop in each cycle, including temperature and humidity reading, operating condition determination, thermal field modulation parameter update, heating array control, and response signal acquisition. This ensures that the acquisition of each response signal is completed under the optimal thermal field conditions corresponding to the current temperature and humidity conditions, achieving time synchronization between temperature and humidity compensation and signal acquisition.

[0130] After the system is powered on, the microcontroller first executes the initialization program. The initialization program completes the register configuration of each peripheral module, loads the parameter data of the deep kernel learning agent model and the synaptic weight data of the spiking neural network decision engine from the storage unit, and sets the initial target temperature of each heating point of the micro heating array to 250℃.

[0131] After initialization, the microcontroller enters the main loop, which executes the following steps in sequence.

[0132] Step 1: The microcontroller sends a measurement command to the digital temperature and humidity sensor via the I²C bus, waits 200μs, and then reads the 16-bit temperature data and 16-bit humidity data. The temperature data is converted into the actual temperature value, and the humidity data is converted into the actual relative humidity value. The converted real-time temperature and humidity values ​​are then stored in memory variables.

[0133] Step Two: The microcontroller reads the temperature and humidity values ​​corresponding to the current thermal field modulation parameters stored in memory. It compares the real-time temperature value with the temperature value corresponding to the current thermal field modulation parameters, and the real-time humidity value with the humidity value corresponding to the current thermal field modulation parameters. If the absolute value of the temperature difference exceeds 3℃ or the absolute value of the humidity difference exceeds 5% relative humidity, the microcontroller inputs the real-time temperature and humidity values ​​into the deep kernel learning proxy model. The deep kernel learning proxy model sequentially executes matrix multiplication and modified linear unit activation function operations in three fully connected layers, outputting a 128-dimensional feature vector to the high-resolution model. In the Gaussian process regression part, a 16-dimensional output vector is calculated based on the radial basis function covariance matrix. After inverse normalization, the 16-dimensional output vector is converted into the target temperature values ​​of 16 heating points. At the same time, the model output also includes heating duration parameters and heating timing parameters. The microcontroller combines the target temperature value, heating duration, and heating timing into new thermal field modulation parameters and writes them into the heating control register. If the absolute value of the temperature difference does not exceed 3℃ and the absolute value of the humidity difference does not exceed 5% relative humidity, the microcontroller skips the call to the deep kernel learning proxy model and maintains the current thermal field modulation parameters unchanged.

[0134] Step 3: The microcontroller controls the micro-heating array to perform differentiated heating on the micro-nano composite sensitive array according to the thermal field modulation parameters in the heating control register. Specifically, the microcontroller reads the heating timing parameters in the thermal field modulation parameters. If the heating timing parameters are in synchronous heating mode, the microcontroller simultaneously writes the duty cycle corresponding to the target temperature value to the pulse width modulation channel of the 16 heating points. The thin-film thermocouple temperature sensor of each heating point provides real-time feedback on the actual temperature. The microcontroller uses a proportional-integral-derivative algorithm to adjust the duty cycle in a closed loop to make the actual temperature approach the target temperature value. If the heating timing parameters are in staggered heating mode, the microcontroller starts each heating point in sequence according to the order specified in the parameters. After each heating point is started, the next heating point is started after 0.5 seconds, until all 16 heating points are started.

[0135] After starting heating, the microcontroller waits according to the heating duration parameter in the thermal field modulation parameters. The heating duration parameter ranges from 3 to 10 seconds. During the waiting period, the microcontroller continuously monitors the actual temperature feedback of each heating point to ensure that the actual temperature of all heating points is stable within the target temperature value ±5℃.

[0136] Step 4: After the heating waiting time ends, the microcontroller starts the analog-to-digital converter to collect the response signal of the micro-nano composite sensitive array. The analog-to-digital converter switches nine input channels in sequence, and each channel continuously collects 16 voltage values ​​at a sampling rate of 1000Hz. The microcontroller calculates the arithmetic mean of the 16 voltage values ​​as the response signal value of that channel. After all nine channels have been collected, the nine response signal values ​​are stored in the data buffer.

[0137] Step 5: The microcontroller inputs the nine response signal values ​​from the data buffer into the spiking neural network decision engine. The spiking neural network decision engine runs in an event-driven manner. The microcontroller detects the difference between the nine response signal values ​​and the corresponding values ​​of the previous cycle. If any difference exceeds 2% of the full scale, pulse calculation is triggered. The pulse calculation iterates for 10 steps with a time step of 10ms. Within each time step, the 128 leaky integrator neurons in the hidden layer update the membrane potential according to the input layer current and synaptic weights. When the membrane potential exceeds the ignition threshold, a pulse is output and the membrane potential is reset. The three neurons in the output layer update the membrane potential according to the pulse input from the hidden layer and the synaptic weights and output a pulse. After the iteration is completed, the number of pulses of the three neurons in the output layer within 10 time steps is counted as the pulse frequency value. If all differences do not exceed 2% of the full scale, the microcontroller skips the pulse calculation and directly uses the pulse frequency value calculated in the previous cycle.

[0138] Step Six: The microcontroller compares the pulse frequency values ​​of the three neurons in the output layer with the pre-stored frequency thresholds in the storage unit. The first-level warning frequency threshold is 5 pulses per 100ms, the second-level warning frequency threshold is 15 pulses per 100ms, and the third-level warning frequency threshold is 30 pulses per 100ms. The microcontroller maintains a continuous pass counter for each warning level. If the pulse frequency corresponding to a certain level reaches or exceeds the threshold of that level at the end of every 200ms time window, the counter for that level is incremented by one; otherwise, it is reset to zero. When the threshold is reached in three consecutive 200ms time windows, the warning for that level is confirmed to be triggered.

[0139] After confirming the triggering of the warning, the microcontroller drives the tri-color LED and the piezoelectric buzzer through the general-purpose input / output port. The first-level warning outputs a green light and a 1000Hz square wave, the second-level warning outputs a yellow light and a 2000Hz square wave, and the third-level warning outputs a red light and a 4000Hz square wave. At the same time, the microcontroller packages the warning level, nine response signal values, real-time temperature value, and real-time humidity value into a 26-byte data frame and sends it to the wireless communication unit through the serial peripheral interface. The wireless communication unit uses Bluetooth Low Energy protocol stack or long-range radio modulation to send the data frame outward, and the broadcast interval is set to 1000ms.

[0140] After completing step six, the microcontroller returns to step one and begins the next cycle of processing, thereby achieving continuous and uninterrupted gas detection, temperature and humidity compensation, signal processing, and graded early warning output.

[0141] This invention proposes a miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation. It integrates a microfluidic chip, a micro-nano composite sensing array, a micro heating array, a digital temperature and humidity sensor, and a control module into a single unit, achieving both handheld portability and long-term continuous operation. By encapsulating the digital temperature and humidity sensor and the micro-nano composite sensing array within the same detection cavity and maintaining synchronized sampling frequencies, the time misalignment between temperature and humidity measurements and the gas response signal is eliminated.

[0142] Based on this, the microcontroller first acquires real-time temperature and humidity data before collecting response signals, calls the deep kernel learning proxy model embedded in the storage unit to calculate thermal field modulation parameters, and controls the micro heating array to perform differentiated heating on each sensitive unit, so that the operating temperature of each sensitive unit actively avoids the sensitive range under the current temperature and humidity conditions. The temperature and humidity compensation is transformed from traditional signal post-processing to active environmental decoupling before response. Within the set temperature range and relative humidity range, the absolute value of the gas concentration detection error can be controlled, and the compensation response has no additional delay.

[0143] This invention further integrates a spiking neural network decision engine and a hierarchical early warning mechanism at the sensor end. The microcontroller directly inputs the multidimensional response signal output by the micro-nano composite sensing array into the spiking neural network, which completes feature extraction and early warning level determination in an event-driven manner. There is no need to upload the data to the host computer for processing, thus realizing independent decision-making and real-time response at the sensing end.

[0144] The spiking neural network employs leaky integration of firing neurons and synaptic plasticity mechanisms, combined with continuous time window integral decision logic, to effectively filter out instantaneous interference caused by environmental fluctuations, thus improving early warning accuracy. The local audible and visual alarm works in conjunction with the wireless communication unit, simultaneously outputting on-site warning signals and remotely transmitted information upon alarm triggering, forming a complete edge-end intelligent sensing and early warning closed loop. Compared to traditional detection systems that rely on host computers for post-correction of temperature and humidity and lack built-in early warning capabilities, this invention achieves improvements in miniaturization integration, environmental adaptability, real-time decision-making, and early warning reliability.

[0145] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0146] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation, characterized in that, include: Microfluidic chip substrate; Microchannels are disposed inside the microfluidic chip substrate, and the microchannels have an air inlet and an air outlet; A micro-nano composite sensing array integrated within the microchannel, the micro-nano composite sensing array comprising micro-nano sensors of at least two different types of sensing materials for outputting response signals; A digital temperature and humidity sensor integrated within the microchannel is used to output real-time temperature and humidity data. A micro heating array is disposed on the microfluidic chip substrate, the micro heating array is located below the micro-nano composite sensing array, and forms thermal coupling with the micro-nano composite sensing array; A control module electrically connected to the micro heating array, the micro-nano composite sensitive array, and the digital temperature and humidity sensor, the control module comprising a microcontroller, a storage unit, and a wireless communication unit; The storage unit contains a temperature and humidity-thermal field coupling proxy model. The microcontroller is configured to call the temperature and humidity-thermal field coupling proxy model to calculate thermal field modulation parameters based on the real-time temperature data and the real-time humidity data, and to control the micro heating array to perform differentiated heating on each micro-nano sensor based on the thermal field modulation parameters. The microcontroller is also configured to acquire the response signal of the micro-nano composite sensitive array, input the response signal into the spiking neural network decision engine, and the spiking neural network decision engine outputs a graded early warning signal based on the input signal. The microcontroller drives the wireless communication unit to send warning information based on the graded warning signal.

2. The miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation according to claim 1, characterized in that: The micro-nano composite sensing array includes multiple sensing units arranged in an array configuration, and the multiple sensing units include at least two of the following: metal oxide semiconductor sensing units, conductive polymer sensing units, and catalytic combustion sensing units. Each sensitive unit has a three-dimensional nanodendritic structure, which is grown on the electrode surface inside the microchannel by electrodeposition. The width of the microchannel is 100μm to 300μm, and a detection cavity is formed inside the microchannel, with a volume of 20μL to 50μL. The digital temperature and humidity sensor and the micro-nano composite sensitive array are encapsulated in the same detection cavity, and the sampling frequency of the digital temperature and humidity sensor is synchronized with the response signal acquisition frequency of the micro-nano composite sensitive array.

3. The miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation according to claim 2, characterized in that: The micro heating array is an independently controllable heating point array, which includes at least 16 heating points, each heating point corresponding to a sensitive unit or a group of sensitive units; The microcontroller controls the heating temperature of the micro heating array to switch as needed within a set temperature range, where the lower limit of the set temperature range is 150°C and the upper limit is 400°C. The thermal field modulation parameters include the target temperature value, heating duration and heating sequence of each heating point. The microcontroller updates the thermal field modulation parameters at a frequency of 50Hz or higher and controls the micro heating array to execute.

4. The miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation according to claim 3, characterized in that: The temperature and humidity-thermal field coupled proxy model is a deep kernel learning proxy model, which is obtained through offline training. The training data includes the optimal thermal field parameters under multiple temperature and humidity combinations. The deep kernel learning proxy model is embedded in the storage unit. When the microcontroller runs, it calls the deep kernel learning proxy model to perform forward inference. The input is real-time temperature data and real-time humidity data, and the output is thermal field modulation parameters. The number of parameters in the deep kernel learning agent model does not exceed 2000 floating-point numbers.

5. A miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation according to claim 4, characterized in that: The spiking neural network decision engine includes an input layer, a hidden layer, and an output layer; The input layer corresponds to each sensitive unit of the micro-nano composite sensitive array and is used to receive the response signal; The hidden layer includes leaky integrated firing neurons, which are connected to each other via synaptic plasticity. The output layer includes three neurons, corresponding to the first-level warning, the second-level warning, and the third-level warning, respectively. The microcontroller is configured to run the spiking neural network decision engine, which operates in an event-driven manner and triggers pulse calculation only when the change in the response signal exceeds a set threshold.

6. The miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation according to claim 5, characterized in that: The microcontroller is also configured to perform time window integration on the pulse frequency of the output layer neurons of the spiking neural network decision engine; The microcontroller determines whether the pulse frequency of the output layer neurons exceeds the frequency threshold of the corresponding warning level within three consecutive time windows. When the frequency threshold is exceeded in three consecutive time windows, the corresponding warning level is triggered. The microcontroller drives the local audible and visual alarm to output the corresponding level of audible and visual signal based on the confirmed warning level, and sends the warning information to the external terminal through the wireless communication unit.

7. A miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation according to claim 6, characterized in that: The microfluidic chip substrate also integrates a micro diaphragm pump, which is connected to the air inlet of the microchannel and is used to actively draw external gas into the microchannel. The outlet of the microchannel is connected to the atmosphere, forming a gas flow path; The microcontroller is configured to control the start-up, shutdown, and flow rate of the micro diaphragm pump, the flow rate of which is matched with the sampling frequency of the micro-nano composite sensing array.

8. A miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation according to claim 7, characterized in that: The control module also includes a local audible and visual alarm, which is electrically connected to the microcontroller. The local sound and light alarm includes a multi-color light-emitting diode and a buzzer. The microcontroller controls the multi-color light-emitting diode to display the corresponding color according to the graded warning signal, and controls the buzzer to emit a prompt sound of the corresponding frequency. The wireless communication unit is a low-power Bluetooth communication unit or a long-range radio communication unit.

9. A miniaturized VOCs gas sensor and early warning system based on temperature and humidity compensation according to claim 8, characterized in that: The microcontroller is configured to acquire real-time temperature and humidity data from the digital temperature and humidity sensor before acquiring the response signal of the micro-nano composite sensitive array. Based on the real-time temperature and humidity data, the microcontroller calls the temperature-humidity-thermal field coupling proxy model to calculate thermal field modulation parameters, controls the micro heating array to execute the thermal field modulation parameters, and then acquires the response signal of the micro-nano composite sensitive array.

10. A miniaturized VOCs gas detection method based on temperature and humidity compensation, applied to the system described in any one of claims 1 to 9, characterized in that, Includes the following steps: Step 1: Obtain real-time temperature and humidity data; Step 2: Input the real-time temperature data and the real-time humidity data into the temperature-humidity-thermal field coupling proxy model stored in the storage unit, and calculate the thermal field modulation parameters; Step 3: Control the micro heating array to perform differentiated heating on the micro-nano composite sensitive array according to the thermal field modulation parameters; Step 4: Acquire the response signal output by the micro-nano composite sensing array; Step 5: Input the response signal into the spiking neural network decision engine, and the spiking neural network decision engine outputs a graded early warning signal; Step Six: Drive the wireless communication unit to send warning information according to the graded warning signal, and drive the local sound and light alarm to output the corresponding level of sound and light signal.

Images