Method for manufacturing colorimetric sensing units and arrays using multi-objective bayesian optimization algorithm

Abstract

The application discloses a method for manufacturing a colorimetric sensing unit and array by using a multi-target Bayesian optimization algorithm. Different formulations of colorimetric sensing units are initially prepared on a hole plate; the hole plate is dried, and the dried hole plate is introduced into a gas test to obtain the values of response time, reversibility, responsivity and sensitivity of each formulation; the values of response time, reversibility, responsivity and sensitivity of each formulation are processed by a Sigmoid function to obtain score values, which are substituted into a multi-target Bayesian optimization algorithm to generate a next round of formulation; colorimetric sensing units are prepared on the hole plate according to the formulation; the steps are repeatedly cycled, and the number of cycles reaches a stop condition to obtain a global optimal formulation; and colorimetric sensing units are prepared according to the global optimal formulation. The method solves the problem that multiple indexes of a sensor are simultaneously optimized and some indexes are easily neglected, and can quickly optimize a CO2 colorimetric sensing array which simultaneously has wide range, high sensitivity, fast response and high reversibility.

Description

Technical Field

[0001] This invention relates to an optimized manufacturing method for colorimetric sensors in the field of biochemical synthesis, specifically a method for manufacturing colorimetric sensor units and arrays using a multi-objective Bayesian optimization algorithm in the sensor performance optimization process. Background Technology

[0002] Carbon dioxide sensing is needed in many fields, including building ventilation, exhaust treatment, carbon capture, utilization and storage, deep-sea diving, and aerospace. In practical applications, CO2 detection concentrations can range from 400 ppm in the air to greater than 30%. Various efforts have been made to construct CO2 sensors based on different mechanisms, such as those based on gas chromatography, electrodes, electrochemistry, chemielectricity, optics, and acoustics.

[0003] Colorimetric methods have garnered significant attention due to their low cost and ease of use, making them an ideal choice for developing portable and disposable CO2 sensors. Current research on colorimetric CO2 sensors focuses on detection limits, liquid-phase detection, or high-selectivity detection. Few colorimetric sensors developed using a systematic approach simultaneously achieve high sensitivity, wide detection range, short response time, and good reversibility.

[0004] In theory, a CO2 colorimetric sensor array possessing all the aforementioned advantages can be prepared by adjusting the types and proportions of raw materials. However, adjusting the types and proportions of raw materials involves nearly 10 variables. Manually handling such high-dimensional variables is not only tedious but also prone to getting trapped in local optima.

[0005] To address this issue, optimization algorithms can be used to optimize the types and proportions of raw materials in colorimetric sensor arrays. However, most reported optimization algorithms for biochemical synthesis and material preparation only have one optimization objective, such as catalytic efficiency, yield, or quantum efficiency. In the development of CO2 colorimetric sensors mentioned earlier, multiple objectives, such as sensitivity, responsivity, response time, and reversibility, significantly impact its practicality. Different objectives require different degrees of optimization. Current technologies lack implementation methods that can simultaneously achieve multiple optimization objectives for colorimetric sensor arrays. Summary of the Invention

[0006] To address the problems existing in the background technology, this invention provides a method for optimizing sensor performance in machine learning-based colorimetric sensors. This method solves the problem that it is easy to overlook one aspect when optimizing multiple sensor indicators simultaneously, and accurately and quickly obtains a CO2 colorimetric sensor array with wide range, high sensitivity, fast response, and high reversibility.

[0007] The technical solution adopted in this invention is:

[0008] For colorimetric sensing units within a fixed detection range, the method is as follows:

[0009] 1) Initially prepare N colorimetric sensing units with different formulations on an N-well plate, with one colorimetric sensing unit prepared on each well; the initial N colorimetric sensing units with different formulations can be randomly generated.

[0010] 2) Dry the N-well plate, and then pass the dried N-well plate through different concentrations of the test gas within a fixed detection range to obtain the response time, reversibility, responsivity and sensitivity values ​​of the colorimetric sensing unit corresponding to each formulation.

[0011] 3) The response time, reversibility, responsivity and sensitivity values ​​of each formulation and its corresponding colorimetric sensing unit are processed by the Sigmoid function to obtain the score value, which is then substituted into the multi-objective Bayesian optimization algorithm to generate the next round of N formulations.

[0012] 4) Prepare N colorimetric sensing units on an N-well plate in the same manner as step 1) according to N formulas;

[0013] 5) Repeat steps 2) to 4) continuously to form a loop. After several rounds of looping, the stopping condition is reached to obtain the globally optimal formula. Prepare the colorimetric sensing unit according to the globally optimal formula.

[0014] Each colorimetric sensing unit is mainly composed of a polytetrafluoroethylene (PTFE) film as the sensing substrate, which is first added to the bottom of the wells of the well plate, and then the source solution is added to the wells of the well plate and mixed. The source solution includes tetrabutylammonium hydroxide (TBAH), ethyl cellulose (EC), polyethylene glycol (PEG, Mw=200), pH indicator and solvent. The solvent includes isopropanol (IPA) and deionized water. The pH indicator is selected from one of thymol blue (TB), m-cresol purple (CP), phenol red (PR) and cresol red (CR).

[0015] The formulation includes tetrabutylammonium hydroxide, ethyl cellulose, polyethylene glycol, pH indicator, and solvent concentrations, wherein the solvent concentrations include the concentrations of isopropanol and deionized water.

[0016] The N mentioned is greater than 8, that is, N>8.

[0017] Experiments revealed that in the aforementioned colorimetric sensing unit, a porous polytetrafluoroethylene (PTFE) film was used as the substrate, and tetrabutylammonium hydroxide, cresol red, polyethylene glycol, and ethyl cellulose were used to construct the colorimetric sensing unit. Isopropanol and water were used to adjust the affinity between the formulation solution of the sensing unit and the substrate. The formulation solution of the sensing unit refers to the solution obtained by mixing the loading materials together according to the formulation before loading them onto the substrate.

[0018] The mass percentage of tetrabutylammonium hydroxide is between 1% and 44%, the mass ratio of cresol red to polyethylene glycol is between 1:16 and 1:32, the mass percentage of ethyl cellulose is between 2.5% and 22.4%, and the mass percentage of aqueous phase is between 13% and 18%.

[0019] Specifically, a single colorimetric sensing unit can detect gas concentrations ranging from 0.04% to 30%.

[0020] In step 2), specifically: the N-well plate is dried in an oven at 50°C for 1 hour, then the dried N-well plate is fixed in the gas chamber and the gas containing CO2 to be tested is introduced for a period of time before measurement. The RGB values ​​of each well are obtained by taking an image of the N-well plate with a camera, and then the ventilation curve is obtained by analyzing and processing the RGB values. The values ​​of response time, reversibility, responsivity and sensitivity are extracted from the ventilation curve.

[0021] In the CO2-N2 mixture, the volume fraction of CO2 used for detection can be in any range of 0.04%-30%.

[0022] In step 3), the values ​​of response time, reversibility, responsivity, and sensitivity of the colorimetric sensing unit are processed by the Sigmoid function to obtain a score value. Specifically, only the response time, reversibility, and responsivity are individually transformed and calculated using the Sigmoid function, and then multiplied by the sensitivity value in a weighted manner to obtain the score value, as shown below:

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031] in, It is a rating value; , , These are the target values ​​for the responsivity, response time, and reversibility of the colorimetric sensing unit. It is the sensitivity of the colorimetric sensor unit; and These are the first and second scaling factors used to adjust different target ranges; This represents the time elapsed from the colorimetric sensor unit's exposure to the gas at the target concentration until the response value reaches 95% of its final stable value; it characterizes the sensor unit's response time. It is the response value when exposed to a CO2 atmosphere. This is the response value 240 seconds after exposure to the CO2 atmosphere was stopped; , These represent the threshold values ​​of the sigmoid function applied to the three objectives of response value, response time, and reversibility, respectively. These are the adjustment parameters of the slope of the sigmoid function near the control threshold, which applies to the three objectives of response value, response time, and reversibility. Expressed in terms of carbon dioxide concentration Independent variable, response value The slope obtained by performing linear regression on the response data collected from the dependent variable. This indicates the concentration of carbon dioxide.

[0032] Will , , and Multiplication is used as the weighted score value of the colorimetric sensing unit, and then a multi-objective Bayesian optimization algorithm is established by simultaneously optimizing multiple objectives on demand in the Bayesian algorithm.

[0033] This invention uses the Sigmoid function in its method to convert and calculate response time, reversibility, and responsivity. In this way, the Sigmoid function solves the problem of not being able to obtain good results in the multi-objective optimization process, and solves the problem of not being able to obtain good results in the multi-objective optimization process of colorimetric sensor performance, and solves the problem of not being able to obtain good results in the multi-objective optimization process of gas colorimetric sensor performance.

[0034] This invention designs and uses the Translate function (Sigmoid) to convert multiple technical indicators of a sensor. This conversion relationship is a numerical expression of the experience of sensor developers, which can help optimization algorithms understand the prior knowledge of technicians in sensor optimization scenarios. The converted target value can be used for multi-objective evaluation and simultaneous optimization of the sensor, and quickly develop sensors that meet the objectives.

[0035] In step 3), the six parameters and score values ​​of the N recipes are input into a multi-objective Bayesian optimization algorithm to construct a seven-dimensional space. The multi-objective Bayesian optimization algorithm is used to optimize and iterate to generate new six parameters for each of the N recipes with the goal of maximizing the score value.

[0036] The six parameters in the N formulas are the concentrations of five of the following: tetrabutylammonium hydroxide, ethyl cellulose, polyethylene glycol, pH indicator, isopropanol, and deionized water, as well as the type of pH indicator. Since the sum of the concentrations of tetrabutylammonium hydroxide, ethyl cellulose, polyethylene glycol, pH indicator, isopropanol, and deionized water is 100%, determining the concentration of only five of these components will allow us to determine the concentration of the remaining component.

[0037] In step 5):

[0038] During the iterative process, the parameters of the acquisition function in the multi-objective Bayesian optimization algorithm are adjusted so that: in the early-stage iterations (1-N), the multi-objective Bayesian optimization algorithm tends to explore when generating the next recipe; in the (N+1)th iteration (early-late stage), the multi-objective Bayesian optimization algorithm tends to mine when generating the next recipe. N is a preset threshold.

[0039] The timing of the iteration (early or late) is determined by a combination of the search space, the number of search rounds, and the number of data points in each round. The specific formula is as follows:

[0040]

[0041] in, It is an indicator used to evaluate the early or late stages of an optimization round. is the number of experimental data points in each round, k is the number of rounds completed, m is the dimension of the search space, and d i It is the width of the i-th dimension variable.

[0042] Based on this metric, the parameters used to control biased exploration and biased mining in the acquisition function are updated in each round of search. Specifically:

[0043]

[0044] in, These are the input and output variables of the function to be optimized; , These are the hyperparameters used by the function to regulate the balance between exploration and mining; These are the mean and variance estimated by the surrogate model; These are the cumulative distribution function and probability density function of the variable, respectively.

[0045] , The formula for updating with each iteration round is as follows:

[0046] 𝜉𝑘,=0.01×exp(−𝑆𝑠𝑡𝑎𝑔𝑒+1)

[0047] .

[0048] In step 5), the stopping condition is that the variance of the scores of N formulas is less than a preset variance threshold and the average score of N formulas is greater than a preset average threshold. Finally, the formula with the highest score is taken as the global optimal formula.

[0049] The present invention uses the Sigmoid function in the method to combine the multi-objective weighted multiplication score, which solves the problem that it is easy to lose sight of one thing when optimizing multiple indicators of the sensor at the same time. It can simultaneously optimize and obtain a CO2 colorimetric sensor array with wide range, high sensitivity, fast response and high reversibility.

[0050] The colorimetric sensor array is divided into colorimetric sensor units with different detection ranges. For each colorimetric sensor unit, the manufacturing is optimized according to the above method, and the colorimetric sensor array is composed of all colorimetric sensor unit arrays.

[0051] The colorimetric sensing unit / array described in this invention is a gas colorimetric sensing unit / array, specifically a CO2 colorimetric sensing unit / array.

[0052] The beneficial effects of this invention are:

[0053] This invention avoids investing too many resources in its optimization. By using a multi-objective Bayesian optimization algorithm to guide the iterative process, a CO2 colorimetric sensor array with high sensitivity, reversibility, and fast response time can be obtained in fewer iterations. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the equipment used to implement the automated preparation and testing process of the colorimetric sensor unit in this invention. (a) Opentrons automated liquid preparation platform, (b) functional area in the Opentrons automated liquid preparation platform, (c) gas testing platform.

[0055] Figure 2 This is a schematic diagram of a test image of the prepared sensing unit.

[0056] Figure 3 The following are schematic diagrams illustrating the performance and scoring of four typical sensing units before and after applying the sigmoid function: (a) ventilation test curve; (b) sigmoid function curves and scoring values ​​for different individual targets; (c) weighted scores before and after applying the sigmoid function.

[0057] Figure 4For the process optimization diagram: the score distribution of all sensing units in different cycles under different CO2 concentration ranges in Examples 1-6: (a) Example 1, (b) Example 2, (c) Example 3, (d) Example 4, (e) Example 5 and (f) Example 6.

[0058] Figure 5 Schematic diagram of the optimization results: (a) Ventilation test curve of the optimization results of Example 1; (b) Optimal formulations obtained from Examples 1-6; (c) Response values ​​of the formulations obtained from Examples 1-6 at different CO2 concentrations in their respective ranges. Detailed Implementation

[0059] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0060] The embodiments of the present invention are as follows:

[0061] Example 1

[0062] This embodiment uses a multi-objective Bayesian optimization algorithm to guide the optimization of a colorimetric sensor unit with a CO2 detection range of 0.04%-0.2%. The specific process is as follows:

[0063] (1) The equipment used for the preparation and testing of the CO2 colorimetric sensing unit for experimental verification is as follows: Figure 1 As shown, it includes an automated liquid dispensing system, Opentrons (OT-2 Liquid Handler, Opentrons, USA), and a ventilation testing system. The structure of Opentrons is as follows... Figure 1 As shown in Figure a, this is a platform with automated pipetting functionality. The shaded area at the bottom represents the functional modules involved in the pipetting process, and their layout is as follows: Figure 1 As shown in b. Where in Figure 1 A shaking device (BioShake 3000 elm, Qinstruments, Germany) is fixed at the upper left corner. The source solution is added... Figure 1 In the lower left well plate (b), the source solution comprises tetrabutylammonium hydroxide (TBAH), ethyl cellulose (EC), polyethylene glycol (PEG, Mw=200), a pH indicator, and a solvent. The solvent includes isopropanol (IPA) and deionized water. The pH indicator is selected from one of thymol blue (TB), m-cresol purple (CP), phenol red (PR), and cresol red (CR). Opentrons can automatically prepare formulations for 96 colorimetric sensing units in a single experiment.

[0064] Systems for ventilation testing, such as Figure 1As shown in Figure c, a camera (MV-CE060-10UC, Hikvision, China) is installed at the top of a darkroom, with LED surface light sources installed on both sides. A gas chamber with a transparent cover is fixed at the bottom, with an inlet and an outlet connected to both ends of the chamber. During testing, the prepared sensor units are placed in the gas chamber and sealed. A mass flow controller (AST10-HLM, ASLT, China) is used to control the CO2 exposure concentration. The camera at the top takes a picture of all colorimetric sensor units in the gas chamber every 2 seconds. A schematic diagram of the photographs is shown below. Figure 2 As shown, the shape and color RGB values ​​of each sensing unit are extracted using an open-source computer vision algorithm (OpenCV).

[0065] (2) 96 initial formulations were randomly generated, and colorimetric sensing units corresponding to these 96 formulations were prepared on the same PTFE matrix using the Opentrons automated liquid preparation system according to the process of liquid preparation-shaking-dropping.

[0066] The solution preparation process involves transferring the source solutions from different source solution well plates into a 96-well plate; the mixing process involves using a mixing device to mix the different source solutions in the 96-well plate; and the dropping process involves dropping the mixed formulation solution from the 96-well plate onto the PTFE of the sensing unit well plate.

[0067] Specifically, for each well, the formulation is repeatedly applied at random locations four times, and the average of the four test results is taken as the test result for that well / formula, in order to reduce error.

[0068] Finally, the prepared colorimetric sensing unit was dried in an oven at 50°C for 1 hour.

[0069] (3) Fix the dried sensing unit to, for example Figure 1 In the gas chamber at the bottom of the test device shown in c, a CO2-N2 mixture was introduced at a total flow rate of 10 L / min for 240 seconds, followed by purging with pure N2 gas at the same flow rate for 240 seconds.

[0070] In this embodiment, the volume fractions of CO2 in the CO2-N2 mixture are 0.04%, 0.08%, 0.12%, 0.16%, and 0.2%, respectively.

[0071] (4) Experimental measurement: The RGB values ​​of 96 formulations were extracted by analyzing the images captured by the camera. The response time, reversibility, responsivity and sensitivity of the colorimetric sensor unit corresponding to each formulation were extracted from the ventilation curve and substituted into the multi-objective Bayesian optimization algorithm to generate the next round of 96 formulations.

[0072] The specific implementation evaluates the performance of the colorimetric sensor in four aspects, then corrects the score value of each performance by superimposing a nonlinear function, and finally obtains the weighted formula.

[0073] The four performance parameters of a colorimetric sensor include response time, reversibility, responsivity, and sensitivity.

[0074] The response time is the time to reach equilibrium; the reversibility is the ratio of the response value 240 seconds after exposure to CO2 atmosphere is stopped to the maximum value when exposed to CO2 atmosphere; the responsivity is the RGB color change before and after exposure to CO2 atmosphere; and the sensitivity is the slope of the response value under different CO2 exposure concentrations.

[0075] The specific response is expressed using the Euclidean distance ∆E as follows:

[0076]

[0077] in, These represent the values ​​for the red, green, and blue color channels, respectively.

[0078] For responses only, reversibility, and responsivity, the sigmoid function is used for further calculation. For sensitivity, the sigmoid function is not used for further calculation.

[0079] For response time, reversibility, and responsiveness, the average of the measurement results for different CO2 volume fractions is taken as the final result; for sensitivity, the slope is calculated by combining the measurement results for different CO2 volume fractions.

[0080] Using the sigmoid function, the score approaches 0 when response time, reversibility, or responsiveness is below a certain threshold, and approaches 1 when it is above that threshold. The scores for response time, reversibility, responsiveness, and sensitivity are then multiplied to obtain a weighted score. This way, the final score decreases significantly when any single objective falls below the threshold. Simultaneously, it avoids allocating excessive resources to optimizing a single objective that exceeds the threshold.

[0081] After obtaining the transformation result of the sigmoid function, a multi-objective Bayesian optimization algorithm guides the iterative process, generating 96 recipes for the next round. The Bayesian optimization process is as follows: First, a Gaussian process is used as an alternative statistical model to describe the prior distribution, and expected values ​​and confidence levels are assigned to all unevaluated data points; then, the acquisition function is used to evaluate the prior distribution and propose suggestions. In each round, the tendency towards mining (preferring higher score values) and exploration (preferring higher uncertainty) is adjusted by setting different acquisition function parameters, thereby generating the recipes for the next round. After several iterations, a CO2 colorimetric sensor array with near-globally optimal performance is obtained. The acquisition functions used include:

[0082]

[0083] These are the input and output variables of the function to be optimized; These are the hyperparameters used by the function to regulate the balance between exploration and mining; These are the mean and variance estimated by the surrogate model; These are the cumulative distribution function and probability density function of the variable, respectively.

[0084] 5) Repeat steps 2) to 4) continuously to form a loop. After several rounds of looping, the stopping condition is reached to obtain the global optimal formula. According to the global optimal formula, N colorimetric sensing units are prepared on an N-well plate to form the final colorimetric sensing array.

[0085] During the iterative process, the parameters of the acquisition function in the multi-objective Bayesian optimization algorithm are adjusted so that: in the earlier iterations, the multi-objective Bayesian optimization algorithm tends to explore when generating the next round of recipes; in the later iterations, the multi-objective Bayesian optimization algorithm tends to mine when generating the next round of recipes. The early or late iteration is determined by a combination of the search space, the number of search rounds, and the number of data points in each round, as shown in the specific formula:

[0086]

[0087] in, It is an indicator used to evaluate the early or late stages of an optimization round. It is the number of data points in each round of experiments. is the number of rounds completed, and m is the dimension of the search space. It is the first The width of the dimensional variable. Based on this metric, the parameters in the acquisition function used to control biased exploration and biased mining are updated in each search round. Specifically:

[0088]

[0089]

[0090]

[0091] in, These are the input and output variables of the function to be optimized; , These are the hyperparameters used by the function to regulate the balance between exploration and mining; These are the mean and variance estimated by the surrogate model; These are the cumulative distribution function and probability density function of the variable, respectively. , The formula for updating with each iteration round is as follows:

[0092]

[0093]

[0094] The stopping condition is that the variance of the scores of the N recipes is less than a preset variance threshold, and the average score of the N recipes is greater than a preset average threshold.

[0095] Example 2

[0096] This embodiment uses a multi-objective Bayesian optimization algorithm to guide the optimization of a colorimetric sensor unit with a CO2 detection range of 0.2%-1%. The method is the same as in embodiment 1, except that the volume fraction of CO2 in the CO2-N2 mixture introduced in step (3) is 0.2%, 0.4%, 0.6%, 0.8%, and 1%, respectively.

[0097] Example 3

[0098] This embodiment uses a multi-objective Bayesian optimization algorithm to guide the optimization of a colorimetric sensor unit with a CO2 detection range of 1%-3%. The method is the same as in embodiment 1, except that the volume fraction of CO2 in the CO2-N2 mixture introduced in step (3) is 1%, 1.5%, 2%, 2.5%, and 3%, respectively.

[0099] Example 4

[0100] This embodiment uses a multi-objective Bayesian optimization algorithm to guide the optimization of a colorimetric sensor unit with a CO2 detection range of 3%-10%. The method is the same as in embodiment 1, except that the volume fraction of CO2 in the CO2-N2 mixture introduced in step (3) is 3%, 4.75%, 6.5%, 8.25%, and 10%, respectively.

[0101] Example 5

[0102] This embodiment uses a multi-objective Bayesian optimization algorithm to guide the optimization of a colorimetric sensor unit with a CO2 detection range of 10%-20%. The method is the same as in embodiment 1, except that the volume fraction of CO2 in the CO2-N2 mixture introduced in step (3) is 10%, 12.5%, 15%, 17.5%, and 20%, respectively.

[0103] Example 6

[0104] This embodiment uses a multi-objective Bayesian optimization algorithm to guide the optimization of a colorimetric sensor unit with a CO2 detection range of 20%-30%. The method is the same as in embodiment 1, except that the volume fraction of CO2 in the CO2-N2 mixture introduced in step (3) is 20%, 22.5%, 25%, 27.5%, and 30%, respectively.

[0105] Comparison without the sigmoid function

[0106] Typical ventilation curves of four sensing units in Example 1 are selected for demonstration: one is curve d, which shows better performance, and the other three are test curves a, b, and c, which show defective sensing units.

[0107] Their test curves are as follows Figure 3 As shown in figure a. Curve a has an excessively long response time, failing to reach equilibrium even after 240 seconds of exposure. Curve b has a very low response value, showing almost no response. Curve c exhibits poor reversibility.

[0108] Figure 3 The topmost curve (b) is the sigmoid function curve. When using the sigmoid function, a threshold is set for the score of each target. The response time of curve a, the response value of curve b, and the reversibility score of curve c are all below the threshold; therefore, their scores after sigmoid calculation will be close to 0.

[0109] Figure 3c shows the weighted scores of the four curves before and after the application of the sigmoid function. Before applying the sigmoid function, curve a had the highest score. However, its long response time should not have made it a good sensing unit. This problem was corrected after applying the sigmoid function.

[0110] like Figure 3 As shown on the right side of c, the test curve scores of all defective sensing units are significantly reduced, which demonstrates that the sigmoid function setting in this invention can significantly improve the rationality of the score, thereby improving the speed and accuracy of the optimization algorithm.

[0111] Test analysis of the example:

[0112] A. Optimization process of multiple iterations in Examples 1-6.

[0113] Figure 4 The values ​​of 'af' correspond to the distribution of score values ​​for different formulations during the optimization process in Examples 1-6. In the first experiment, the initial 96 samples were randomly generated, and most performed poorly, resulting in score distribution concentrated near 0. In subsequent batches of experiments, the score distribution of the generated formulations shifted towards higher-scoring regions, with the highest score steadily increasing. After four iterations, the highest score stabilized, indicating that a quasi-global maximum had been reached. It can be seen that using a multi-objective Bayesian algorithm combined with a high-throughput experimental approach, a quasi-global optimum can be found in the high-dimensional variable space within four iterations.

[0114] B. Formulation and performance of the CO2 colorimetric sensors obtained in Examples 1-6.

[0115] Taking Example 1 as an example, the ventilation test curve of the optimized colorimetric sensor formulation is as follows: Figure 5 As shown in Figure a, the formula obtained using the multi-objective Bayesian optimization algorithm can achieve an extremely short response time (within a few seconds), is completely reversible, and has good sensitivity. The color varies significantly under different CO2 concentrations, making it possible to achieve precise quantification.

[0116] The six final optimized formulations obtained from Examples 1-6 are as follows: Figure 5 As shown in b, the response values ​​of these formulations at different CO2 concentrations are as follows: Figure 5 As shown in c.

[0117] Four noteworthy phenomena were observed. First, the CR indicator exhibited the best performance across the entire CO2 range. It was hypothesized that CP and TB would perform well at low CO2 concentrations. Indeed, they do show good response values ​​at lower CO2 concentrations, but their sensitivity is not as high as CR. PR, on the other hand, could not detect CO2 concentrations below 1% and had poor sensitivity at higher CO2 concentrations, possibly due to its color—the ΔE value from yellow to red in the RGB spectrum is very small.

[0118] Secondly, according to Figure 5 b. In the lower CO2 concentration range (<3%), less TBAH is preferred; in the higher CO2 concentration range (>3%), the optimal dose of TBAH is close to 10%. This phenomenon aligns with the intuition and assumption that more CO2 molecules are needed to change the pH as the TBAH concentration increases.

[0119] Furthermore, the aqueous phase percentage in all optimal formulations is below 20%. The aqueous phase percentage refers to the content of TBAH, PEG, and deionized water. Since the PTFE substrate is hydrophobic, excessive aqueous phase leads to incompatibility between the sensing unit and the substrate, resulting in uneven distribution of the formulation solution on the porous PTFE surface after drying. Additionally, PEG does not evaporate during drying, further increasing CO2 mass transfer resistance and significantly prolonging the response time. Based on the characteristics of the sigmoid function, if the response time exceeds a threshold, the weighted score will be very low.

[0120] Finally, it was found that the optimal dosages of CR, PEG, and EC appeared to be independent of the CO2 concentration range. EC primarily served to modify surface morphology. The optimal dosage of PEG was closely related to the dosage of CR. This is likely because PEG provided an environment mimicking aqueous solutions. PEG is somewhat like water: it can form hydrogen bonds to bind tightly to CR and can donate protons. Unlike water, PEG does not evaporate and can be retained within the sensing unit. Furthermore, as a surfactant, PEG can improve CO2 mass transfer by reducing the viscosity of the loading or by making the loaded components more uniformly dispersed.

[0121] However, since PEG (Mw=200) is liquid and highly hydrophilic, adding too much PEG will lead to incompatibility with the hydrophobic substrate and reduce the score.

[0122] In the final implementation, all the colorimetric sensor unit arrays obtained in Examples 1-6 are combined into a colorimetric sensor array to achieve the calibration and manufacturing of a colorimetric sensor array with a wide CO2 concentration range (0.04-30%).

Claims

1. A method for rapidly manufacturing a colorimetric sensing unit using a multi-objective Bayesian optimization algorithm, characterized in that: 1) Initially prepare N colorimetric sensing units with different formulations on an N-well plate; 2) Dry the N-well plate, and then pass gas through the dried N-well plate to test and obtain the response time, reversibility, responsivity and sensitivity values ​​of the colorimetric sensing unit corresponding to each formulation; 3) The response time, reversibility, responsivity and sensitivity values ​​of each formulation and its corresponding colorimetric sensing unit are processed by the Sigmoid function to obtain the score value, which is then substituted into the multi-objective Bayesian optimization algorithm to generate the next round of N formulations. In step 3), the values ​​of response time, reversibility, responsivity, and sensitivity of the colorimetric sensing unit are processed by the Sigmoid function to obtain a score value. Specifically, only the response time, reversibility, and responsivity are individually transformed and calculated using the Sigmoid function, and then multiplied by the sensitivity value in a weighted manner to obtain the score value, as shown below: Where Obj is the rating value; ΔE Obj Δt Obj reversibility These are the responsivity, response time, and reversibility conversion target values ​​of the colorimetric sensing unit, respectively. sensitivity b1 is the sensitivity of the colorimetric sensor unit; b2 is the first and second scaling factors; Δt represents the time elapsed from the exposure of the colorimetric sensor unit to the gas of the concentration to be measured until the response value reaches 95% of the final stable value, which characterizes the response time of the sensor unit; ΔE is the response value exposed to CO2 atmosphere; ΔE0 is the response value 240 seconds after the exposure to CO2 atmosphere is stopped. th ΔE ,th Δt ,th reversibility These represent the thresholds of the sigmoid function applied to the three objectives of response value, response time, and reversibility, respectively; s ΔE s Δt s reversibility These are the adjustment parameters for the slope of the sigmoid function near the control threshold, which acts on the three objectives of response value, response time, and reversibility; slope(CO2,ΔE) represents the slope of the sigmoid function with respect to the carbon dioxide concentration C. CO2 C is the slope obtained by performing linear regression on the response data collected, where C is the independent variable and ΔE is the response value, and ΔE is the dependent variable. CO2 Indicates carbon dioxide concentration; 4) Prepare N colorimetric sensing units on an N-well plate according to N formulas; 5) Repeat steps 2) to 4) continuously until the stopping condition is met after several cycles to obtain the globally optimal formula. Prepare the colorimetric sensing unit according to the globally optimal formula. The timing of the iteration (early or late) is determined by a combination of the search space, the number of search rounds, and the number of data points in each round, and is set according to the following formula: Among them, S stage It is an indicator used to evaluate the early or late stages of the optimization round, N batch is the number of experimental data points in each round, k is the number of rounds completed, m is the dimension of the search space, and d i It is the width of the i-th dimension variable; Combined with indicator S stage In each round of search, update the parameters in the acquisition function used to control biased exploration and biased mining: Where X and Y are the input and output variables of the function to be optimized; ξ and λ are the hyperparameters of the function used to regulate the balance between exploration and mining; μ(X) and σ(X) are the mean and variance estimated by the surrogate model; Φ(X) and φ(X) are the cumulative distribution function and probability density function of the variables, respectively. The formulas for updating ξ and λ with the number of iterations are as follows: 。 2. The method for manufacturing a colorimetric sensing unit using a multi-objective Bayesian optimization algorithm according to claim 1, characterized in that: Each colorimetric sensing unit is mainly obtained by first adding a polytetrafluoroethylene (PTFE) film to the bottom of the well plate, then adding the source solution to the well plate, and mixing them. The source solution includes tetrabutylammonium hydroxide, ethyl cellulose, polyethylene glycol, pH indicator, and solvent. The solvent includes isopropanol (IPA) and deionized water. The pH indicator is selected from thymol blue, m-cresol purple, phenol red, and cresol red.

3. The method for manufacturing a colorimetric sensing unit using a multi-objective Bayesian optimization algorithm according to claim 1, characterized in that: The formulation includes tetrabutylammonium hydroxide, ethyl cellulose, polyethylene glycol, pH indicator, and solvent concentrations, wherein the solvent concentrations include the concentrations of isopropanol and deionized water.

4. The method for manufacturing a colorimetric sensing unit using a multi-objective Bayesian optimization algorithm according to claim 1, characterized in that: In the aforementioned colorimetric sensing unit, a porous polytetrafluoroethylene film is used as the substrate, and tetrabutylammonium hydroxide, cresol red, polyethylene glycol, and ethyl cellulose are used to construct the colorimetric sensing unit.

5. The method for manufacturing a colorimetric sensing unit using a multi-objective Bayesian optimization algorithm according to claim 1, characterized in that: In step 2), specifically: the N-well plate is dried in an oven at 50°C for 1 hour, then the dried N-well plate is fixed in the gas chamber and the gas to be tested is introduced for a period of time before measurement. The RGB values ​​of each well are obtained by taking an image of the N-well plate with a camera. Then, the ventilation curve is obtained by analyzing and processing the RGB values. The values ​​of response time, reversibility, responsivity and sensitivity are extracted from the ventilation curve.

6. The method for manufacturing a colorimetric sensing unit using a multi-objective Bayesian optimization algorithm according to claim 1, characterized in that: In step 3), the six parameters and score values ​​of the N recipes are input into a multi-objective Bayesian optimization algorithm to construct a seven-dimensional space. The multi-objective Bayesian optimization algorithm is used to optimize and iterate to generate new six parameters for each of the N recipes with the goal of maximizing the score value.

7. A method for manufacturing a colorimetric sensing unit using a multi-objective Bayesian optimization algorithm according to claim 1, characterized in that: In step 5): During the iterative process, the parameters of the acquisition function in the multi-objective Bayesian optimization algorithm are adjusted so that: in the first to Nth iterations, the multi-objective Bayesian optimization algorithm tends to explore when generating the next round of recipes; in the N+1th late-stage iteration, the multi-objective Bayesian optimization algorithm tends to mine when generating the next round of recipes.

8. A method for manufacturing a colorimetric sensing unit using a multi-objective Bayesian optimization algorithm according to claim 1, characterized in that: In step 5), the stopping condition is that the variance of the scores of N formulas is less than a preset variance threshold and the average score of N formulas is greater than a preset average threshold. Finally, the formula with the highest score is taken as the global optimal formula.

9. A method for manufacturing a colorimetric sensor array using a multi-objective Bayesian optimization algorithm, characterized in that: The colorimetric sensor array is divided into colorimetric sensor units with different detection ranges. For each colorimetric sensor unit, it is manufactured in an optimized manner according to the method described in claim 1. The colorimetric sensor array is composed of all colorimetric sensor unit arrays.

Images