Intelligent self-adaptive flame ionization detector and control method thereof

Abstract

The invention provides an intelligent self-adaptive flame ionization detector which comprises an instrument, an ionization part and a pretreatment part are mounted on the instrument, the ionization part is used for detecting sample airflow, and the pretreatment part is used for pretreating the sample airflow; a gas path unit is mounted on the instrument and is used for injecting a sample gas flow into the pretreatment part, and then the sample gas flow is injected into the ionization part for ionization. The gas path unit quantitatively injects the sample gas into the pretreatment part, then injects the sample gas into the ionization part, preheats the sample gas through the pretreatment part, and then injects the sample gas into the ionization part for subsequent ionization, and meanwhile, the ionization part can heat the sample gas, so that a double-temperature-zone structure is formed; due to the double-temperature-zone independent temperature control design, differential precise adaptation of different boiling range components is achieved, and the technical bottleneck that all-component analysis cannot be considered in a traditional single-temperature-zone'compromise temperature 'is broken through.

Description

Technical Field

[0001] This invention relates to the field of chemical analysis instrument technology, and in particular to an intelligent adaptive flame ionization detector and its control method. Background Technology

[0002] Flame ionization detectors (FIDs) are core detection equipment in fields such as environmental monitoring, industrial safety, and emergency response. (See attached image.) Figure 5 Includes 101 - combustion chamber; 102 - flame; 103 - collecting electrode; 104 - high voltage. As applications expand to outdoor and mobile scenarios, existing portable FIDs face three major technical challenges: Insufficient analytical adaptability: Traditional FID employs a single-temperature zone design, and its fixed operating temperature is a compromise optimized for specific types of compounds. When faced with complex mixtures containing both low-boiling-point components (such as methane and ethane) and high-boiling-point components (such as benzene compounds and polycyclic aromatic hydrocarbons), it is difficult to achieve optimal simultaneous analysis of all components, leading to biases in quantitative results.

[0003] Limited battery life: The heating and temperature control module of the FID is the main source of power consumption in portable devices, typically accounting for more than 50% of the total power consumption. The currently widely used PID (proportional-integral-derivative) control method, due to its inherent response lag and frequent power fine-tuning to suppress temperature fluctuations, results in significant steady-state energy waste, severely limiting the instrument's working time on a single charge.

[0004] Severe environmental interference: The on-site environment is complex and variable. Factors such as atmospheric pressure, ambient temperature, and unavoidable physical vibrations during operation can severely interfere with the extremely weak ion current signal of FID (usually at the pA level), manifesting as baseline drift, random noise, and spurious peaks. This not only reduces the accuracy and reliability of the detection results but also worsens the instrument's detection limit.

[0005] Therefore, a device is needed to solve the above problems. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the defects of the existing technology. The present invention proposes an intelligent adaptive flame ionization detector and its control method.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: An intelligent adaptive flame ionization detector includes: an instrument equipped with an ionization section and a pretreatment section, wherein the ionization section detects a sample gas flow, and the pretreatment section pretreats the sample gas flow; a gas path unit is installed on the instrument for injecting the sample gas flow into the pretreatment section, after which the sample gas flow is injected into the ionization section for ionization; a central controller installed on the instrument, wherein the central controller stores an AI model and has a built-in algorithm; a memory is installed in the central controller for storing predictive thermal model parameters and network weights of the AI ​​correction model; the central controller is configured to perform the following collaborative tasks through the built-in software algorithm: independently controlling the temperature of the pretreatment zone and the ionization zone according to a preset analysis mode using a predictive low-power isothermal control method; and employing a machine learning-based real-time signal correction method to dynamically compensate for interference in the main analysis signal by real-time fusion of the main analysis signal output by the detection module and the signal from the auxiliary environmental sensor, and outputting a corrected signal.

[0008] Preferably, the predictive low-power constant temperature control method includes the following steps: establishing a predictive thermal model describing the characteristics of the temperature control zone detection module, such as heat capacity and heat dissipation coefficient; during operation, periodically predicting the compensation energy required to maintain the target temperature in the next time period based on the current temperature and the thermal model; and supplying the compensation energy in the form of energy pulses at the beginning of the period using high-frequency pulse width modulation.

[0009] Preferably, the machine learning model used in the real-time signal correction method based on machine learning is a pre-trained recurrent neural network or a variant thereof (long short-term memory network or gated recurrent unit); the model is trained to learn the nonlinear mapping relationship between a time series input containing "[disturbed main analysis signal + multiple auxiliary environmental signals]" and an output of "undisturbed ideal main analysis signal".

[0010] Preferably, the central controller is further equipped with a power management unit for power supply; it also includes a user interface unit installed on the central controller, the user interface unit being connected to an external display screen for displaying results.

[0011] Preferably, the instrument also includes a triaxial accelerometer mounted on it, wherein the triaxial accelerometer is a TDKMPU-6050 six-axis motion sensor used to collect motion data of the instrument; and an atmospheric pressure sensor mounted on it to detect the atmospheric pressure value of the environment, wherein the atmospheric pressure sensor is a Bosch BMP388 high-precision pressure sensor.

[0012] Preferably, the ionization section includes an ionization chamber for ionization; a first heating element is installed outside the ionization chamber; a collecting electrode is installed inside the ionization chamber; a first temperature sensor is installed outside the first heating element for detecting the heating temperature of the first heating element; and a first heat insulation layer is installed outside the first heating element to prevent heat from the first heating element from diffusing outward.

[0013] Preferably, the pretreatment section includes a second heating element; it also includes a second temperature sensor installed outside the second heating element; and a second heat insulation layer is installed outside the second heating element to prevent the heat of the second heating element from dissipating outward.

[0014] Preferably, the gas path unit includes a sampling pump and a proportional valve, used to quantitatively inject sample gas into the pretreatment section and then into the ionization section. The sample gas is preheated by the pretreatment section and then injected into the ionization section for subsequent ionization. At the same time, the ionization section can also heat the sample gas, thus forming a dual-temperature zone structure.

[0015] Preferably, the first heating element and the second heating element constitute a heating system, and both the first heating element and the second heating element are flexible polyimide film heaters, with a temperature zone formed inside the first heating element and the second heating element; the sample gas outputs an FID signal value after being ionized.

[0016] A method for controlling an intelligent adaptive flame ionization detector, characterized by comprising the following steps: Step 1: Predictive Low-Power Precision Temperature Control The goal of this step is to efficiently and stably establish and maintain the temperature in both temperature zones; The thermal model establishment is also known as factory calibration: In the factory calibration procedure, the instrument automatically performs a "thermal characteristic self-learning" process; the central controller performs step heating and natural cooling on two temperature zones with different high-frequency pulse width modulation duty cycles and records the temperature change curves; based on these data, the system identification algorithm (such as the least squares method) is used to determine the thermal model parameters of each temperature zone, which can be simplified as: T(t+1)=T(t)+k1*P_in(t)-k2*(T(t)-T_ambient); the calculated thermal gain coefficient k1 and heat dissipation coefficient k2 are stored in the memory; Runtime control: When the user selects an analysis method (such as "high boiling point mode", T1=250°C, T2=200°C), the central controller starts a predictive control cycle (5-second cycle): Prediction: At the beginning of the cycle, the central controller reads the current temperature T(t) and uses the thermal model to predict the temperature drop ΔT_loss caused only by natural heat dissipation in the next 5 seconds; Calculate the compensation energy: Calculate the total energy E_needed required to restore the temperature to the target value and offset the expected heat loss; Pulse heating: The central controller calculates the high-frequency pulse width modulation pulse width required to supply energy to E_needed, and outputs the pulse at 100% power for the first few hundred milliseconds of the 5-second cycle, and then turns off the power for the rest of the cycle; Intelligent standby: If the instrument is not used for more than 5 minutes, it will automatically enter standby mode and reduce the temperature to a low maintenance temperature (such as 90°C) with extremely low power consumption; when the user operates it again, the system uses the thermal model to calculate the fastest heating path and performs full-power pulse heating to restore the working temperature in a short time (such as within 60 seconds). Step Two: Real-time Analysis and AI Intelligent Correction Multidimensional data stream acquisition: After the temperature zone stabilizes, the analysis begins; the central controller synchronously acquires a data frame at a frequency of 10Hz, which includes: [FID signal value, raw ADC reading, atmospheric pressure sensor reading, accelerometer X-axis reading, accelerometer Y-axis reading, accelerometer Z-axis reading]; AI model training (completed offline): Data acquisition: During the research and development phase, the instrument was placed in a vibration table and an environmental chamber. Under the conditions of zero-point gas and standard gas of different concentrations, various controllable vibrations (such as the spectrum of simulated walking and vehicle bumps) and pressure changes were applied, and millions of the above data frames were recorded. At the same time, a high-precision benchtop instrument was used to record "real" interference-free signals as tags. Model architecture: It adopts a neural network architecture containing two LSTM layers (32 hidden units per layer) and a fully connected output layer; Training: The model is trained using the collected data, with Adam as the optimizer and mean squared error (MSE) as the loss function; the goal is to enable the model to learn to accurately predict the true signal value from input data frames with interference. Deployment: After training, the model is quantized using TensorFlowLite and its network weights are stored in the memory of the central controller; Real-time inference and correction: When used in the field, the central controller takes each acquired data frame as the latest input of a time series window and sends it to the deployed AI model; the model performs forward inference once, and its output is a "clean" signal that has been compensated for pressure and vibration interference in real time. Step 3: Results Output and Effects The high signal-to-noise ratio signal after AI correction has an extremely stable baseline compared to the original signal which is full of glitches and drift, truly reflecting the changes in sample concentration, and is ultimately displayed on the screen of the user interface unit.

[0017] Compared with the prior art, the beneficial effects of the present invention include: 1. The gas path unit quantitatively injects the sample gas into the pretreatment section and then into the ionization section. The sample gas is preheated in the pretreatment section and then injected into the ionization section for subsequent ionization. At the same time, the ionization section can also heat the sample gas, thus forming a dual-temperature zone structure. The independent temperature control design of the dual temperature zones enables differentiated and precise adaptation to components with different boiling ranges, breaking the technical bottleneck of the traditional single-temperature zone "compromise temperature" which cannot take into account the analysis of all components.

[0018] 2. The central controller controls the temperature of the temperature zone. Compared with the traditional FID heating and temperature control module, the heating is intelligent, which prevents energy waste caused by long-term constant temperature and improves the instrument's endurance.

[0019] 3. By training an AI model and simulating the vehicle's operating environment, it learns to accurately predict the true signal value from input data frames with interference. Attached Figure Description

[0020] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein: Figure 1 This is a schematic diagram of the structure of an intelligent adaptive flame ionization detector proposed in this invention; Figure 2 This is an electronic control block diagram of an intelligent adaptive flame ionization detector proposed in this invention; Figure 3 This is a schematic diagram of the integrated process of predictive low-power isothermal control and AI signal correction working together in an intelligent adaptive flame ionization detector proposed in this invention. Figure 4 This is a simulation comparison diagram of the original FID signal and the signal corrected by the method of this invention under the same interference conditions of an intelligent adaptive flame ionization detector proposed in this invention. Figure 5 This is a schematic diagram of the structure of a single-temperature zone FID in the prior art; The following are the labeling elements in the diagram: 1. Instrument; 2. Ionization section; 21. Ionization chamber; 22. First temperature sensor; 23. Collector electrode; 24. First heating element; 25. First insulation layer; 3. Pretreatment section; 31. Second temperature sensor; 32. Second insulation layer; 33. Second heating element; 4. Gas path unit; 5. Central controller; 51. Memory; 52. Power management unit; 53. User interface unit; 6. Atmospheric pressure sensor; 7. Triaxial accelerometer. Detailed Implementation

[0021] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0022] According to one embodiment of the present invention, Figures 1-5 As shown.

[0023] A smart adaptive flame ionization detector specifically includes the following core components and structure: Instrument 1 serves as the main frame of the entire detector, on which the ionization unit 2 and the pretreatment unit 3 are securely mounted with bolts. The ionization unit 2 detects the sample gas flow, while the pretreatment unit 3 is responsible for pretreating the sample gas flow. The gas path unit 4 is also fixed to instrument 1 with bolts and is responsible for quantitatively injecting the sample gas flow into the pretreatment unit 3. After pretreatment, the sample gas flow is then introduced into the ionization unit 2 for ionization treatment.

[0024] The central controller 5 is securely fixed to the instrument 1 with bolts. The central controller 5 not only stores advanced AI models but also incorporates efficient algorithms. Its internal memory 51 is used to store predictive thermal model parameters and AI correction model network weights. The software algorithms embedded within the central controller 5 can efficiently execute a series of collaborative tasks: on the one hand, it employs a predictive low-power isothermal control method to independently and precisely control the temperature of the pretreatment and ionization zones according to a preset analysis mode, ensuring the detection process is conducted in a stable temperature environment; on the other hand, it uses a machine learning-based real-time signal correction method to dynamically compensate for interference in the main analysis signal by fusing the main analysis signal output from the detection module with the signals from the auxiliary environmental sensors, ultimately outputting an accurate and reliable corrected signal.

[0025] In this embodiment, the predictive low-power isostatic control method has rigorous steps: First, a predictive thermal model is constructed that can accurately describe the characteristics of the temperature control zone detection module, such as heat capacity and heat dissipation coefficient; during the operation of the device, the compensation energy required to maintain the target temperature in the next time period is predicted periodically based on the current temperature and the thermal model; then, the compensation energy is precisely supplied in the form of energy pulses at the beginning of the period using high-frequency pulse width modulation, thereby achieving low-power and accurate isostatic control.

[0026] The machine learning model used in real-time signal correction methods is a recurrent neural network (RNN) or its variants, such as Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU), that has been pre-trained on a large amount of data. This carefully trained model can learn a complex nonlinear mapping relationship between a time-series input containing "[disturbed main analysis signal + multiple auxiliary environmental signals]" and an output containing "disturbed ideal main analysis signal," thereby achieving accurate signal correction.

[0027] The central controller 5 is also equipped with a power management unit 52 to provide a continuous and stable power supply. Meanwhile, the user interface unit 53, installed on the central controller 5, connects to an external display screen, allowing users to easily and intuitively view the test results.

[0028] To more comprehensively perceive equipment status and environmental information, multiple sensors are also fixed in this embodiment with bolts. The triaxial accelerometer 7 uses a TDK MPU-6050 six-axis motion sensor, which can accurately collect motion data of instrument 1 and provide support for monitoring the operating status of the equipment; the atmospheric pressure sensor 6 uses a Bosch BMP388 high-precision pressure sensor, which can accurately detect the atmospheric pressure value of the environment and provide more comprehensive environmental parameters for the detection process.

[0029] The ionization unit 2 includes an ionization chamber 21, a core component for ionization. A first heating element 24 is installed outside the ionization chamber 21 to provide a suitable temperature environment for the ionization process. A collecting electrode 23 is also installed inside the ionization chamber 21 to collect the signals generated by ionization. A first temperature sensor 22 is installed outside the first heating element 24 to detect the heating temperature of the first heating element 24 in real time and ensure the accuracy of temperature control. In addition, a first heat insulation layer 25 is provided outside the first heating element 24 to effectively prevent the heat of the first heating element 24 from spreading outward and to ensure the temperature stability of the ionization unit 2.

[0030] The pretreatment unit 3 also has a reasonable structural design, including a second heating element 33 to provide heat for the pretreatment of sample gas; a second temperature sensor 31 is installed on the outside of the second heating element 33 to monitor temperature changes in real time; at the same time, a second heat insulation layer 32 is also provided on the outside of the second heating element 33 to prevent heat loss and ensure the temperature stability of the pretreatment unit 3.

[0031] The gas path unit 4 includes a sampling pump and a proportional valve, which can accurately and quantitatively inject sample gas into the pretreatment section 3. After preheating in the pretreatment section 3, the sample gas is then injected into the ionization section 2 for subsequent ionization. This design creates a dual-temperature zone structure between the ionization section 2 and the pretreatment section 3. The independent temperature control design of the dual temperature zones enables precise adaptation to the differences in components with different boiling ranges, successfully breaking through the technical bottleneck of the traditional single-temperature zone "compromise temperature" which cannot take into account the analysis of all components, and greatly improving the accuracy and comprehensiveness of the detection.

[0032] The first heating element 24 and the second heating element 33 together form the heating system. Both are flexible polyimide thin-film heaters, a material with good flexibility and heating performance. A temperature zone is formed inside the first heating element 24 and the second heating element 33, providing a suitable temperature environment for the processing of the sample gas. After the sample gas is ionized, the output is an FID signal value, providing a basis for subsequent data analysis and processing.

[0033] A method for controlling an intelligent adaptive flame ionization detector specifically includes the following steps: Step 1: Predictive Low-Power Precision Temperature Control The goal of this step is to efficiently and stably establish and maintain the temperature in both temperature zones.

[0034] Thermal Model Establishment (Factory Calibration): During the factory calibration procedure, Instrument 1 automatically executes a "thermal characteristic self-learning" process. The central controller 5 performs step heating and natural cooling on two temperature zones with different high-frequency pulse width modulation duty cycles, recording the temperature change curves. Based on this data, a system identification algorithm (such as the least squares method) is used to determine the thermal model parameters for each temperature zone, which can be simplified as: T(t+1) = T(t) + k1*P_in(t) - k2*(T(t) - T_ambient). The calculated thermal gain coefficient k1 and heat dissipation coefficient k2 are stored in memory 51.

[0035] Runtime control: When the user selects an analysis method (such as "high boiling point mode", T1=250°C, T2=200°C), the central controller 5 initiates a predictive control cycle (5-second cycle): Prediction: At the start of the cycle, the central controller 5 reads the current temperature T(t) and uses the thermal model to predict the temperature drop ΔT_loss caused only by natural heat dissipation in the next 5 seconds.

[0036] Calculate the compensation energy: Calculate the total energy E_needed required to restore the temperature to the target value and offset the expected heat loss.

[0037] Pulse heating: The central controller 5 calculates the high-frequency pulse width modulation pulse width required to supply energy to E_needed, and outputs the pulse at 100% power for the first few hundred milliseconds of the 5-second cycle, and then turns off the power for the rest of the cycle.

[0038] Intelligent standby: If the instrument 1 remains inactive for more than 5 minutes, it will automatically enter standby mode, lowering the temperature to a low maintenance temperature (e.g., 90°C) with extremely low power consumption. When the user operates the instrument again, the system uses a thermal model to calculate the fastest heating path and performs full-power pulse heating to restore the operating temperature in a short time (e.g., within 60 seconds).

[0039] Step Two: Real-time Analysis and AI Intelligent Correction Multidimensional data stream acquisition: After the temperature zone stabilizes, the analysis begins. The central controller 5 synchronously acquires a data frame at a frequency of 10Hz. This data frame contains: [FID signal value, raw ADC reading, atmospheric pressure sensor reading, accelerometer X-axis reading, accelerometer Y-axis reading, accelerometer Z-axis reading].

[0040] AI model training (completed offline): Data Acquisition: During the R&D phase, Instrument 1 was placed in a vibration table and an environmental chamber. With zero-point gas and standard gases of different concentrations introduced, various controllable vibrations (such as the spectrum of simulated walking and vehicle bumps) and pressure changes were applied, and millions of the aforementioned data frames were recorded. At the same time, a high-precision benchtop instrument 1 was used to record "real" interference-free signals as tags.

[0041] Model architecture: It adopts a neural network architecture containing two LSTM layers (32 hidden units per layer) and a fully connected output layer.

[0042] Training: The model is trained using the collected data, with Adam as the optimizer and mean squared error (MSE) as the loss function. The goal is for the model to learn to accurately predict the true signal value from input data frames with interference.

[0043] Deployment: After training, the model is quantized using tools such as TensorFlowLite (e.g., INT8 quantization), and its network weights are stored in the memory 51 of the central controller 5.

[0044] Real-time inference and correction: In field use, the central controller 5 takes each acquired data frame as the latest input to a time-series window and feeds it into the deployed AI model. The model performs one forward inference iteration, and its output is a "clean" signal that has been compensated for pressure and vibration disturbances in real time.

[0045] Step 3: Results Output and Effects like Figure 4 As shown, the high signal-to-noise ratio signal 9 after AI correction has an extremely stable baseline compared to the original signal 8 which is full of glitches and drift, and truly reflects the changes in sample concentration. It is finally presented on the screen of the user interface unit 53.

[0046] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. An intelligent, self-adapting flame ionization detector, comprising: The instrument is characterized in that it is equipped with an ionization unit and a pretreatment unit, wherein the ionization unit is used to detect the sample gas flow, and the pretreatment unit is used to pretreat the sample gas flow. The instrument is equipped with a gas path unit, which is used to inject sample gas flow into the pretreatment section, and then the sample gas flow is injected into the ionization section for ionization. It also includes a central controller installed on the instrument, which stores AI models and has algorithms embedded in it; The central controller is equipped with a memory used to store the parameters of the predictive thermal model and the network weights of the AI ​​correction model. The central controller is configured to perform the following collaborative tasks through internally embedded software algorithms: independently controlling the temperature of the preprocessing zone and the ionization zone according to a preset analysis mode by adopting a predictive low-power isostatic control method; and using a machine learning-based real-time signal correction method to dynamically compensate for interference in the main analysis signal by fusing the main analysis signal output by the detection module and the signals from the auxiliary environmental sensors, and outputting a corrected signal.

2. The intelligent adaptive flame ionization detector of claim 1, wherein, The predictive low-power constant temperature control method includes the following steps: establishing a predictive thermal model that describes the characteristics of the temperature control zone detection module, such as heat capacity and heat dissipation coefficient; during operation, periodically predicting the compensation energy required to maintain the target temperature in the next time period based on the current temperature and the thermal model; and supplying the compensation energy in the form of energy pulses at the beginning of the period using high-frequency pulse width modulation.

3. The intelligent adaptive flame ionization detector of claim 1, wherein, The machine learning-based real-time signal correction method employs a pre-trained recurrent neural network or its variants (long short-term memory network or gated recurrent unit). This model is trained to learn the nonlinear mapping relationship between a time-series input containing "[disturbed main analysis signal + multiple auxiliary environmental signals]" and an output of "undisturbed ideal main analysis signal".

4. The intelligent, self-adapting flame ionization detector of claim 1, wherein, The central controller is also equipped with a power management unit for power supply; it also includes a user interface unit installed on the central controller, which is connected to an external display screen for displaying results.

5. The intelligent, self-adapting flame ionization detector of claim 1, wherein, It also includes a triaxial accelerometer mounted on the instrument, which uses a TDK MPU-6050 six-axis motion sensor to collect motion data of the instrument; and an atmospheric pressure sensor mounted on the instrument to detect the atmospheric pressure value of the environment, which uses a Bosch BMP388 high-precision pressure sensor.

6. The intelligent, self-adapting flame ionization detector of claim 1, wherein, The ionization section includes an ionization chamber for ionization; a first heating element is installed outside the ionization chamber; a collecting electrode is installed inside the ionization chamber; a first temperature sensor is installed outside the first heating element for detecting the heating temperature of the first heating element; and a first heat insulation layer is installed outside the first heating element to prevent heat from the first heating element from diffusing outward.

7. The intelligent adaptive flame ionization detector of claim 6, wherein, The pretreatment section includes a second heating element; it also includes a second temperature sensor installed outside the second heating element; and a second heat insulation layer is installed outside the second heating element to prevent the heat of the second heating element from dissipating outward.

8. The intelligent adaptive flame ionization detector according to claim 1, characterized in that, The gas path unit includes a sampling pump and a proportional valve, which are used to quantitatively inject sample gas into the pretreatment section and then into the ionization section. The sample gas is preheated by the pretreatment section and then injected into the ionization section for subsequent ionization. At the same time, the ionization section can also heat the sample gas, thus forming a dual-temperature zone structure.

9. The intelligent adaptive flame ionization detector according to claim 7, characterized in that, The first heating element and the second heating element constitute a heating system, and both the first heating element and the second heating element are flexible polyimide film heaters. The inner sides of the first heating element and the second heating element form a temperature zone; the sample gas outputs an FID signal value after being ionized.

10. A method for controlling an intelligent adaptive flame ionization detector, characterized in that, Specifically, the following steps are included: Step 1: Predictive Low-Power Precision Temperature Control The goal of this step is to efficiently and stably establish and maintain the temperature in both temperature zones; The thermal model establishment is also known as factory calibration: In the factory calibration procedure, the instrument will automatically perform a "thermal characteristic self-learning" process; the central controller will perform step heating and natural cooling on the two temperature zones with different high-frequency pulse width modulation duty cycles and record the temperature change curves; based on these data, the system identification algorithm (such as the least squares method) is used to determine the thermal model parameters of each temperature zone, which can be simplified as: T(t+1)=T(t)+k1*P_in(t)-k2*(T(t)-T_ambient); The calculated thermal gain coefficient k1 and heat dissipation coefficient k2 are stored in memory; Runtime control: When the user selects an analysis method (such as "high boiling point mode", T1=250°C, T2=200°C), the central controller starts a predictive control cycle (5-second cycle): Prediction: At the beginning of the cycle, the central controller reads the current temperature T(t) and uses the thermal model to predict the temperature drop ΔT_loss caused only by natural heat dissipation in the next 5 seconds; Calculate the compensation energy: Calculate the total energy E_needed required to restore the temperature to the target value and offset the expected heat loss; Pulse heating: The central controller calculates the high-frequency pulse width modulation pulse width required to supply energy to E_needed, and outputs the pulse at 100% power for the first few hundred milliseconds of the 5-second cycle, and then turns off the power for the rest of the cycle; Intelligent standby: If the instrument is inactive for more than 5 minutes, it will automatically enter standby mode and lower the temperature to a low maintenance temperature (such as 90°C) with extremely low power consumption; when the user operates it again, the system uses the thermal model to calculate the fastest heating path and performs full-power pulse heating to restore the working temperature in a short time (such as within 60 seconds). Step Two: Real-time Analysis and AI Intelligent Correction Multidimensional data stream acquisition: After the temperature zone stabilizes, the analysis begins; the central controller synchronously acquires a data frame at a frequency of 10Hz, which includes: [FID signal value, raw ADC reading, atmospheric pressure sensor reading, accelerometer X-axis reading, accelerometer Y-axis reading, accelerometer Z-axis reading]; AI model training (completed offline): Data acquisition: During the research and development phase, the instrument was placed in a vibration table and an environmental chamber. With zero-point gas and standard gas of different concentrations introduced, various controllable vibrations (such as the spectrum of simulated walking and vehicle bumps) and pressure changes were applied, and millions of the above data frames were recorded. At the same time, a high-precision benchtop instrument was used to record "real" interference-free signals as tags. Model architecture: It adopts a neural network architecture containing two LSTM layers (32 hidden units per layer) and a fully connected output layer; Training: The model is trained using the collected data, with Adam as the optimizer and mean squared error (MSE) as the loss function; the goal is to enable the model to learn to accurately predict the true signal value from input data frames with interference. Deployment: After training, the model is quantized using TensorFlowLite and its network weights are stored in the memory of the central controller; Real-time inference and correction: When used in the field, the central controller takes each acquired data frame as the latest input of a time series window and sends it to the deployed AI model; the model performs forward inference once, and its output is a "clean" signal that has been compensated for pressure and vibration interference in real time. Step 3: Results Output and Effects The high signal-to-noise ratio signal after AI correction has an extremely stable baseline compared to the original signal which is full of glitches and drift, truly reflecting the changes in sample concentration, and is ultimately displayed on the screen of the user interface unit.

Images