Odor prediction method for aqueous polymer composition

Abstract

A method and system (400) for predicting odor of an aqueous polymer composition, such as a polymeric coating, the method comprising: analyzing characterizing the aqueous polymer composition with a detector (4013), thereby producing concentration data of volatile organic compounds in the aqueous polymer composition from the analytical characterization; inputting the concentration data to a decision tree ensemble, the decision tree ensemble configured to predict an odor intensity of the aqueous polymer composition after polymerization based on the concentration data; and outputting the predicted odor intensity of the aqueous polymer composition from the decision tree ensemble.

Description

Technical Field

[0001] The present invention relates to a method and system for predicting the odor of waterborne polymer compositions and coatings made therefrom by machine learning, particularly suitable for coating application. Background Art

[0002] As an environmentally friendly alternative to solvent-based compositions, aqueous or water-based binders or coating compositions are becoming increasingly important. In the architectural coating industry, particularly for indoor applications, some manufacturers and end users are also keen on the smell of aqueous compositions. Conventional odor evaluation of binders and coating compositions depends primarily on human sensory panels. This odor panel test is laborious and time-consuming, which is also often subjective, and usually requires repetition to obtain consistent test results. In addition, long-term exposure to odor may cause potential harm to odor panel members. Existing computer-implemented odor prediction methods are usually developed for relatively simple systems (such as tobacco), which are not suitable for aqueous polymer compositions or coatings with more complex chemical compositions in the coating industry. These aqueous polymer compositions usually contain various types (e.g., more than 30 types) of volatile organic compounds, with concentrations in a wide range (e.g., from parts per billion to parts per million of aqueous polymer compositions), and the interactions between these compounds usually have a significant impact on the odor of aqueous polymer compositions. Therefore, it is more challenging to develop methods or systems for odor prediction of aqueous polymer systems.

[0003] It would therefore be desirable to provide a method for predicting the odor of an aqueous polymer composition or a coating made therefrom. Summary of the invention

[0004] The present invention provides a novel computer-assisted method and system without the above problems. The method of the present invention includes inputting a novel combination and concentration of volatile organic compounds (VOCs) in a waterborne polymer composition or a coating made therefrom into a specific supervised machine learning module to predict the odor intensity of the composition or coating. The machine learning module used in the present invention is a decision tree ensemble, which is trained to predict the odor intensity of a waterborne polymer composition or coating using a training data set using multiple training samples. The training data set contains concentration data and the corresponding actual odor intensity of a waterborne polymer composition or coating containing VOCs as assessed by a human panelist. This method or system implements a standardized and automated method that can accurately predict odors by assessing odor intensity. Compared with odor assessments performed by human panelists, the method of the present invention greatly improves test consistency and accuracy without being affected by changes in panelists, experimental conditions, and environmental conditions. At the same time, the method can also greatly improve the productivity and efficiency of rapidly evaluating dozens of samples in a sequence, thereby reducing the labor cost and time of training and evaluation, and also preventing panelists from being exposed to harmful substances due to inhalation.

[0005] In a first aspect, the present invention is a method for predicting the odor of an aqueous polymer composition. The method comprises:

[0006] analytically characterizing the aqueous polymer composition with a detector, thereby generating concentration data of volatile organic compounds in the aqueous polymer composition from the analytical characterization;

[0007] inputting the concentration data into a decision tree ensemble configured to predict the odor intensity of the aqueous polymer composition based on the concentration data; and

[0008] A predicted odor intensity of the aqueous polymer composition is output from the decision tree ensemble.

[0009] In a second aspect, the present invention is a method for predicting the odor of a coating. The method comprises:

[0010] analyzing and characterizing the coating using a detector to generate concentration data of volatile organic compounds in the coating, wherein the coating is obtained by drying an aqueous polymer composition;

[0011] inputting the concentration data into a decision tree ensemble configured to predict an odor intensity of the coating based on the concentration data; and

[0012] The predicted odor intensity of the coating is output from the decision tree ensemble.

[0013] In a third aspect, the present invention is a system for predicting the odor of an aqueous polymer composition or a coating made therefrom. The system comprises:

[0014] a detector configured to analytically characterize the aqueous polymer composition or the coating, thereby generating concentration data of volatile organic compounds from the analytical characterization; and a computing device having a decision tree ensemble deployed thereon, the computing device configured to input the concentration data and output a predicted odor intensity of the aqueous polymer composition or the coating. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Fig. 1 shows a schematic diagram of a simplified example of a decision tree used for prediction.

[0016] Figure 2 A flowchart of an odor prediction method according to an example of the present invention is shown.

[0017] Figure 3 A schematic diagram of a cloud-based server cluster according to an example of the present invention is shown.

[0018] Figure 4 A schematic block diagram of an odor prediction system according to an example of the present invention is shown. DETAILED DESCRIPTION

[0019] When no date is indicated by a test method number, the test method refers to the most current test method as of the priority date of this document.

[0020] Products identified by their trade names refer to compositions available under those trade names on the priority date of this document.

[0021] “And / or” means “and, or as an alternative.” Unless otherwise indicated, all ranges are inclusive.

[0022] "Smell" refers to the sensation perceived by the olfactory nerves through the nose.

[0023] "Odor intensity" is a measure of the intensity of an odor based on initial perception. Odor intensity can be assessed based on the criteria described in the VDA 270 standard. The German Association of the Automotive Industry (VDA) developed the VDA 270 standard (Determination of the odor characteristics of motor vehicle trim materials).

[0024] "Actual odor intensity" is the odor intensity rated by a human panelist according to the odor panel test described in the Examples section below. "Volatile organic compound" ("VOC") refers to any organic compound having a normal boiling point of 250 degrees Celsius (°C) or less at a pressure of 101.3 kilopascals (kPa).

[0025] "VOC profile" refers to a dataset containing the identification of VOCs and their concentrations.

[0026] By "aqueous" polymer composition is meant herein a composition comprising a polymer present in an aqueous medium (e.g., polymer particles dispersed in an aqueous medium). "Aqueous medium" herein refers to water and 0 to 30 wt % of a water-miscible compound, such as, for example, an alcohol, a glycol, a glycol ether, a glycol ester, or a mixture thereof, based on the weight of the medium.

[0027] "Acrylic (co)polymer" refers herein to a homopolymer of an acrylic monomer or a copolymer comprising structural units of an acrylic monomer and one or more additional monomers. "Acrylic" in the present invention includes (meth)acrylic acid, (meth)alkyl acrylates, (meth)acrylamide, (meth)acrylonitrile and modified forms thereof, such as (meth)hydroxyalkyl acrylates. Throughout this document, the word fragment "(meth)acryloyl" refers to both "methacryloyl" and "acryloyl". For example, (meth)acrylic acid refers to both methacrylic acid and acrylic acid, and methyl (meth)acrylate refers to both methyl methacrylate and methyl acrylate. Specific examples of acrylic (co)polymers include acrylic homopolymers, styrene acrylic copolymers or mixtures thereof.

[0028] A "structural unit" of a named monomer (also referred to as a "polymerized unit") refers to the residue of the monomer after polymerization, i.e., the polymerized monomer or the monomer in polymerized form. For example, the structural unit of methyl methacrylate is shown below: The dashed lines indicate the points of attachment of the structural units to the polymer backbone.

[0029] "Machine learning" refers to a set of methods that 'learn' from data to improve performance on a specific task. Machine learning algorithms build models based on historical data (also called training data) to make predictions as model outputs.

[0030] A "decision tree" is a type of machine learning where the model has a tree-like structure to represent a 'test' on a series of attributes. Each internal node represents a 'test', and each leaf node represents a decision following the outcome of the attribute.

[0031] The method of the present invention can be used to predict the odor of an aqueous polymer composition or a coating made therefrom (collectively referred to as a "test sample"). In particular, the method can be used to predict the odor of an aqueous polymer composition.

[0032] The method of the present invention includes characterizing the aqueous polymer composition or coating using a detector to generate concentration data of volatile organic compounds in the aqueous polymer composition or coating.

[0033] Aqueous polymer compositions, particularly in coating applications, typically contain various types of VOCs, such as more than twenty types of VOCs. These VOCs can be present in a wide range of concentrations, from parts per billion to parts per million compositions. The types of VOCs and the interactions between these VOCs typically have a significant effect on the perceived odor of the aqueous polymer composition. Therefore, the odor intensity from a mixture of individual chemicals cannot reflect the actual odor of the aqueous polymer composition containing these individual chemicals. By using concentration data of VOC mixtures in aqueous compositions obtained from analytical characterization to train a specific machine learning module, the present invention can achieve surprisingly higher prediction accuracy than a module that generates predicted odor intensity by training with input of individual chemical structure data (such as chemical molecule descriptors) or by measuring individual chemicals for training.

[0034] Waterborne polymer composition

[0035] The aqueous polymer composition useful in the present invention may comprise one or more acrylic acid (co)polymers. The acrylic acid (co)polymer may comprise any one or any combination of more than one type of alkyl (meth)acrylate. "Alkyl" means a linear, branched or cyclic alkyl group. The alkyl (meth)acrylate may be a C1-C10 alkyl ester of (meth)acrylate. 20 -、C1-C 10 - or C1-C8-alkyl esters, including for example methyl acrylate, methyl methacrylate, ethyl acrylate, butyl acrylate, butyl methacrylate, 2-ethylhexyl acrylate, isobutyl (meth) acrylate, hexyl (meth) acrylate, lauryl (meth) acrylate, stearyl (meth) acrylate, cyclohexyl (meth) acrylate, benzyl (meth) acrylate, oleyl (meth) acrylate, palmityl (meth) acrylate, nonyl (meth) acrylate, decyl (meth) acrylate, dodecyl (meth) acrylate, pentadecyl (meth) acrylate, hexadecyl (meth) acrylate, octadecyl (meth) acrylate or mixtures thereof. Desirably, the alkyl esters of (meth) acrylic acid include methyl methacrylate, methacrylate, ethyl acrylate, butyl methacrylate, butyl acrylate, 2-ethylhexyl acrylate or mixtures thereof. The acrylic (co)polymer may contain structural units of alkyl (meth)acrylate at a concentration of 5 wt % to 100 wt %, and may be 15 wt % to 99 wt %, 30 wt % to 98 wt %, 50 wt % to 95 wt %, 55 wt % to 90 wt %, or 60 wt % to 90 wt %, based on the weight of the acrylic (co)polymer.

[0036] The acrylic (co)polymers useful in the present invention may or may not contain structural units of one or more ethylenically unsaturated monomers with at least one functional group selected from amide, urea, carbonyl, carboxyl, carboxylic anhydride, hydroxyl, sulfonic acid, sulfonate, phosphoric acid or phosphate ester groups; or their salts (hereinafter referred to as "functional monomers"). Suitable functional monomers may include α, β-ethylenically unsaturated carboxylic acids, which include acid-containing monomers such as methacrylic acid, acrylic acid, itaconic acid, maleic acid or fumaric acid; or monomers with acid-forming groups that produce or can subsequently be converted to such acid groups (such as anhydrides, (meth)acrylic anhydride or maleic anhydride); phosphorus-containing monomers such as vinylphosphonic acid, allylphosphonic acid, (meth)acrylic acid alkyl phosphates such as (meth)acrylic acid ethyl ester, (meth)acrylic acid phosphopropyl ester, (meth)acrylic acid phosphobutyl ester; The functional monomers include acrylamide, acrylic acid, methacrylic acid, phosphoethyl (meth)acrylate, sodium salt of 2-acrylamido-2-methyl-1-propanesulfonic acid, and ammonium salt of 2-acrylamido-2-methyl-1-propanesulfonic acid; sodium vinyl sulfonate; sodium salt of allyl ether sulfonic acid; monomers with carbonyl-containing groups, such as acetoacetoxyethyl methacrylate (AAEM) and diacetone acrylamide (DAAM); acrylamide; methacrylamide; or mixtures thereof. Desirably, the functional monomers include acrylamide, acrylic acid, methacrylic acid, phosphoethyl (meth)acrylate, sodium salt of 2-acrylamido-2-methyl-1-propanesulfonic acid, or mixtures thereof. The acrylic (co)polymer may contain structural units of the functional monomer at a concentration of 0 wt % to 10 wt %, based on the weight of the acrylic (co)polymer, and may be 0.1 wt % to 8 wt %, 0.3 wt % to 5 wt % or 0.5 wt % to 3 wt % or 0.7 wt % to 2 wt %.

[0037] The acrylic (co)polymers useful in the present invention may or may not contain structural units of one or more additional ethylenically unsaturated nonionic monomers in addition to the (meth)acrylic acid alkyl esters and the above-mentioned functional monomers. The term "nonionic monomer" refers to a monomer that does not carry an ionic charge between pH = 1-14. Suitable additional ethylenically unsaturated nonionic monomers may include vinyl aromatic monomers such as styrene and substituted styrenes (including, for example, α-methylstyrene, p-methylstyrene, tert-butylstyrene, vinyltoluene), glycidyl (meth)acrylate, ethylenically unsaturated monomers with at least one alkoxysilane functional group, including vinyltrialkoxysilanes such as vinyltrimethoxysilane and (meth)acryloyloxyalkyltrialkoxysilanes such as (meth)acryloyloxyethyltrimethoxysilane and (meth)acryloyloxypropyltrimethoxysilane; α-olefins such as ethylene, propylene and 1-decene; vinyl acetate, vinyl acetate, vinyl butyrate, vinyl versatate and other vinyl esters; acrylonitrile; or mixtures thereof. Desirably, additional ethylenically unsaturated nonionic monomers include styrene, vinyl acetate or mixtures thereof. The acrylic (co)polymer may or may not contain structural units of multifunctional nonionic monomers such as butadiene, divinylbenzene and allyl (meth)acrylate, typically in a concentration of 0 to 5 wt % based on the weight of the acrylic (co)polymer, and may be 0 to 2 wt %, 0.1 to 1 wt % or 0.1 to 0.5 wt %.

[0038] The acrylic (co)polymers useful in the present invention may contain structural units of additional ethylenically unsaturated nonionic monomers at a concentration of 0% to 95% by weight, based on the weight of the acrylic (co)polymer, and may be 1% to 85%, 2% to 70%, 5% to 50% or 10% to 45% by weight. Desirably, the acrylic (co)polymer may contain structural units of styrene at a concentration of 0% to 60% by weight, based on the weight of the acrylic (co)polymer, and may be 5% to 55%, 10% to 50% or 20% to 40% by weight. Alternatively, the acrylic (co)polymer may contain structural units of vinyl acetate at a concentration of 0% to 95% by weight, based on the weight of the acrylic (co)polymer, and may be 5% to 90%, 10% to 85%, 20% to 80% or 50% to 70% by weight.

[0039] The acrylic (co)polymer that can be used in the present invention can be a vinyl acrylic (co)polymer comprising 5% to 30% of structural units of alkyl (meth)acrylates, 70% to 95% of structural units of vinyl acetate, and 0% to 5% of structural units of functional monomers. Alternatively, the acrylic (co)polymer is a styrene acrylic copolymer comprising 30% to 60% of structural units of styrene, 40% to 70% of structural units of alkyl (meth)acrylates, and 0% to 5% of structural units of functional monomers. Alternatively, the acrylic (co)polymer can comprise 95% to 100% of structural units of alkyl (meth)acrylates and 0% to 5% of structural units of functional monomers.

[0040] The aqueous polymer composition useful in the present invention may contain one or more types of VOCs. Depending on the synthesis process such as the polymerization process used to prepare the aqueous polymer composition, the VOCs in the aqueous polymer composition may be selected from aldehydes, ketones, acrylic monomers, alcohols, saturated esters, ethers, aromatic hydrocarbons or mixtures thereof.

[0041] The aqueous polymer composition may be a binder emulsion or an aqueous coating composition. Commercially available aqueous polymer compositions include, for example, PRIMAL TM DC-420 emulsion, PRIMAL TM DC-430V emulsion, PRIMAL TM SF-155 emulsion, PRIMAL TM DC-460 emulsion, PRIMAL TM DC-480 emulsion, PRIMAL TM LE-328V emulsion, PRIMAL TM AS-2010 emulsion, PRIMAL TM AS-356 emulsion, PRIMAL TM SF-508 emulsion, PRIMAL TM SF-105 emulsion, PRIMAL TM SF-308 emulsion, PRIMAL TM LE-318V emulsion or mixtures thereof (PRIMAL is a trademark of The Dow Chemical Company).

[0042] The aqueous polymer composition that can be used for the present invention may contain or not contain pigment, extender or their mixture. Pigment may include particulate inorganic material that can make substantial contribution to the opacity or hiding power of coating. The refractive index of such material is usually greater than 1.8. The example of suitable pigment includes titanium dioxide (TiO2), zinc oxide, zinc sulfide, iron oxide, barium sulfate, barium carbonate or their mixture. The aqueous polymer composition may contain or not contain one or more extenders. Extender may include particulate inorganic material that is usually less than or equal to 1.8 and greater than 1.5 refractive index. The example of suitable extender includes calcium carbonate, aluminum oxide (Al2O3), clay, calcium sulfate, aluminosilicate, silicate, zeolite, mica, diatomaceous earth, solid or hollow glass, ceramic beads and opaque polymer (such as ROPAQUE available from Dow Chemical Company) TM Ultra E opaque polymer (ROPAQUE is a trademark of The Dow Chemical Company), or mixtures thereof. The pigment and / or extender may be present in a concentration of 0 wt % to 40 wt %, 5 wt % to 30 wt %, 10 wt % to 25 wt %, or 15 wt % to 20 wt %, based on the weight of the aqueous polymer composition. The aqueous polymer composition may also include or not include any one or combination of the following additives commonly used for coating application: defoamers, thickeners, dispersants, biocides, and coalescing agents.

[0043] Analytical characterization

[0044] The method of the present invention includes characterizing an aqueous polymer composition with a detector analysis, thereby generating concentration data of VOCs in the aqueous polymer composition. "Concentration data of VOCs", also referred to as "VOC concentration data", refers to a plurality of concentration values ​​of VOCs in the aqueous polymer composition. Alternatively, the concentration data may include concentration values ​​of those VOCs, each of which has a concentration of 0.1 parts per million (ppm) or more, 0.2 ppm or more, 0.3 ppm or more, or even 0.5 ppm or more of the weight of the aqueous polymer composition. The concentration of VOCs refers to the weight concentration of VOCs in ppm based on the weight of the aqueous polymer composition.

[0045] Surprisingly, it was also found that the concentration of some specific VOCs (such as compounds having 1 to 11 carbon atoms and a normal boiling point below 220°C at a pressure of 101.3 kPa) plays a crucial role in the prediction accuracy of the odor intensity of such waterborne polymer compositions. Desirably, the concentration data input to the decision tree ensemble includes the concentration of VOCs selected from the following: acetone, 2-methylpropanol, 1-butanol, methyl methacrylate, butyl acetate, 4-heptanone, 2-heptanone, butyl ether, styrene, butyl acrylate, anisole, propionic acid, butyl ester, methyl ethylbenzene, 3-methyl-4-heptanone, propenylbenzene, propylbenzene, benzaldehyde, acetophenone, butyl methacrylate, isobutyl vinyl acetate, butyric acid, butyl ester, 2-butenoic acid, butyl ester, diethylbenzene isomers, cyclohexyl methacrylate (2-propenoic acid, 2-methyl-, cyclohexyl ester), 2-ethylhexyl acrylate, xylene, ethylbenzene or mixtures thereof. More desirably, the concentration data comprises or consists of concentrations of VOCs selected from the group consisting of 2-methylpropanol, methyl methacrylate, butyl acetate, 2-heptanone, 3-methyl-4-heptanone, styrene, butyl acrylate, propionic acid, butyl ester, 2-ethylhexyl acrylate, ethylbenzene, xylene, benzaldehyde, or mixtures thereof.

[0046] Analytical characterization of aqueous polymer compositions can include analytical characterization techniques with a lower limit of detection (LLOD) <0.1ppm. LLOD (also referred to as analytical sensitivity) is the minimum amount of an analyte that can be reliably detected. Common analytical characterization techniques can be gas chromatography (GC) coupled with different detectors, such as GC-MSD (mass selective detector) (which is used interchangeably with "GC-MS" hereinafter), GC-FID (flame ionization detector) and GC-ECD (electron capture detector). GC-MS typically includes a GC instrument and an MSD. The detector in the GC-MS can be used to identify VOCs by the MS spectrum of the VOCs, and can also be used to obtain the peak area of ​​the VOCs to further quantify their concentrations. By comparing the peak area of ​​the VOCs in the aqueous polymer composition with the peak area of ​​an external standard, the concentration of each VOC can be obtained. Conventional GC-MS can be used, such as the Agilent 6890 gas chromatograph coupled with the Agilent 5975C MSD. Various sample preparation techniques for GC-MS can be used, including, for example, solid phase microextraction (SPME) coupled to GC-MS (also referred to as "SPME GC-MS"); needle trap microextraction (NTM) coupled to GC-MS; and Tenax cartridges (TC) coupled to GC-MS. Conventional NTM GC-MS and TC GC-MS are available from Shinwa Ltd. (Japan) and Gerstel Co., Ltd. (Germany), respectively.

[0047] Desirably, SPME coupled to a GC-MS comprising SPME is coupled to a GC equipped with an MSD instrument for extracting, identifying and quantifying VOCs from aqueous polymer compositions. SPME is a sample preparation technique for integrating operations including sample collection, extraction, and enrichment of analytes from the sample headspace. The SPME technique in the present invention is coupled to a GC and can be used to extract analytes (such as VOCs) from liquid samples such as aqueous polymer compositions. The usual procedure for SPME includes two steps: (1) distributing the analyte between the extraction phase and the sample matrix, and (2) subsequently desorbing the concentrated extract into an analytical instrument such as a GC-MS. The SPME procedure can be performed manually or automatically. For example, a multipurpose sampler (MPS) can be used to automate the SPME procedure.

[0048] Decision Tree Ensemble

[0049] The VOC concentration data generated by the analytical characterization is used as input to a decision tree ensemble configured to predict odor intensity of the waterborne polymer composition based on the VOC concentration data, for example, the decision tree ensemble is trained to predict odor intensity.

[0050] A decision tree is a model of decisions and their possible outcomes (such as the probability or predicted value of an event outcome) using a tree structure. A decision tree model (an example of a supervised learning model) is trained using input data paired with output data by optimizing the model parameters to minimize the difference between the model's predicted values ​​and the actual values.

[0051] Then, given a set of input data for a test sample (also referred to as a "new sample"), the trained model can predict a new output. A "decision tree ensemble" (also referred to as a "decision tree ensemble model") refers to a model that combines multiple decision trees. "Multiple" means two or more, and can be 10 or more, while typically 1000 or less or 100 or less.

[0052] Figure 1 A diagram showing a simplified example of a decision tree with p features as input X = {x1, x2…x p}The decision process of predicting y. For example, in a decision tree using two features x1 and x2, each node is split into different branches based on the condition of x1 or x2. Based on the condition of x1, the decision tree is first split into 3 branches with different ranges of x1. When x1>0.5, the decision tree output is y=5. When x1<0.5, there is another branch split based on the condition of x2. In these new branches, under the condition of x1<0.5, depending on the different conditions of the value of x2, the predicted value of y can be 2, 3, or 4. For example, when x1<0.5 and x2>0.6, the predicted y is 4.

[0053] Decision trees can be built for regression by using standard deviation reduction to produce leaf nodes. In decision tree regression tasks (where the predicted y is a continuous value), decision trees are built top-down from the root node to the leaf nodes and involve partitioning data with similar values ​​by standard deviation. Decision trees have two or more branches, and each branch represents the value of the attribute being tested. Each leaf node represents a decision about the target. The count m represents the total number of points in the node, Compute attribute Z = {z1,z2…z m}, as follows:

[0054]

[0055] Standard deviation (S) for tree branching:

[0056]

[0057] The coefficient of variation (CV) is used to stop branching:

[0058]

[0059] The sum of the standard deviations of multiple attributes (S(target,C)) can be calculated by combining the standard deviations of the individual standard deviations (S) with the probability (P) of condition (c) occurring under all conditions of the decision (C):

[0060]

[0061] Standard Deviation Reduction (SDR) can be achieved after splitting the dataset on features. A decision tree is built to find the highest SDR, which is calculated by the difference between the node standard deviation before differentiation (S(target)) and the node standard deviation after differentiation (S(target,C)):

[0062] SDR(target,C)=S(target)-S(target,C)

[0063] Stop branching after the CV reaches a specified threshold. Sometimes branching is also limited by other criteria, including the maximum depth of the tree and the minimum number of samples per node.

[0064] The decision tree ensemble method combines several decision trees to produce better prediction performance than using a single tree. The decision tree ensemble used in the present invention is trained to predict the odor intensity of a sample by establishing a correlation model from VOC concentration to odor intensity rated by human sensory panelists (hereinafter also referred to as "actual odor intensity") during the training phase. During the training phase, the concentration of VOCs in the training samples is used as the input of the decision tree ensemble model, and the actual odor intensity of these training samples rated by the panelists is used to guide the learning of the output to minimize the difference between the actual value and the predicted value. The trained decision tree ensemble model can be used to predict the odor intensity based on the VOC concentration of new samples.

[0065] First, the concentration of each of the selected VOCs for the total p features is collected: {P1,P2…P p}, where each feature group P contains n samples: P = {p1, p2…p n}, these samples are rescaled so that all values ​​fall on the interval [0,1]:

[0066]

[0067] where max(P) is the maximum concentration of a VOC in all training samples, and min(P) is the minimum concentration of this VOC in all training samples.

[0068] The decision tree ensemble model takes the rescaled input from the training data and predicts the odor intensity as output. i With the predicted value The mean square error (MSE) between them is used to optimize the model parameters, as shown in the following equation (I):

[0069]

[0070] Where n is the total number of data points, y i is the actual value of the odor intensity rated by the human panelists, i is a sample data point, and is the model-predicted value for odor intensity.

[0071] MSE is a metric used to measure the average of the squares of the differences between the predicted values ​​and the actual values. RMSE stands for Root Mean Squared Error, which is a metric used to measure the differences between the predicted values ​​and the actual values. The RMSE percentage is the square root of the MSE normalized by the population mean into a dimensionless number expressed as a percentage. The RMSE percentage can be calculated as follows:

[0072]

[0073] Where n, y i , i and As defined in equation (I) above, and is the average of the actual odor intensities rated by the human panelists. The lower the RMSE percentage, the better the model fits the dataset.

[0074] R 2 represents the discriminant coefficient, which is a common performance measure for regression, calculating the proportion of variance explained by the regression model. Usually R 2 The range is from 0 to 1 and can be calculated according to the following equation (III):

[0075]

[0076] where y i , i and As defined in equation (I) above, and is the average of the actual odor intensities rated by human panelists. 2 The higher it is, the better the model fits the dataset.

[0077] A plurality of decision tree-based ensemble learning models may be used in the present invention, including, for example, a random forest (RF) model, a gradient boosting (GB) model, and an extreme gradient boosting (XGB) model.

[0078] The RF model uses over bagging, which creates several data subsets from the training samples and performs random selection of training features to develop many decision trees with the selected samples and features. The decision trees are averaged to obtain the final RF model. The RF model can effectively reduce the variance of a single decision tree.

[0079] The GB model uses boosting, another ensemble technique for combining multiple weak learners to produce a strong learner. In the GB model, each individual decision tree is learned sequentially, where early learners learn a simple model of the data, and later learners analyze the prediction errors by learning from the errors of early learners. It uses a gradient descent algorithm that allows a differentiable loss function to optimize the weights of sequential learners to recover the difference between the actual value and the predicted value.

[0080] The XGB model provides several adjustments to the GB algorithm, including changing the gradient descent algorithm in the GB algorithm to the Newton-Raphson optimization algorithm. In addition to the original GB algorithm, the XGB model also uses additional randomization parameters, proportional shrinkage in leaf nodes, tree penalties, and feature selection.

[0081] The RF model, GB model, and XGB model can be established in a computing environment such as Python, R, or MATLAB. These models are available in open source libraries such as the Scikit-learn library. For example, the RF and GB models can be established with the Scikit-learn package, as described in "Scikit-learn: Machine Learning in Python", Pedregosa, F. et al., Journal of Machine Learning Research, 2011, 12, 2825-2830. The XGB model can be established with the XGBoost package, such as "XGBoost: A Scalable Tree Boosting System", Chen, T. and Guestrin, C., Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. nd ACM SIGKDDInternational Conference on Knowledge Discovery and Data Mining), 2016, 785–794, Association for Computing Machinery (ACM), New York, USA.

[0082] Hyperparameters for decision tree ensemble models can be default settings in machine learning model libraries such as the Scikit-learn library. A "hyperparameter" is a defined value used to control the learning process that is set before the training process of the model begins. Hyperparameters are different from other parameters of the model because other parameters are learned during the training process. Examples of hyperparameters include the number of estimators, which indicates how many trees are used for the ensemble learning model, and the minimum number of samples per split, which indicates the criteria for making a split in a decision tree model.

[0083] A decision tree ensemble (i.e., a trained decision tree ensemble) useful in the present invention is trained to predict the odor intensity of an aqueous polymer composition using a training data set using a plurality of training samples such as concentration data of an aqueous composition containing VOCs. The training data set comprises VOC concentration data for each training sample (as input) paired with corresponding actual odor intensity data for such training sample as rated by a human panelist (as output) to yield a trained decision tree ensemble. A validation data set may be utilized to validate the training results obtained using validation samples to monitor model prediction accuracy. The decision tree ensemble model may also be tested using a test data set using another set of test samples that are prepared separately at a different time than the samples used to generate the training data set. Models that meet certain thresholds for training performance, validation accuracy, and test accuracy are selected.

[0084] In the present invention, the concentration of the selected VOC in each training sample (such as the above-mentioned waterborne polymer composition or the coating described below) is input into the decision tree ensemble model to train the model. Desirably, the training data set contains concentration data of volatile organic compounds, each of which has a concentration of ≥0.1 ppm by weight based on the weight of the training sample (i.e., the training waterborne polymer composition). More desirably, the training data set contains concentration data of volatile organic compounds including acetone, 2-methylpropanol, 1-butanol, methyl methacrylate, butyl acetate, 4-heptanone, 2-heptanone, butyl ether, styrene, butyl acrylate, anisole, propionic acid, butyl ester, methyl ethyl benzene, 3-methyl-4-heptanone, propenyl benzene, propyl benzene, benzaldehyde, acetophenone, butyl methacrylate, isobutyl vinyl acetate, butyric acid, butyl ester, 2-butenoic acid, butyl ester, diethyl benzene or isomers, cyclohexyl methacrylate, 2-ethylhexyl acrylate, xylene, ethyl benzene or mixtures thereof. The training samples are analyzed and characterized using methods such as GC-MS (as input) to determine the concentration of VOCs, and the odor intensity is evaluated based on human sensory results (i.e., the evaluation of human panelists), also known as "actual odor intensity". The actual odor intensity of the training samples can be evaluated by an odor panel test according to the VDA 270 standard on a scale of 1 to 6, where 1 = not perceptible; 2 = perceptible, not disturbing; 3 = clearly perceptible, but not disturbing; 4 = disturbing; 5 = strongly disturbing; and 6 = unacceptable (further details are provided in the Examples section below). The decision tree ensemble model is trained to simulate the correlation between the VOC concentration data and the odor intensity of the waterborne polymer composition as rated by the panelists. The decision tree ensemble model in the present invention can be trained to provide, for example, the odor intensity of the waterborne polymer composition as rated by the panelists by training R 2>0.85, as calculated by equation (III) above. The trained decision tree ensemble model can be further validated using a validation dataset using multiple validation samples. The validation dataset contains VOC concentration data of volatile organic compounds in each validation sample (as input) and corresponding actual odor intensity data of such validation samples rated by human panelists (as output).

[0085] The trained decision tree ensemble model can then be used to predict the odor of an aqueous polymer composition tested as a test sample (also referred to as a "new sample") using a test data set that contains concentration data of VOCs in the test sample obtained from analytical characterization. A user can input the concentration data of the VOCs of the test sample into the trained decision tree ensemble through a web-based user interface (details are described below).

[0086] In one embodiment of the present invention, a decision tree ensemble model is trained using a total of 39 data pairs containing actual odor intensities rated by human panelists, split into a training data set with 31 data pairs and a validation data set with 8 data pairs. A "data pair" herein refers to the VOC concentration data of a particular sample and its corresponding odor intensity, as rated by the panelists. The model is then evaluated or tested using a new data set collected in a new batch of experiments to assess the predictive performance of the model.

[0087] Figure 2 A flow chart of an odor prediction method according to an example of the present invention is shown. The method includes extracting raw data by identifying the VOC type to be analyzed, obtaining a VOC profile containing VOC concentration data, and collecting odor intensity data from an odor panel test (more details are provided in the Examples section below). The odor intensity data is used together with the concentration data as a data pair for model development. The method also includes training and validating multiple models using the obtained concentration data paired with the odor intensity data, and then based on the training R 2 The desired model criteria of training RMSE and validation RMSE are used to select an appropriate model. Underfitting is a situation where the model cannot capture the relationship between the input and output variables. Underfit models tend to have undesirably low training RMSE. 2 . Training R 2 Threshold, that is, training R 2 >0.85, is used to filter out underfit models. However, overfitting of a model is when the model is too closely aligned to the training data, and the learned representation cannot accurately predict the validation data. A validation RMSE percentage threshold, i.e., validation RMSE percentage <30%, is used to filter out overfit models. After the threshold is met, the trained model is used to predict the odor intensity of new samples.

[0088] The method of the present invention may also include adjusting the synthesis process of the aqueous polymer composition based on the predicted odor intensity, such as adjusting the polymerization process, particularly the emulsion polymerization process, used to prepare the polymer in the aqueous polymer composition. The parameters that can be adjusted may include, for example, surfactant type and amount, initiator type and amount, monomer type and source, reaction temperature, steam stripping parameters, or a combination thereof.

[0089] coating

[0090] The present invention also relates to a method for predicting the odor of a coating (also referred to as a "coating film"). The method includes characterizing the coating with a detector analysis to generate concentration data of VOCs in the coating, inputting the concentration data into a decision tree ensemble configured to predict the odor intensity of the coating based on the VOC concentration data; and outputting the predicted odor intensity of the coating from the decision tree ensemble. The coating can be obtained by drying or allowing the above-mentioned aqueous polymer composition (i.e., a dried aqueous polymer composition). The aqueous polymer composition can be applied to a substrate, and the applied polymer composition can be dried or allowed to dry to form a coating. The coating can have a dry film thickness of 50 μm to 60 μm. The aqueous polymer composition can be used alone or in combination with other coatings to form a multilayer coating. The aqueous polymer composition can be applied to the substrate by existing methods including brushing, dipping, rolling and spraying. Drying can be carried out at room temperature (20°C to 25°C) or at an elevated temperature (e.g., 35°C to 60°C). Due to drying, the coating may contain a different VOC profile than the above-mentioned aqueous polymer composition, because some VOCs in the aqueous polymer composition can evaporate from the composition after drying. The coating is suitable for different coating applications, such as architectural coatings, wood coatings, and protective coatings. The analytical characterization of VOC concentration data that can be used to generate the coating is the same as the analytical characterization of the aqueous polymer composition described above. The decision tree ensemble in the method for predicting the odor of the coating is as described above in the method for predicting the odor of the aqueous polymer composition, except that the training samples, validation samples, and test samples used in the decision tree ensemble are coating samples obtained by drying the aqueous polymer composition.

[0091] In particular, the novel combination of VOC concentration data as input with a specific machine learning model (i.e., a decision tree ensemble) enables the method of the present invention to predict the odor intensity of an aqueous polymer composition with higher accuracy than other machine learning models such as an artificial neural network (ANN) model, a partial least squares regression (PLS) model, or a ridge regression (RR) model. The ANN model is based on a collection of connected units or nodes called artificial neurons for predictive modeling. The PLS model finds a linear regression model by projecting the predictor variables and observable variables into a new space. The RR model establishes a multiple regression model in situations where the independent variables are highly correlated by creating a ridge regression estimator.

[0092] In one embodiment, the method can provide high prediction accuracy for predicting the odor of an aqueous polymer composition, such as the training R calculated according to equation (III) above. 2 >0.85 and the test RMSE percentage calculated according to equation (II) above is less than 30% (<30%). In addition, the validation RMSE percentage is <30%, as calculated according to equation (II) above. 2 ” refers to the R of the training dataset 2 , "Validation RMSE percentage" refers to the RMSE percentage of the validation dataset, and "Test RMSE percentage" refers to the RMSE percentage of the test dataset. In particular, when the RF model is used, this method can provide even higher prediction accuracy, showing a test RMSE percentage of less than 20%.

[0093] A system for odor prediction

[0094] The present invention also relates to a system for predicting the odor of an aqueous polymer composition or coating. The system of the present invention includes: a detector configured to analyze and characterize the aqueous polymer composition or the coating, thereby generating concentration data of volatile organic compounds in the aqueous polymer composition or the coating by the analytical characterization; and a computing device, on which a decision tree ensemble is deployed, and the computing device is configured to input the concentration data and output a predicted odor intensity. The decision tree ensemble and the training decision tree ensemble are as described above. For example, using a training data set, a decision tree ensemble deployed on a computing device is trained using multiple training samples, wherein the training data set, the training samples, and the resulting trained decision tree ensemble are as described above. The analytical characterization and the detector are as described above, such as GC-MS, and desirably, SPME GC-MS.

[0095] A computing device useful in the present invention may include a processor and a data storage device, wherein the data storage device has stored thereon computer executable instructions which, when executed by the processor, cause the computing device to perform functions including inputting VOC concentration data into a decision tree ensemble and outputting predicted odor intensity.

[0096] The computing device that can be used for the present invention can be a client device (e.g., a device actively operated by a user), a server device (e.g., a device that provides computing services to client devices), or some other type of computing platform. Some server devices can operate as client devices from time to time to perform specific operations, and some client devices can incorporate server functions.

[0097] Processors useful in the present invention may be in the form of one or more of any type of computer processing element, such as a central processing unit (CPU), a coprocessor (e.g., a math, graphics, neural network, or cryptographic coprocessor), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a network processor, and / or an integrated circuit or controller that performs processor operations.

[0098] The data storage device may include one or more data storage arrays including one or more drive array controllers configured to manage read and write access to a group of hard disk drives and / or solid state drives.

[0099] In some embodiments, computing devices can be deployed to support cluster architectures. The exact physical location, connectivity, and configuration of these computing devices may be unknown and / or unimportant to client devices. Therefore, the computing devices can be referred to as "cloud-based" devices, which can be housed at various remote data center locations, such as cloud-based server clusters. Desirably, the computing devices are cloud-based server clusters, and inputting concentration data into the decision tree ensemble is performed via a web-based user interface that a user can access.

[0100] Figure 3A schematic diagram of a cloud-based server cluster 300 according to one example of the present invention is depicted. Desirably, the operation of the computing device can be distributed between server devices 302, data storage devices 304, and routers 306, all of which can be connected via a local cluster network 308. The amount of server devices 302, data storage devices 304, and routers 306 in the server cluster 300 can depend on the computing tasks and / or applications assigned to the server cluster 300. For example, the server device 302 can be configured to perform various computing tasks of the computing device. Therefore, computing tasks can be distributed between one or more server devices in the server device 302. As an example, the data storage device 304 can store any form of database, such as a structured query language (SQL) database or a trained model checkpoint. In addition, any database in the data storage device 304 can be integral or distributed across multiple physical devices. The router 306 can include a network device configured to provide internal and external communications to the server cluster 300. For example, router 306 may include one or more packet switching and / or routing devices (including switches and / or gateways) configured to (i) provide network communications between server device 302 and data storage device 304 via cluster network 308, and / or (ii) provide network communications between server cluster 300 and other devices via communication link 310 to network 312. Server device 302 may be configured to transmit data to cluster data storage device 304 and receive data from the cluster data storage device. In addition, server device 302 may organize the received data into a web page representation. Such a representation may take the form of a markup language, such as hypertext markup language (HTML), extensible markup language (XML), or some other standardized or proprietary format. In addition, server device 302 may have the ability to execute various types of computerized scripting languages, such as Perl, Python, PHP hypertext preprocessor (PHP), active server pages (ASP), JavaScript, etc. Computer program code written in these languages ​​may facilitate providing web pages to client devices and interaction of client devices with web pages.

[0101] Figure 4A schematic block diagram is shown, which shows an odor prediction system 400 according to an example of the present invention. Desirably, the odor prediction system 400 includes a GC-MS 401 for determining the concentration of volatile organic compounds (VOCs) in a sample, and a cloud-based computing platform 402, such as a cloud-based server cluster. The trained decision tree ensemble model is deployed in the cloud-based computing platform 402 to provide web-based access to end users. The GC-MS 401 includes a sample injector 4011, a GC column 4012, and an MSD 4013. The sample is first analyzed and characterized using the GC-MS 401 to generate VOC concentration data from the sample, and then the resulting concentration data is input into the cloud-based computing platform 402. The odor prediction system 400 may also include a sampling unit 403, which includes a headspace vial 4032 and an SPME fiber 4033. Desirably, a sample 4031 (e.g., an aqueous polymer composition or coating) is added to a headspace vial 4032, and then an SPME fiber 4033 is inserted into the headspace vial 4032 and exposed to the headspace of the vial to extract VOCs in the sample. The VOCs extracted from the sample 4031 are then injected into the sample injector 4011 and analyzed in the GC-MS 401 to generate a VOC profile. The obtained VOC profile is used as input and uploaded to a cloud-based computing platform 402 deployed with the model via a web-based user interface. The model then predicts the odor intensity of each sample based on its VOC profile.

[0102] Example

[0103] Some embodiments of the invention will now be described in the following examples.PRIMAL, FORMASHIELD and RHOPLEX are trademarks of The Dow Chemical Company.

[0104] Methyl methacrylate, butyl acetate, xylene (containing 3 isomers), butyl ether, styrene, butyl acrylate, propionic acid, butyl ester, 3-methyl-4-heptanone and benzaldehyde can be obtained from Sinopharm Chemical Reagent Co., Ltd.

[0105] Thirty-nine waterborne acrylic binders commercially available from various sources (including PRIMAL TM 、FORMASHIELD TM and RHOPLEX TM emulsion, ACRONAL emulsion available from BASF, ARCHSOL available from Wanhua TMEmulsion and RS series emulsion available from BATF) were used as training and verification samples, where 31 samples were randomly selected as training samples and 8 samples were randomly selected as verification samples. These samples are randomly marked with numbers "1" to "39" as sample codes in Tables 2-4.

[0106] Another six acrylic adhesives commercially available from various sources as training samples and validation samples described above were used as new samples for the test model and were randomly labeled with numbers “41” to “46” as sample codes in Table 5.

[0107] PRIMAL containing 100% acrylic acid polymer TM SF-180 emulsion, PRIMAL TM DC-430V Styrene Acrylic Copolymer Emulsion and PRIMAL TM SF-508M styrene acrylic emulsion is commercially available from The Dow Chemical Company.

[0108] The following standard analytical equipment and methods were used to implement and determine the properties and characteristics described herein:

[0109] SPME coupled with GC-MS analysis

[0110] 1) Preparation of external standard

[0111] By mixing methyl methacrylate, butyl acetate, xylene (containing 3 isomers), butyl ether, styrene, butyl acrylate, propionic acid, butyl ester, 3-methyl-4-heptanone and benzaldehyde with PRIMAL TM SF-180 emulsion was mixed to prepare the standard mixture, and each of these chemical compounds was added as PRIMAL TM The prepared standard mixture was further diluted to different concentrations of each chemical compound for use as external standards for the quantification of different VOCs, such as those based on PRIMAL TM Wet weight of SF-180 emulsion, 0.005ppm, 0.01ppm, 0.05ppm, 0.1ppm, 0.5ppm, 1ppm, 5ppm, 10ppm, 50ppm, 100ppm and 500ppm.

[0112] 2) Automated SPME GC-MS test parameters

[0113] Approximately 0.05 g of aqueous binder sample was weighed into a 20 mL headspace vial.

[0114] A multipurpose sampler (MPS) (Zeste Co., Ltd.) with an SPME unit was coupled to the GC. The SPME fiber available from Anple was polydimethylsiloxane / divinylbenzene (PDMS / DVB, 65 mm, catalog 57345-U, Supelco Co., Ltd.). The parameters of the SPME treatment were as follows: incubation temperature: 60°C, extraction time: 30 minutes, desorption time: 90 seconds.

[0115] Based on the conditions listed in Table 1, GC-MS analysis was performed using an Agilent 6890 gas chromatograph coupled to a mass spectrometer detector (Agilent 5975C MSD).

[0116] Table 1. GC-MS conditions

[0117]

[0118]

[0119] Odor Panel Test

[0120] The test was performed according to VDA 270, where the experimental procedure and sample preparation are described below:

[0121] In a sensory test, three test samples were prepared for evaluation by each panelist. Two benchmark marker samples PRIMAL were used in each sensory test. TM DC-430V and PRIMAL TM SF-508M was used as a reference. According to the criteria described in the following VDA 270 standard, PRIMAL TM SF-508M and PRIMAL TM The odor intensity values ​​of the DC-430V emulsion were rated as '1.5' and '5.0' respectively.

[0122] Aliquots of 0.5g of the binder sample are placed in 100mL glass vials with a tasteless cap. Before the assessment, the vials containing the sample were balanced at room temperature for 2 hours. Each panelist accepts a group of three samples independently. Randomly selected marks are assigned to each sample for blind sample identification. The order in which the sample is presented to the panelist is also random.

[0123] For each test, eight to ten trained panelists (certified by SGS. Co., Ltd. for odor intensity training) were invited to perform sensory evaluation on these samples. The variance analysis of the sensory evaluation was automatically calculated by CSAS (Conventional Sensory Analysis System) sensory software (ISENSO Co., Ltd., Shanghai, China).

[0124] The odor intensity is evaluated and rated based on the criteria described in VDA 270 (minimum unit is 0.5): 1. Imperceptible; 2. Perceptible, not disturbing; 3. Clearly perceptible, but not disturbing; 4. Disturbing; 5. Strongly disturbing; 6. Unacceptable.

[0125] For each adhesive sample, the average of the odor intensities rated by the panelists is reported as "Actual Odor Intensity."

[0126] Model

[0127] The random forest (RF) model, gradient boosting (GB) model, artificial neural network (ANN) model, partial least squares regression (PLS) model and ridge regression (RR) model used in the following embodiments (IE 1-3 and CE 1-3) are each independently constructed based on Python Scikit-learn (more details can be found in "Scikit-learn: Machine Learning in Python", Pedregosa, F. et al., Journal of Machine Learning Research, 2011, 12, 2825-2830.). Each of these models was independently established using Scikit-learn 0.23.2 in a Python 3.8 environment.

[0128] The extreme gradient boosting (XGB) model was built using the XGBoost package version 1.4.2 and Python 3.8 environment (for more details see “XGBoost: A scalable tree boosting system”, Chen, T. and Guestrin, C., Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016, 785–794, ACM, New York, USA).

[0129] Each model was trained with a total of 39 data pairs from the training samples, which were split into a training dataset with 31 data pairs (dataset type: "training") and a validation dataset with 8 data pairs (dataset type: "validation"). The input to the model was the concentration of 27 selected VOCs measured by GC-MS, and the output of the model was the odor intensity of the training samples as rated by human panelists according to the above-mentioned odor panel test. The results are given in Tables 2-4. In order to verify the effectiveness and accuracy of the method using each model, the VOC concentrations of six new samples (i.e., test samples) were measured and input into each training model. The predicted odor intensity values ​​were obtained and compared with the manual evaluation results of the odor intensity rated by human panelists in Table 5.

[0130] IE-1

[0131] An automated odor intensity prediction tool was established by combining SPME GC-MS, odor panel test, and RF model. The hyperparameters of the RF model are as follows:

[0132] Number of estimators = 100; minimum number of samples per split = 2; minimum sample per leaf node = 1; minimum weight score of a leaf node = 0; number of features for splits And the minimum value of impurity reduction = 0.

[0133] IE-2

[0134] IE-2 was conducted according to the same procedure as IE-1, except that the RF model was replaced by a GB model with the hyperparameters shown below:

[0135] Learning rate = 0.1; number of estimators = 100; fraction of samples used for a single base learner = 1; minimum number of samples per split = 2; minimum samples per leaf node = 1; minimum weight fraction per leaf node = 0; maximum depth of a single tree = 3; and minimum impurity reduction = 0.

[0136] IE-3

[0137] IE-3 was conducted according to the same procedure as IE-1, except that the RF model was replaced by an XGB model with the hyperparameters shown below:

[0138] Step shrinkage = 0.3; minimum impurity reduction = 0; maximum depth of a single tree = 6; minimum sum of instance weights = 1; maximum incremental step = 0; minimum samples per leaf node = 1; co-sampling by tree, co-sampling by leaf, co-sampling by node = 1; L2 regularization term for weights = 1; and L1 regularization term for weights = 0.

[0139] CE-1

[0140] CE-1 was conducted according to the same procedure as IE-1, except that the RF model was replaced by an ANN model with the hyperparameters shown below:

[0141] hidden_size = 100, activation = ReLU (rectified linear unit), solver = Adam, alpha = 0.0001, batch_size = number of samples, learning_rate = 0.001, max_number_of_iterations = 200, shuffle = true, early_stop = false, beta1 = 0.9, beta2 = 0.999, epsilon = 10 -8 .

[0142] CE-2

[0143] CE-2 was performed according to the same procedure as IE-1, except that the RF model was replaced by a PLS model with the following hyperparameters: number of components = 2, scaling = true, maximum number of iterations = 500, tolerance = 10 -6 , copy = true.

[0144] CE-3

[0145] CE-3 was conducted according to the same procedure as IE-1, except that the RF model was replaced by the RR model with the hyperparameters shown below:

[0146] α = 0.2, fit intercept = true, normalize = false, copy X = true, maximum iterations = 15000, tolerance = 10 -3 , Solver = Auto, Positive = False.

[0147] According to the methods described in the above embodiments (IE-1 to IE-3 and CE-1 to CE-3), the raw data (including VOC concentration and actual odor intensity) of the training samples and the validation samples and the model predictions are presented in Tables 2 to 4. Table 5 gives the odor evaluation results of the test samples. In these tables, the actual odor intensity refers to the odor intensity rated by the panelists according to the above odor panel test, and "\" means that the concentration is below the detection limit.

[0148] Table 2. Odor evaluation of training and validation samples

[0149]

[0150] Table 3. Odor evaluation of training and validation samples

[0151]

[0152] Table 4. Odor evaluation of training and validation samples

[0153]

[0154] Table 5. Odor evaluation of test samples

[0155]

[0156] Based on the data given in Tables 2 to 5, we train R 2 , validation RMSE percentage, and test RMSE percentage are calculated based on equations (II) and (III) above, and the results are given in Table 6. As shown in Table 6, the IE-1 (RF model), IE-2 (GB model), and IE-3 (XGB model) methods show higher training R than CE-1 (ANN model), CE-2 (PLS model), and CE-3 (RR model). 2 . The validation and test results of IE-1 to IE-3 methods all met the requirement of RMSE percentage <30%, while all CE-1 to CE-3 methods provided test RMSE percentages higher than 30%. This indicates that all IE-1 to IE-3 methods based on decision tree ensembles show higher prediction accuracy than CE-1 to CE-3 based on ANN, PLS and RR models, respectively. In particular, IE-1 using the RF model shows higher prediction accuracy (test RMSE percentage = 18.7%) than IE-2 (test RMSE percentage = 21.3%) and IE-3 (test RMSE percentage = 20.6%).

[0157] Table 6. Training R 2 , validation RMSE percentage and test RMSE percentage results

[0158]

Claims

1. A method for predicting the odor of an aqueous polymer composition, the method comprising: analytically characterizing the aqueous polymer composition using a detector, thereby generating concentration data of volatile organic compounds in the aqueous polymer composition from the analytical characterization; inputting the concentration data into a decision tree ensemble configured to predict an odor intensity of the aqueous polymer composition based on the concentration data; as well as A predicted odor intensity of the aqueous polymer composition is output from the decision tree ensemble.

2. The method of claim 1, wherein the decision tree ensemble is trained to predict the odor intensity of the waterborne polymer composition using a training data set using a plurality of training samples, wherein the training data set comprises the concentration data of the volatile organic compounds in each training sample, the concentration data being paired with actual odor intensity data of such training samples as rated by human panelists; thereby obtaining a trained decision tree ensemble.

3. The method of claim 2, wherein the trained decision tree ensemble is validated using a plurality of validation samples using a validation data set, wherein the validation data set comprises the concentration data of the volatile organic compounds in each validation sample, the concentration data being paired with actual odor intensity data of such validation samples as rated by a human panelist.

4. The method of claim 1, wherein the decision tree ensemble is selected from a random forest model, a gradient boosting model, or an extreme gradient boosting model.

5. The method of claim 2, wherein the training data set comprises the concentration data for volatile organic compounds, each of the volatile organic compounds having a concentration of ≥ 0.1 parts per million by weight based on the weight of the aqueous polymer composition.

6. The method of claim 1, wherein inputting the concentration data into the decision tree ensemble is performed via a web-based user interface.

7. The method of claim 1 , wherein the concentration data input to the decision tree ensemble is the concentration of volatile organic compounds, the volatile organic compounds comprising acetone, 2-methylpropanol, 1-butanol, methyl methacrylate, butyl acetate, 4-heptanone, 2-heptanone, butyl ether, styrene, butyl acrylate, anisole, propionic acid, butyl ester, methyl ethylbenzene, 3-methyl-4-heptanone, propenylbenzene, propylbenzene, benzaldehyde, acetophenone, butyl methacrylate, isobutyl vinyl acetate, butyric acid, butyl ester, 2-butenoic acid, butyl ester, diethylbenzene or isomers, cyclohexyl methacrylate, 2-ethylhexyl acrylate, xylene, ethylbenzene or mixtures thereof.

8. The method of claim 1, wherein analytical characterization of the aqueous polymer composition comprises analytical characterization selected from the group consisting of solid phase microextraction coupled to gas chromatography-mass spectrometry, needle trap microextraction coupled to gas chromatography-mass spectrometry, or Tenax cartridge coupled to gas chromatography-mass spectrometry.

9. The method of claim 2, wherein the decision tree ensemble exhibits a test RMSE percentage < 30% and a training R 2 >0.85 indicates prediction accuracy.

10. The method of claim 1, further comprising adjusting a polymerization process used to prepare the aqueous polymer composition based on the predicted odor intensity.

11. The method of claim 1, wherein the aqueous polymer composition comprises an acrylic acid (co)polymer.

12. A method for predicting the odor of a coating, the method comprising: analytically characterizing the coating using a detector, thereby generating concentration data of volatile organic compounds in the coating from the analytical characterization; wherein the coating is obtained by drying an aqueous polymer composition; inputting the concentration data into a decision tree ensemble configured to predict an odor intensity of the coating based on the concentration data; as well as A predicted odor intensity of the coating is output from the decision tree ensemble.

13. A system for predicting the odor of an aqueous polymer composition or a coating made therefrom, the system comprising: a detector configured to analytically characterize the aqueous polymer composition or the coating, thereby generating concentration data of volatile organic compounds from the analytical characterization; as well as A computing device having a decision tree ensemble deployed thereon, the computing device being configured to input the concentration data and output a predicted odor intensity of the aqueous polymer composition or the coating.

14. The system of claim 13, wherein the ensemble of decision trees deployed on the computing device is trained to predict the odor intensity of the aqueous polymer composition or the coating using a plurality of training samples using a training data set, wherein the training data set comprises the concentration data of the volatile organic compounds in each training sample, the concentration data being paired with actual odor intensity data of such training samples as rated by a human panelist; Thus, a trained decision tree ensemble is obtained.

15. The system of claim 13, wherein the computing device is a cloud-based server cluster and inputting the concentration data into the decision tree ensemble is performed via a web-based user interface.

Images