A rapid quantitative analysis system and method for a continuous flow reaction process

Abstract

The application discloses a continuous flow reaction process rapid quantitative analysis system and method. The application combines standard substance mixing design, FTIR spectrum real-time acquisition and deep convolution modeling, realizes accurate prediction of the concentrations of reactants and products in continuous flow, and has good promotion adaptability and industrial application prospect. The application forms a dynamic concentration gradient by adjusting the flow rate ratio of a multi-component solution, generates multi-concentration samples in the whole process under continuous flow state, and significantly improves the modeling efficiency and real-time monitoring capability. In addition, the application uses a self-designed detection structure to optimize the spectrum acquisition module, combines constant temperature control and degassing treatment, effectively reduces the influence of flow fluctuation, temperature interference and bubbles on the signal, enhances the system stability and integration, realizes efficient in-situ spectrum detection, can be expanded to various continuous reaction monitoring scenes, and has actual deployment and industrial promotion value.

Description

Technical Field

[0001] This invention relates to the fields of chemical process analysis, spectral signal processing and intelligent modeling technology, specifically a rapid quantitative analysis system and method for continuous flow reaction processes. Background Technology

[0002] With the rapid development of continuous flow technology, online real-time monitoring has become a core element in ensuring process stability, improving product quality, and optimizing resource utilization. However, current continuous flow reaction analysis systems still have many bottlenecks:

[0003] Traditional methods rely heavily on offline standard sample preparation, which is cumbersome, time-consuming, and results in sparse sample distribution. This makes it impossible to fully capture the continuous concentration evolution under different reaction orders such as first, second, and third order, which limits the generalization performance of the model to complex kinetics. Absorption peaks in multi-component systems overlap significantly. Spectroscopic acquisition under flow conditions is easily disturbed by factors such as bubbles and temperature drift, resulting in poor data stability. Existing kinetic analyses are mostly limited to single-order assumptions and lack full coverage and seamless integration of multiple reaction orders.

[0004] Meanwhile, while existing online FTIR (Fourier Transform Infrared Spectroscopy) systems can achieve preliminary reaction monitoring, the modeling process requires manual intervention in sample collection and relies heavily on linear algorithms, making it difficult to handle multi-level nonlinear spectra. Microfluidic mixers, although promoting efficient mixing, are not deeply integrated with multi-level equivalent flow rate control to automatically generate training samples. Deep learning applications in quantitative spectroscopy are mostly limited to offline or single-level scenarios and have not been organically integrated with continuous flow systems. Summary of the Invention

[0005] To address the problems existing in the aforementioned background technologies, this invention discloses a rapid quantitative analysis system and method for continuous flow reaction processes through systematic integration and innovative optimization. Its core objective is to construct an online quantitative analysis system for continuous flow reaction processes based on the generation of multi-level equivalent dynamic concentration gradients and deep learning modeling. This system automates the generation of high-density samples covering multiple reaction order characteristics through precise flow rate control, optimizes detection hardware to improve signal quality, employs a 1D-CNN algorithm to overcome interference bottlenecks, and extends to the calculation of reaction kinetic parameters.

[0006] To achieve the above objectives, the present invention proposes the following technical solution:

[0007] A first aspect of the present invention provides a quantitative analysis system for a continuous flow reaction process, comprising:

[0008] An infusion device is used to generate dynamic concentration gradient samples covering the entire concentration range in real time under flowing conditions by adjusting the volumetric flow rate ratio of the precursor solution to the product solution.

[0009] The three-way mixing module is used to mix liquids from different channels at different flow rate ratios and guide the mixed solution through the detection channel;

[0010] The constant temperature control module is used to stabilize the temperature of the mixture at the reaction temperature to ensure the stability of spectral acquisition.

[0011] FTIR infrared probe module, used to acquire the infrared spectrum of mixed liquids in real time;

[0012] The host computer is used to preprocess the collected infrared spectra and input them into a one-dimensional convolutional neural network model to predict the concentration of the target components, and then carry out reaction kinetic analysis based on the concentration-time data.

[0013] A second aspect of the present invention provides a method for quantitative analysis of a continuous flow reaction process, applied to the above-mentioned apparatus, comprising the following steps:

[0014] Step 1: Prepare the stock solution required for spectral acquisition.

[0015] Step 2: A horizontal flow pump is used to regulate the volumetric flow rate ratio of the reaction precursor solution to the product solution, thereby generating dynamic gradient samples covering the entire concentration range in real time.

[0016] Step 3: The fluid enters the preheating chamber for temperature control, and then flows into the three-way mixing module to complete the mixing.

[0017] Step four: The mixture flows into the detection structure, and the infrared fiber optic probe collects the infrared spectral signal in real time and sends it to the host computer.

[0018] Step 5: After preprocessing the collected spectral data, input it into the depth model for modeling.

[0019] Step 6: Construct a one-dimensional convolutional neural network model, input the processed spectrum, and output the concentration values ​​of the corresponding components.

[0020] Step 7: Replace the precursor solution and control the solution mixing concentration ratio by adjusting the flow rate to simulate different reaction stages; repeat the above steps to obtain real-time spectral data of the reaction system.

[0021] Step 8: Import the collected spectra into the trained one-dimensional convolutional neural network model to predict the concentration of each component in the reaction system.

[0022] Step 9: Perform parameter fitting based on concentration-time data at different temperatures, calculate kinetic parameters, and complete the reaction kinetic analysis.

[0023] Compared with the prior art, the present invention has the following beneficial technical effects:

[0024] (1) By adjusting the flow rate ratio of the multi-component solution to form a dynamic concentration gradient, a multi-concentration sample is generated throughout the process under continuous flow, which significantly improves modeling efficiency and real-time monitoring capability.

[0025] (2) Using one-dimensional convolutional neural network (1D-CNN) for deep modeling, key features can be automatically extracted from high-dimensional infrared spectra, overcoming the accuracy limitations of traditional linear methods in multi-component overlapping and nonlinear signal scenarios, and ensuring the accuracy of target component concentration prediction.

[0026] (3) By optimizing the spectral acquisition module through the self-designed detection structure, combined with constant temperature control and degassing treatment, the influence of flow fluctuations, temperature interference and bubbles on the signal is effectively reduced, the system stability and integration are enhanced, and efficient in-situ spectral detection is achieved.

[0027] (4) Based on real-time concentration-time data, kinetic fitting can be carried out to calculate key parameters such as reaction rate constant and activation energy, providing quantitative basis for reaction mechanism research and process optimization, and expanding the application value of the system in process analysis.

[0028] (5) The system has a modular structure, good compatibility and scalability, and is suitable for real-time monitoring of reactants or products and identification of reaction status in continuous processes. It can be extended to various continuous reaction monitoring scenarios and has practical deployment and industrial promotion value. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the structure of a continuous flow reaction process analysis system according to an embodiment of this application;

[0030] Figure 2 for Figure 1 A schematic diagram of the detection structure;

[0031] Figure 3 This is a diagram of the D-CNN network structure in Embodiment 1 of this application;

[0032] Figure 4 This is a schematic diagram showing the changes in concentration and flow rate over time in the test examples of this application;

[0033] Figure 5 This is a graph showing the relationship between residence time and flow rate versus time in the test cases of this application;

[0034] Figure 6 This is a graph showing the instantaneous concentration changes of reactants and products in the test examples of this application. Detailed Implementation

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] like Figure 1 As shown, this application provides a quantitative analysis system for continuous flow reaction processes based on Fourier transform infrared spectroscopy (FTIR) and a one-dimensional convolutional neural network (1D-CNN), comprising:

[0037] Two sets of infusion devices control the flow rates of the precursor solution and the product solution respectively to form a dynamic concentration gradient;

[0038] The three-way mixing module (micro mixer) is used to efficiently mix two liquids at different flow rate ratios.

[0039] The constant temperature control module is used to stabilize the temperature of the mixture at the reaction temperature to ensure the stability of spectral acquisition.

[0040] The FTIR infrared probe module is used to acquire the infrared spectrum of the mixed liquid in real time. After preprocessing, the acquired spectrum is input into the 1D-CNN model to predict the concentration of the target component, and then reaction kinetic analysis is carried out based on the concentration-time data.

[0041] Furthermore, the infusion device includes: a storage bottle 1 for storing the reaction precursor solution and the product solution; a horizontal flow pump 2 for quantitatively delivering the aforementioned solutions; and a degassing machine 3 for removing air bubbles from the liquid. Based on this, the reaction precursor solution and the product solution are quantitatively delivered from the storage bottle by two sets of horizontal flow pumps, and then sequentially passed through a degassing machine to remove air bubbles from the liquid, thereby ensuring stable system operation and accurate spectral detection.

[0042] Furthermore, the three-way mixing module uses a T-type three-way mixer 4 to mix and guide the mixed solution through the detection channel, thereby achieving rapid and efficient mixing of the precursor solution and the product solution. Its multi-channel structure design provides a high surface area-to-volume ratio, promoting shear diffusion of the two phases under low Reynolds number conditions, improving mixing efficiency, and shortening the required channel length for mixing.

[0043] In particular, the T-type three-way mixer 4 is made of stainless steel, which has good chemical compatibility and high temperature resistance, and is suitable for a variety of chemical reaction environments.

[0044] Furthermore, the constant temperature control module is a water bath heating device equipped with a precision temperature control system, which can achieve constant temperature control within different ranges, with a control accuracy within ±0.1℃. The solution undergoes constant temperature regulation by this module before entering the detection structure 10, which helps eliminate the interference of temperature fluctuations on the infrared absorption spectrum and improves the consistency and stability of the modeling data.

[0045] Furthermore, its working method is as follows: water enters the preheating chamber 7 from the preheating chamber inlet 5, preheats the preheating coil 8 in the preheating chamber 7, and then flows out from the preheating chamber outlet 6 for water circulation to achieve constant temperature control.

[0046] Furthermore, the FTIR infrared detection module is composed of an infrared fiber optic probe 9 and a detection structure 10. The detection structure 10 connects the left and right pipes, with a narrow channel in the middle ensuring that the probe can fully contact the solution without leaving any dead volume of solution, and is fixed and sealed with a sealing rubber sleeve 10-3. The infrared fiber optic probe 9 is vertically installed at the side opening of the detection structure 10. This probe is in direct contact with the liquid and is used to collect the infrared spectral signal of the mixed solution in real time.

[0047] After the solution enters through the solution inlet 10-1, it flows through the fiber optic probe 9 and contacts the ATR crystal 9-1 at its tip. The solution's spectral information is collected, and then it flows out through the solution outlet 10-2. Figure 2 .

[0048] In summary, the workflow of this system is as follows:

[0049] The solution in the storage bottle 1 is pumped out by the horizontal flow pump 2, flows through the degasser 3 to remove air bubbles from the liquid, and is preheated by the preheating coil 8 in the preheating chamber 7. After mixing, it flows through the detection structure 10, where the infrared fiber optic probe 9 collects spectral information. This information is then processed by the infrared spectrometer 12 and uploaded to the host computer 13. After detection, the liquid finally flows into the waste liquid bottle 11.

[0050] This application also provides a quantitative analysis method for continuous flow reaction processes based on FTIR infrared spectroscopy and a one-dimensional convolutional neural network, applied to the above-mentioned system, including the following steps:

[0051] Step 1: Prepare the stock solution required for spectral acquisition.

[0052] Step 2: Two sets of adjustable-speed infusion devices output the component stock solution and the mixed solution respectively. A high-precision horizontal flow pump is used to regulate the volumetric flow rate ratio of the reactant reservoir and the product reservoir. Dynamic gradient samples covering the entire concentration range are generated in real time in the micro mixer. In each cycle, the following combination is performed: first-order equivalent segment (linear decay), second-order equivalent segment (square root decay), and third-order equivalent segment (cubic root decay), and this is repeated for 4 cycles.

[0053] It is worth noting that this step requires ensuring that the sum of the solution flow rates in the two pipes remains unchanged after the flow rate ratio is changed.

[0054] Furthermore, this step also includes: calculating the parameters of the dynamic gradient sample, specifically:

[0055] (1) Reactant flow rates in each equivalent segment:

[0056] Based on the law of conservation of mass, the instantaneous concentration is calculated, and a dynamic concentration gradient is generated. The formulas for calculating the reactant flow rate in each equivalent segment are as follows:

[0057] First-order equivalent segment (linear decay):

[0058] Second-order equivalent segment (square root attenuation):

[0059] Third-order equivalent segment (cubic root attenuation):

[0060] in, , , The reaction flow rates corresponding to the three equivalent stages are: The total constant flow rate of the two solutions; This represents the normalized time within the corresponding series segment, with a value ranging from 0 to 1.

[0061] (2) Product flow rate :

[0062] Determined by the difference between the total flow rate and the reactant flow rate:

[0063]

[0064] (3) Instantaneous concentrations of reactants and products:

[0065] Based on the law of conservation of mass, the concentrations of each substance in the solution are calculated using the initial solution concentration and the flow rate ratio, thus obtaining the instantaneous concentrations of reactants and products in the mixed solution:

[0066] Instantaneous concentration of reactants:

[0067] Instantaneous concentration of product:

[0068] in, The initial concentration of the reactant reservoir solution. This represents the initial concentration of the product storage solution.

[0069] (4) Functional relationship between flow velocity and residence time:

[0070] The numerical change in the instantaneous theoretical residence time of the liquid caused by the change in instantaneous flow velocity is calculated, yielding the following formula:

[0071]

[0072] in, It is a momentary pause. It is the residence time of dead volume. It is the coefficient of flow rate variation, where t is time. It's the flow rate. It is the reaction volume.

[0073] Calculate the actual residence time of a single liquid from entry to exit, due to the continuous change in flow velocity:

[0074]

[0075] in, It is the time of entry into the reactor. The moment of outflow from the reactor.

[0076] make have to:

[0077]

[0078] The relationship between flow rate and time is as follows:

[0079]

[0080] in, Indicates flow rate, It is the system volume. It is the residence time of dead volume. This represents the coefficient of velocity variation.

[0081] Step 3: The fluid enters the preheating chamber for temperature control, and then flows into the T-type three-way mixer to complete efficient mixing.

[0082] Furthermore, the mixture is temperature-controlled by a preheating coil before entering the infrared detection module. The temperature can be adjusted according to the required temperature for the reaction, with an error of about 0.1 degrees Celsius.

[0083] Step four: The mixed liquid flows into the detection structure, and the inserted ATR infrared fiber optic probe collects the infrared spectral signal in real time and sends it to the host computer.

[0084] Furthermore, the method for real-time acquisition of infrared spectral signals is specifically as follows:

[0085] Using an FTIR infrared spectrometer, the system resolution is [missing information]. The flow rate of the advection pump changes at 1-second intervals, and the spectrometer is set to acquire one spectrum every 5 seconds.

[0086] Step 5: The collected spectral data is preprocessed by SG smoothing, baseline correction, etc., and then input into the depth model for modeling.

[0087] Step 6: Construct a 1D-CNN model with a structure consisting of 3 convolutional layers, 3 max pooling layers, 1 flattening layer, and 2 fully connected layers. Input the processed spectrum into the 1D-CNN model and output the concentration values ​​of the corresponding components.

[0088] In this embodiment, all infrared spectral sample data is divided into three parts: 75%, 15%, and 10%. 10% is designated as the test set, 15% as the validation set, and 75% as the training set, used for training model parameters, adjusting hyperparameters, and ultimately evaluating model performance, respectively. The model is trained using the Adam optimizer with an initial learning rate of 0.001, and the entire training process consists of a maximum of 100 rounds.

[0089] Furthermore, such as Figure 3 As shown, the 1D-CNN model structure specifically includes, along the input-to-output direction, one convolutional layer, one max-pooling layer, one convolutional layer, one max-pooling layer, one convolutional layer, one max-pooling layer, one flattening layer, and two fully connected layers, used to complete data feature extraction. Specifically:

[0090] The first layer of the model uses 64 filters to scan the spectrum and look for basic features. The second and third layers progressively increase the number of filters to 128 and 256, respectively, combining and extracting more complex patterns from the spectrum. A pooling layer is connected after each convolutional layer to reduce the data size and retain key information. A flattening layer converts the multidimensional feature map into a one-dimensional vector, and two fully connected layers integrate all the extracted features, finally outputting the predicted concentration values ​​of the three substances.

[0091] To prevent the model from overlearning on limited sample data and thus degrading prediction performance, this application employs a combination of regularization techniques. L1 and L2 combined constraints are applied to the network weights to prevent the model from becoming overly reliant on individual extreme values. Simultaneously, a portion of neurons are randomly and temporarily dropped during training, forcing the network to learn more robust features.

[0092] In addition, the training process is closely monitored. Once the model's performance on the validation set stops improving, training will be terminated early, and the learning rate will be automatically lowered to seek a better solution.

[0093] Step 7: Replace the precursor solution and control the solution mixing concentration ratio by adjusting the flow rate to simulate different reaction stages; repeat the above steps to obtain real-time spectral data of the reaction system.

[0094] Step 8: Import the collected spectra into the trained 1D-CNN model to predict the concentration of each component in the reaction system.

[0095] Step 9: Based on the concentration-time data at different temperatures, perform parameter fitting, calculate kinetic parameters such as reaction rate constant and activation energy, and complete the reaction kinetic analysis.

[0096] Furthermore, step nine includes:

[0097] Kinetic fitting was performed on concentration-time data at different temperatures (30°C, 40°C, 50°C, 60°C), assuming the reaction is first-order and the reaction rate satisfies:

[0098]

[0099] The infrared spectral data of the complete reaction process can be represented by a two-dimensional matrix D. The infrared spectrum D of the reaction process can be expressed as:

[0100]

[0101] in, It is a matrix showing the concentration changes of each substance during the reaction process. It is a spectral matrix of pure matter. It is the number of substances participating in the reaction. This is the measurement error matrix.

[0102] First, assign a given value to the rate constant k to be calculated in the kinetic model, and then calculate the concentration matrix at this time. .

[0103] Next, calculate the spectral matrix of the pure components:

[0104] in, It is a concentration matrix The pseudo-inverse matrix.

[0105] Finally, calculate the error matrix:

[0106] Adjust the parameters and repeat the above steps until the sum of squares of all elements in the error matrix reaches its minimum. At this point, the rate constant is assigned the final estimated value of the rate constant.

[0107] Using the actual measured spectrum of the pure substance as an additional constraint in the iterative process, the objective function FIR is shown in the following equation:

[0108] ,

[0109] in, These are the parameters of the dynamic model to be fitted; This represents the total number of parameters in the dynamic model to be fitted. and These represent the measured pure spectrum and the calculated pure spectrum of the l-th substance, respectively.

[0110] Based on the relationship between the rate constant and temperature, using the Arrhenius equation: ,by right Linear fitting, slope is The activation energy can be calculated. .

[0111] in, This is the universal gas constant. Pre-exponential factors, Thermodynamic temperature.

[0112] To verify the effectiveness of the method described in this application, a typical binary reaction system was selected for spectral acquisition and concentration modeling experiments under continuous flow conditions. The specific steps are as follows:

[0113] Step 1: React 2 mol / L 2,5-hexanedione-methanol solution and 2 mol / L ethanolamine-methanol solution to obtain a 1 mol / L product solution for sample preparation. Using 2,5-hexanedione-methanol solution as starting material A and 1-(2-hydroxyethyl)-2,5-dimethylpyrrole-methanol solution as starting material B as examples, the calculations below will be explained.

[0114] Step 2: Two sets of advection pumps are used to control the volumetric flow rates of the precursor solution and the product solution, respectively. The flow rate changes are controlled according to the program, with the advection pump flow rate changing at 1-second intervals. The spectrometer is set to acquire one spectrum every 5 seconds, and the total flow rate is 10 mL / min. -1 Each experiment involved four cycles, and the variation pattern of the flow rate ratio was obtained as shown in the figure. Figure 4 (a)

[0115] Step 3: The fluid enters the constant temperature preheating module for temperature control, and then flows into the T-type micro-mixer for efficient mixing. Assume the concentration of the 2,5-hexanedione-methanol solution in solution A is 2 mol·L⁻¹. -1 The product solution has a concentration of 1 mol / L in solution B. The concentrations of each component in the mixed solution change with the flow rate ratio as follows: Figure 4 As shown in Figure (b), the changes in the concentration of the component solution at different reaction stages are simulated.

[0116] Step 4: The mixed liquid enters the detection structure, and an infrared fiber optic probe is vertically inserted into its side opening and connected to an FTIR spectrometer to acquire the infrared spectrum of the reaction mixture in real time. The real-time acquisition wavenumber range is 4000–650 cm⁻¹.-1 The FTIR spectral signal was obtained. The spectral data was transmitted to the host computer via serial port and processed using Python. During the acquisition process, 6–8 spectral data points were acquired for each combination of temperature and flow rate ratio, for a total of 624 spectral samples.

[0117] Step 5: After preprocessing such as Savitzky-Golay smoothing, baseline correction, and vector normalization, the spectral data is input into a 1D-CNN neural network for modeling. The activation function is ReLU and the optimizer is Adam.

[0118] Step 6: Construct a 1D-CNN model. The test case in this application uses a one-dimensional convolutional neural network with three layers to construct the prediction model. The specific parameters are shown in Table 1.

[0119] Table 1 1D-CNN Structure Parameters

[0120]

[0121] Step 7: Prepare ethanolamine-methanol solution and 2,5-hexanedione-methanol solution, and connect them to two sets of horizontal flow pumps respectively. Stabilize the temperature of the continuous flow coil reactor at four gradients (30℃, 40℃, 50℃, and 60℃) using a constant temperature water bath to simulate different reaction temperature conditions. Employ a continuous flow rate adjustment strategy to... , The instantaneous residence time and flow rate are calculated by substituting into the formula. The flow rate is continuously adjusted at 1-second intervals to ensure coverage of the entire process from the initial stage of the reaction to equilibrium.

[0122] Step 8: Perform spectral detection at the reaction liquid outlet. The ATR fiber optic probe acquires infrared spectra in real time. The spectra are preprocessed by baseline correction and Savitzky-Golay filtering to eliminate noise and baseline drift.

[0123] Step 9: Input the preprocessed spectrum into the trained 1D-CNN model to output the instantaneous concentrations of reactants and products.

[0124] Step 10: Based on the concentration-time data at different temperatures, perform parameter fitting, calculate kinetic parameters such as the reaction rate constant and activation energy, and complete the reaction kinetic analysis.

[0125] In the test case of this application, to further analyze the impact of flow rate adjustment on the reaction process, online monitoring of the reaction was conducted under continuous flow rate adjustment. Based on the previously established 1D-CNN model, a quantitative analysis of the reaction process was performed. Figure 5This paper demonstrates the relationship between residence time, flow rate, and time, intuitively illustrating the equivalence between flow rate regulation and reaction time. By designing a linear relationship between residence time and time, and regulating the flow rate to continuously distribute residence time from short to long, the entire transformation process is simulated, reflecting the precise control of residence time achieved by the method in this application. Figure 5 In the middle (a), the curve of the stay time changes with time. Figure 5 In the middle (b), the flow velocity changes over time. Figure 5 In the middle (c), the curve shows the change in flow velocity with residence time.

[0126] Instantaneous concentration changes of reactants and products, such as Figure 6 As shown in the figure, each point represents the concentration quantification result of a spectral data point during a continuous flow rate change reaction process. The quantification results are consistent with those of traditional quantification methods, reflecting the effectiveness of achieving multispectral acquisition and spectral quantitative analysis in a short time and monitoring the entire reaction process through flow rate control. This demonstrates the feasibility and accuracy of sample acquisition, modeling, and quantitative analysis using the flow rate ratio control method.

[0127] In summary, this application achieves accurate prediction of reactant and product concentrations in continuous flow by combining standard substance mixing design, real-time FTIR spectral acquisition, and deep convolution modeling, demonstrating good adaptability for widespread application and promising prospects for industrial use.

[0128] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A rapid quantitative analysis system for a continuous flow reaction process, characterized in that, include: An infusion device is used to generate dynamic concentration gradient samples covering the entire concentration range in real time under flowing conditions by adjusting the volumetric flow rate ratio of the precursor solution to the product solution. The three-way mixing module is used to mix liquids from different channels at different flow rate ratios and guide the mixed solution through the detection channel; The constant temperature control module is used to stabilize the temperature of the mixture at the reaction temperature to ensure the stability of spectral acquisition. FTIR infrared probe module, used to acquire the infrared spectrum of mixed liquids in real time; The host computer is used to preprocess the collected infrared spectra and input them into a one-dimensional convolutional neural network model to predict the concentration of the target components, and then carry out reaction kinetic analysis based on the concentration-time data.

2. The rapid quantitative analysis system for a continuous flow reaction process according to claim 1, characterized in that, The infusion device includes: Storage bottle (1) is used to store the reaction precursor solution and product solution; A horizontal flow pump (2) is used to quantitatively deliver the solution; The degasser (3) is used to remove air bubbles from the liquid to ensure the stable operation of the system and the accuracy of spectral detection.

3. The rapid quantitative analysis system for a continuous flow reaction process according to claim 2, characterized in that, The constant temperature control module uses a water bath heating device.

4. A rapid quantitative analysis system for a continuous flow reaction process according to claim 1 or 3, characterized in that, The FTIR infrared detection module includes: The detection structure (10) is used to connect the constant temperature control module and the infrared spectrometer (12). An infrared fiber optic probe (9) is vertically installed on the side of the detection structure (10) and in direct contact with the solution to collect the infrared spectral signal of the mixed solution in real time.

5. The rapid quantitative analysis system for a continuous flow reaction process according to claim 4, characterized in that, The testing organization (10) includes: The solution inlet (10-1) is connected to the constant temperature control module; the solution outlet (10-2) is connected to the waste liquid bottle (11); A narrow channel is provided between the solution inlet (10-1) and the solution outlet (10-2) to ensure that the infrared fiber optic probe (9) is in full contact with the solution while preventing dead volume solution from remaining. A sealing rubber sleeve (10-3) is provided at the side opening where the infrared fiber optic probe (9) is installed, for fixing and sealing.

6. A rapid quantitative analysis method for a continuous flow reaction process, characterized in that, The apparatus applied to any one of claims 1-5 comprises the following steps: Step 1: Prepare the stock solution required for spectral acquisition; Step 2: Use a horizontal flow pump (2) to regulate the volumetric flow rate ratio of the reaction precursor solution to the product solution, and generate dynamic gradient samples covering the entire concentration range in real time. Step 3: The fluid enters the preheating chamber (7) for temperature control, and then flows into the three-way mixing module to complete the mixing. Step 4: The mixed liquid flows into the detection structure (10), and the infrared fiber optic probe (9) collects the infrared spectrum signal in real time and sends it to the host computer (13). Step 5: After preprocessing the collected spectral data, input it into the depth model for modeling; Step 6: Construct a one-dimensional convolutional neural network model, input the processed spectrum, and output the concentration values ​​of the corresponding components; Step 7: Replace the precursor solution and control the solution mixing concentration ratio by adjusting the flow rate to simulate different reaction stages; repeat the above steps to obtain real-time spectral data of the reaction system. Step 8: Import the collected spectra into the trained one-dimensional convolutional neural network model to predict the concentration of each component in the reaction system; Step 9: Perform parameter fitting based on concentration-time data at different temperatures, calculate kinetic parameters, and complete the reaction kinetic analysis.

7. The rapid quantitative analysis method for a continuous flow reaction process according to claim 6, characterized in that, Each period of the dynamic gradient sample includes: In the first-order equivalent segment, the reaction flow rate decreases linearly. In the second-order equivalent segment, the reaction flow rate decreases in a square root manner; In the third-order equivalent segment, the reaction flow velocity exhibits a cube root decay.

8. The rapid quantitative analysis method for a continuous flow reaction process according to claim 7, characterized in that, Step two also includes: Based on the law of conservation of mass, instantaneous concentration is calculated, dynamic concentration gradient is generated, and the reaction flow rate of each equivalent segment is obtained. The product flow rate is determined by the difference between the total flow rate and the reactant flow rate; The concentrations of each substance in the solution are calculated by using the initial solution concentration and flow rate ratio. Based on the law of conservation of mass, the instantaneous concentrations of reactants and products in the mixed solution are obtained.

9. A rapid quantitative analysis method for a continuous flow reaction process according to claim 6 or 8, characterized in that, The 1D-CNN model includes: The input spectral signal is scanned and basic features are found using a convolutional layer; the number of filters is gradually increased to combine and extract more complex patterns from the spectral signal. A pooling layer is connected after each convolutional layer to reduce the data size and retain key information; A flattening layer is used to convert the multidimensional feature map into a one-dimensional vector. Finally, a fully connected layer is used to integrate all the extracted features and output the concentration prediction value.

10. The rapid quantitative analysis method for a continuous flow reaction process according to claim 8, characterized in that, Step nine includes: Kinetic models were established by fitting the concentration-time data at different temperatures. Infrared spectral data of the complete reaction process are represented using a two-dimensional matrix; Assign a value to the rate constant, calculate the concentration matrix at this point, and further calculate the pure component spectral matrix to obtain the error matrix; adjust the parameters and repeat the above calculation until the sum of squares of each element in the error matrix reaches its minimum. At this point, the assigned value of the rate constant is the final estimated value of the rate constant. The objective function is established by using the pure substance spectrum obtained from actual measurements as an additional constraint on the iterative process; the activation energy is calculated using the Arrhenius equation based on the relationship between the rate constant and temperature.

Images