Accurate Plastic Blend Analysis Using Mid-Infrared Spectroscopy

A recent study examined a new method to quantify plastic blends using virtually generated spectra. This study, which was published in Talanta, was led by Xiao-Li Chu of the Sinopec Research Institute of Petroleum Processing Co., Ltd. in China (1). Chu and the team demonstrated that by combining pure plastic mid-infrared (MIR) spectra with the Beer–Lambert Law, virtual spectra can be created that can be used for blend analysis without needing experimental calibration data sets, which take an inordinate amount of time to put together.

Close-up side shot of microplastics lay on people hand. Concept of water pollution and global warming. Climate change idea. Microplastics concept in food and water or sea. | Image Credit: © Deemerwha studio - stock.adobe.com

What Environmental Impact Do Plastic Blends Have?

Plastic blends are nonrenewable resources that are routinely discarded in the environment (2). For example, packaging is the main user of plastic blends and makes up approximately 39% of all plastic pollution (2). Because of the harm these plastics have on the environment, scientists are investigating new methods to accurately quantify these plastics to reduce their environmental impact. Traditionally, MIR-based chemometric methods have been used for this purpose, but these methods do not do a good job of acquiring sufficient spectral data to build reliable predictive models (1).

What Did The Researchers Study Entail?

In their study, Chu and the team developed a technique to simulate plastic blend spectra by mathematically combining MIR data from individual plastics according to their mass fractions. By doing this, the researchers were able to effectively generate realistic training data without needing to physically mix and analyze countless blend samples (1).

Four experimental groups were created to test their method. These groups were labeled A, B, C, and D. Group A served as the control, building a quantitative model based exclusively on real spectra from polyethylene (PE), polypropylene (PP), and polystyrene (PS) ternary blends, which are some of the common types of plastic blends routinely found in the environment (1,2). Groups B, C, and D used virtual spectra to construct models, evaluate generalization performance, and assess applicability in mid-infrared hyperspectral imaging (MIR-HSI), respectively (1).

What Were The Key Findings Of The Study?

The experimental groups provided insight into how effectively virtual data could be used to predict plastic composition. For example, Group C contained 208 virtual spectra to train three types of predictive models. These predictive models were partial least squares regression (PLSR), a one-dimensional convolutional neural network (CNN1D), and a two-dimensional convolutional neural network using Gramian Angular Fields (GAF-CNN2D) (1). When tasked with predicting the plastic composition of 66 real ternary blends, the three models in Group C achieved determination coefficients (R² values) of 0.9872, 0.9879, and 0.9944, respectively, which confirms that the modeling strategy using the virtual data was effective (1).

There were also several important findings unveiled from Group D. This experimental group investigated how the models would respond in less-than-ideal analytical conditions. Using a fusion of mid-wave and long-wave infrared bands, the PLSR and GAF-CNN2D models maintained strong performance despite spectral limitations and the addition of Gaussian noise, achieving R² values of 0.9852 and 0.9895 (1).

These outcomes indicated to the research team that their method could be capable of being used for on-site plastic analysis through MIR-HSI technology.

What Is The Overall Impact Of This Study?

The overall impact of this study is that it replaced experimental sample blends with mathematically generated virtual ones, and it showed that it can be effective. The study also indicates that automating applications of MIR-HIS could become more common in detecting and analyzing plastics from recycling plants to polluted ecosystems (1).

In their study, the researchers also provided valuable insight into model design by optimizing feature band selection, spectral interpretation, and evaluation metrics (1). The researchers also used attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy was used to acquire the MIR data, and spectral absorption peaks were correlated with the concentration of individual plastics to guide the modeling process (1).

Why Is Further Validation Needed For This Study?

The researchers’ method needs further validation because they tested their models in a laboratory setting. However, in the real world, non-ideal conditions not only should be expected, but they are common. Plastic waste is often contaminated with oils, dust, or other residues and may exhibit surface roughness or degradation (1). All these factors can distort spectral signals and challenge the linear assumptions of the Beer-Lambert law (1).

As a result, Chu and the team assert at the end of their article that environmental variables, and how they impact model reliability, need to be investigated in more detail so that modelling techniques can be adjusted so that they can handle variability in field-collected samples (1).

References

1. Yang, J.; Xu, Y.-P.; Chu, X.-L. Quantitative Analysis of Plastic Blends Based on Virtual Mid-infrared Spectroscopy Combined with Chemometric Methods. Talanta 2025, 292, 128006. DOI: 10.1016/j.talanta.2025.128006
2. Brizga, J.; Hubacek, K.; Feng, K. The Unintended Side Effects of Bioplastics: Carbon, Land, and Water Footprints. One Earth 2020, 3 (1), 45–53. DOI: 10.1016/j.oneear.2020.06.016