Predicting Formulation Component GC-MS Response Factor Using Quantitative Structure-activity Relationships Coupled with Artificial Neural Networks
By Jessica Lum, Madeline Schultz, and Erik Sapper, Department of Chemistry and Biochemistry, California Polytechnic State University
identification, measurement, and reduction of volatile organic compounds (VOCs) has been a key motivator in recent coatings research and development efforts. Analytical methods for determining VOC levels in organic coatings continue to improve, as chromatographic and spectroscopic approaches afford a means of quantifying VOC content directly in waterborne as well as solventborne coatings.
Heuristic methods for estimating the volatility of formulation components are common but are not extensively validated using quantitative structure-property relationships. Thus, a clearer link between component transport through an evolving coating matrix during curing processes, the bulk volatility of a compound, and the elution and quantification of compounds in a gas chromatograph (GC) still must be made to promote innovation in this area.
To address these issues, digital tools such as molecular descriptors and machine learning models are being combined with experimental measurements to better understand the time-dependent mechanistic nature of VOCs in coatings and to enable predictive control over the volatility and in-coating behavior of newly developed formulation components.
Here, we present the development and validation of a molecular structure-based neural network for the prediction of response factor for formulation components in a gas chromatography (GC) analysis. This represents an important step in creating large-scale computational design tools that enable in silico formulation, optimization, and end-use property prediction of environmentally benign coatings.
Consumer and market demand within the coatings industry continues to put pressure on formulators to create high-performance coatings that also have adequately low levels of volatile organic compounds. An ongoing challenge is the creation and optimization of important end-use coating properties while still meeting environmental regulation specifications.
As formulators are urged to innovate more quickly, it has become apparent that traditional empirical and Edisonian (guess-and-test) methods, even statistically designed methods of formulation discovery, must be augmented with newer technologies, such as those represented by digitization, automation, machine learning, and artificial intelligence.
There is also an increased emphasis on understanding chemical and physical interactions within the formulation at all stages of the paint production, application, and film-forming process. The growing consumer demand for environmentally benign “green” coatings has led to a push within the paint industry for improved predictive models and developmental workflows that make use of these next generation technologies.
Consider, for example, the recent South Coast Air Quality Management District (SCAQMD) Test Method 319 (Determination of Exclusion Status for Compounds in Film-Forming Coatings), where measurement, estimation, or prediction of the low vapor pressure of a formulation component may lead to its exclusion in VOC calculation and reporting.
Environmentally conscious consumers and regulatory agencies such as the U.S. Environmental Protection Agency (EPA) have continued to drive the paint and coatings industry towards greener formulating methods, such as shifting from solvent-based to water-based coatings as a method of reducing VOCs.
Throughout the late 1960s and 1970s, there was an increased concern regarding air pollution and the detrimental effects to both human and environmental health.
From this pollution arose the need to define and regulate the effect of paints and coatings on the local environment by limiting the amount of certain additives in paint which are damaging to the environment.
The EPA identified volatile organic compounds as “any compound of carbon, excluding carbon monoxide, carbon dioxide, carbonic acid, metallic carbides or carbonates, and ammonium carbonate, which participates in atmospheric photochemical reactions, except those designated by the EPA as having negligible photochemical reactivity.”1
The EPA calculates compliance with VOC content regulations (Title 40, Chapter I, Subchapter C, Part 59, Subpart D—National Volatile Organic Compound Emission Standards for Architectural Coatings S: 63 FR 48877, Sept. 11, 1998, §59.406) according to Equation 1:
In Equation 1, VOC content = grams of VOC per liter of coating; Wv = mass of total volatiles, in grams; Ww = mass of water, in grams; Wec = mass of exempt compounds, in grams; V = volume of coating, in liters; Vw = volume of water, in liters; and Vec = volume of exempt compounds, in liters.
In 1984, the EPA introduced Method 24 to quantify the amount of VOCs in coatings and inks sold in the United States. Method 24 is an indirect method of VOC determination, wherein the water content, solids content, and density of the coating are directly measured and used to back-calculate the amount of VOCs by a mass difference approach.
Method 24 is insufficient for waterborne coatings with low VOC content, as the indirect method erroneously determines small mass fractions of VOCs as compared to the much larger water weight percent, with exponentially increasing error below VOC content of approximately 250 g/L.
As coatings shifted from solvent-based formulations to more environmentally friendly water-based formulations, the insufficiencies in this method motivated the need for new standardized regulatory methods and measurement procedures. Despite the need for improved methods, EPA Method 24 is currently the regulatory method federally mandated across the United States.
States and regions throughout the United States have various guidelines that extend beyond federal rules. California, particularly the Los Angeles air basin, has faced, and continues to face, high prevalence of air pollution known as “smog,” a portmanteau coined in the 1900s to describe the uniquely industrial mixture of smoke and fog becoming increasingly prevalent in large urban areas.
Regulatory agencies such as the California Air Resource Board (CARB), and more specifically SCAQMD, formed the most stringent regulations in the United States to reduce the local effects of this increasing pollution. Method 313 is a direct method for the measurement and quantitation of VOCs using a gas chromatograph with flame ionization detector (GC-FID) applied to samples with less than 150g/L of VOCs.
The complexity of this method is the main deterrent to its use. VOCs are quantified via multilevel calibration curves generated for each analyte used in the coating formulation.2 Relative response factors allow for the calculation of volatiles through this direct method. The regulation of VOCs is relative to the retention time of methyl palmitate. Compounds that elute prior to methyl palmitate are not included in the calculation of volatiles per liter coating. The complexity and laborious sample preparation associated with this method render its use undesirable and drove the innovation of a new standard: ASTM D6886.
ASTM D6886 is a non-regulatory analytical method suitable for the analysis of coatings with less than 150g/L of VOCs, which resulted from an in-depth study by California Polytechnic State University for the California Air Resource Board.3 This method does not define a VOC as Method 313 does, rather it identifies and quantifies all volatiles within a formulation. Although it is not regulatory in nature, it has been widely adopted by SCAQMD as it provides for a less labor-intensive direct measurement of VOC content in coatings as compared to Method 313.
Like Method 313, GC-FID is used in ADTM D6886 to quantify the volatile compounds present in the material. This method utilizes an internal standard, ethylene glycol diethyl ether (EGDE), for the calculation of response factors for an analyte of interest, as discussed in subsequent sections of this manuscript. Herein all response factors discussed were collected according to ASTM D6886.
Globally, VOCs are regulated by federal and local governments. Looking beyond the United States, Europe developed ISO 11890, a widely employed direct method for the analysis of samples with expected VOC content between 0.1% and 15% by mass.4
While Method 313 defines a VOC as anything that elutes before methyl palmitate, ISO 11890 defines a VOC as compounds with a boiling point below 250 °C. This is dictated by EU Directive 2004/42/EU.4 ASTM-D6886 and ISO 11890 are very similar in practice, with direct measurements preformed via GC-FID, and primarily differ in the associated VOC determination that follows as dictated by regulatory agencies within relevant regions.
Here, we combine structure-property relationships, neural networks, and gas chromatographic analytical methods to create a digitally enabled workflow that can support the formulator chemist while evolving as quickly as the regulations themselves.
We present a multipronged approach to working with, measuring, and understanding the nature of VOCs in coatings formulations. First, we present a method of improved prediction for quantifying the response factor (RF) of compounds being analyzed by gas chromatography, as a means of augmenting and expediting VOC determination by ASTM D6886 and other chromatographic approaches.
Ongoing work is employing vapor pressure (VP) prediction and measurement to improve the working definition of VOC as it applies to coating production, application, and film-formation processes. Finally, we propose new directions for incorporating these predictive approaches into the formulation development process.