Life is Plastic? Detecting the Presence of Micro-Plastics in Food and Drink Containers

What we eat and drink has a significant impact on our health. Unfortunately, anything we eat and drink increasingly contains micro-plastics, tiny fragments of plastic material that result from erosion of plastic objects. Indeed, estimates suggest that humans can ingest up to a credit card worth of micro-plastics each week. We contribute a novel wearable system for detecting the presence of micro-plastics in food and drink containers using optical sensing, low-cost micro-controllers, and signal processing techniques that analyze the contents of the containers. We validate our approach through benchmarks using different plastic materials and concentrations, demonstrating that our approach can identify micro-plastics with over 91% accuracy and classify the type of plastic with over 88% accuracy. Our solution offers an innovative yet low-cost pervasive sensing method for improving food safety and detecting containers that are dangerous to use or otherwise faulty.


ABSTRACT
What we eat and drink has a significant impact on our health.
Unfortunately, anything we eat and drink increasingly contains micro-plastics, tiny fragments of plastic material that result from erosion of plastic objects.Indeed, estimates suggest that humans can ingest up to a credit card worth of micro-plastics each week.We contribute a novel wearable system for detecting the presence of micro-plastics in food and drink containers using optical sensing, low-cost micro-controllers, and signal processing techniques that analyze the contents of the containers.We validate our approach through benchmarks using different plastic materials and concentrations, demonstrating that our approach can identify micro-plastics with over 91% accuracy and classify the type of plastic with over 88% accuracy.Our solution offers an innovative yet low-cost pervasive sensing method for improving food safety and detecting containers that are dangerous to use or otherwise faulty.

INTRODUCTION
What we eat and drink has a significant impact on our health [25].
Unfortunately, we increasingly ingest plastics with estimates suggesting that humans may ingest even a credit card (≈ 5 grams) worth of plastics each week [20].This is due to micro-plastics, tiny speckles -or fragments -of plastic which shed from the surfaces of plastic objects and mix up with drinks and foods before being ingested.While current scientific knowledge on the health effects of micro-plastics remains limited, studies on aquatic flora and fauna have shown that they can indeed be highly harmful [18].What makes micro-plastics particularly dangerous is that they are very difficult to observe.According to definition, micro-plastics are fragments that are at most 5 mm in diameter and most of them are much smaller, measured in micrometers [2].Detecting micro-plastics usually requires laboratory analysis [17], with techniques ranging from optical microscope based techniques to chemical analysis and different spectroscopic techniques [14].When eating or drinking something from a plastic container, taking a sample of the contents and taking it to a laboratory for analysis naturally is not feasible.Thus, there would be a demand for solutions that can be used to test for the presence of micro-plastics in everyday consumption.While this might seem excessive at first glance, unfortunately studies are increasingly finding high contents of micro-plastics in everyday drink and food items.For example, bottled water [23] and other bottled drinks have been shown to contain micro-plastics [7,23], and also the contents of beer bottles can be contaminated by micro-plastics [17].Similarly, food containers can contain significant concentrations of micro-plastics, particularly when the food is heated in a microwave or when they contain hot water or other liquids [13].Even if the contents are not served in a plastic container, they likely have been packaged with plastic materials and thus this is not simply a question of the type of container that is used.
The present paper contributes a novel wearable solution for detecting the presence of micro-plastics in food and drink containers.Our solution re-purposes optical turbidity sensors to analyze liquids placed in a container, and uses signal processing and machine learning techniques to estimate whether the container is likely to be contaminated by micro-plastics.The turbidity sensor measures reflectivity of the liquid and we detect the presence of micro-plastics by analyzing changes in the reflectivity profile as the sample of the contents passes by the sensor.While other compounds, such as different medicine, can also affect the reflectivity profile, these typically even out over time whereas micro-plastics are persistent.Thus, changes in the reflectivity profile not just inform whether the contents are affected by micro-plastics, but also provide insights into changes in the overall composition of the contents.Figure 1 shows a conceptual design of our approach with the idea being that the sensor could be used similarly to a necklace, a keychain, or a stirrer.A prototype design of our system is shown in Figure 2. The solution takes advantage of low-cost off-the-shelf components which cost few tens of $, making our solution an affordable and easy-to-use solution for monitoring personal food and drink safety.Through experiments considering different concentrations, plastic types, and liquids, we demonstrate that we can detect the presence of micro-plastics in liquids with over 91% accuracy and can classify the type of plastic with over 88% accuracy.

DETECTING MICRO-PLASTICS
In here, we analyse the feasibility of using optical turbidity sensor to discriminate different types of micro-plastics.The idea is to dip the sensor into a container which has liquid contents, and then determine whether the contents are contaminated by examining the resulting turbidity values.Note that while we focus on containers that are filled with liquids, our solution is applicable to common food and drink containers.Indeed, any food that is heated usually releases liquids and this liquid that binds the microplastics shedding from the container.Thus, the likelihood of the container to release microplastics can be estimated prior to using it for serving food.As part of our experiments we also demonstrate that we can use the turbidity values to detect the type of plastics that the container has.We next briefly explain the theoretical background for our solution, after which we present our prototype design.
Theoretical Background.Turbidity characterises the relative clarity of a liquid and is measured by examining the amount of light that is reflected by the liquid.Turbidity is reported in nephelometric turbidity units or NTUs which correspond to the number of suspended particles in the liquid scattering and reflecting light.Thus, the lower the turbidity, the clearer the water looks like whereas water that has high turbidity looks more cloudy.Note that turbidity itself is not dangerous, but that it depends on the nature of the compounds that increase turbidity.For example, rivers rarely look clear yet are not necessary dangerous as the higher turbidity often results from organic materials shedding from the river bed to the water.Drinking water from the tap or a bottle is usually expected to have low turbidity, with recommended turbidity levels being below 5 NTU and in most developed countries we would expect to see water having a turbidity below 1 NTU [21].Thus, if we measure water placed on a drink or food container, we are expected to see a low turbidity unless the water contains foreign compounds.In our case, we focus specifically on plastics by assuming the user is responsible for pouring the liquid into the container and measuring its contents.Thus, deliberate modifications of the contents, such as effervescent or drugs, are outside of the scope of our work, even if similar principles can be adopted for monitoring them [26].
Plastic Types.The most common way to classify plastics is to use their resin identification code or RIC which refers to the type of resin that is used in the manufacturing process.The most commonly occurring plastics belong to six RIC categories (RIC 1-6 with RIC = 7 referring to other plastics).Of these categories, we focus on five RIC categories (1-2 and 4-6) corresponding to the most common plastics in consumer products and the most common sources of microplastics: polyethylene terephthalate -PET (transparent), highdensity polyethylene -HDPE, low-density polyethylene -LDPE (translucent), polypropylene -PP, polystyrene -PS.The one missing category is PVC (RIC = 3) which is common as a building material but less common as consumer product.Thus, it is the least likely source of microplastics.The RIC affects besides the manufacturing process, also the characteristics of the resulting plastic.Of these plastics, PET (RIC = 1) has higher density than water which means that it generally should sink.Though, the most common source of PET plastics are plastic bottles which may trap air inside the plastic molecules, resulting in some of the PET particles remaining afloat.PS (RIC = 6) has a density of around 1 g/cm 3 which means that it mostly remains suspended in liquid, whereas the other materials LDPE, HDPE, PP all float [22].Note that the density of a plastic is not fully constant as it depends on the molecular integrity of the material which changes over time as a result of degradation of the material.Liquids also differ in density and thus the behavior of the plastics changes if a different liquid than water is considered [15].Prototype Design.We built a custom prototype that uses a Red board DEV-13975 micro-controller as the control unit and the  DFRobot SEN0189 gravity turbidity sensor for estimating the turbidity level of liquids.The micro-controller costs approximately $25 whereas the turbidity sensor costs around $20. The turbidity sensor is split into a probe part that consists of a LED and a photoreceptor that measures the amount of light decay from the light source (see Figure 1), and a control part that integrates the other component.The light measurement is then converted into a voltage which can be used to estimate the overall turbidity (in NTUs) by using a conversion formula.Optimally, the sensor would be calibrated for different types of liquids.We simply use the default equation given by the sensor data sheet which converts voltage to turbidity values: Here  is the turbidity level and  is the voltage value.As discussed, a key challenge for detecting plastics is that they have different densities which affects where exactly they reside within the liquid.We mitigate this challenge by stirring the liquid and integrating an IMU sensor to the top of the prototype to detect when the contents are stirred.This allows to compare turbidity levels in the liquid as the potential particles inside it are stationary and when they are in motion.As the IMU sensor we use Crowtail CRT35038I which costs approximately $30 though also cheaper options can be used.
Figure 3b shows the prototype in action.

EXPERIMENT
We validate our prototype's capability to detect microplastics through a controlled experiment where we vary the type and concentration of plastics.
Microplastics.We consider five most common plastic materials occurring as microplastics and create microplastics by using a scrapper to shed particles from consumer items of the corresponding material.The plastics products used in our experiments are commonly found in households and environment: water bottles (PET), the cap of water bottles (HDPE), the cover of a food containers (LDPE), food containers (PP), plastic cups (PS); see Figure 3a.To ensure the plastics are sufficiently small, we further filter them using a 1-mm mesh sieve, thus ensuring that all samples have a diameter of less than 5 mm -the maximum size for being classified as microplastics [2].Micro-plastic particles having diameter smaller than 1 mm can be challenging to observe by the naked eye -especially when the micro-plastics are transparent -unless they are present in sufficiently high quantities.
Experiment Procedure.For each plastic, we followed the same procedure.First, we filled a glass jar with 100 ml of water and then used the prototype to measure the turbidity of the water.We chose 100 ml as this was sufficient for submerging the probe part of the sensor.Next, we added 1 g of plastics into the container.The mixture was then stirred for 30 seconds after which we continued to take measurements for one more minute.Through early tests we found that the sensor has some delay in reacting to changes in the liquid composition and one minute duration was found sufficient to react to these changes.Note that the probe part of the turbidity sensor need to be submerged fully to ensure maximum capture surrounding particles; see Figure 3b.The beginning and end of the stirring is detected by examining the IMU measurements of the prototype.We then progressively increased the concentration of water -thus diluting the concentration of plastics -by adding 50 ml of liquid at a time and repeating the same measurement process for each concentration.The final concentration that was used for measurements was 300 ml.To avoid any plastic contamination, the jar and the sensor probes were carefully cleaned before use.

RESULTS
Presence of Micro-Plastics.We first assess the potential of using the turbidity sensors to detect the presence of micro-plastics.We focus on plastic particles suspended in water and later demonstrate that the results are not limited to analysing water.Figure 4 shows the results for different plastics types and concentrations.When the concentration of plastics is high, all plastics show significantly higher turbidity values than the water without any plastics.At lower concentrations, the turbidity values start to contain more variation and some plastics become more difficult to recognize.This is due to different behaviors between the plastics.Indeed, some plastics float, others are suspended in the liquid, and some sink to the bottom.As we controlled concentration by adding more liquid, lower concentrations for plastics that sink or are suspended may fail to come in contact with the sensor.Thus, the results contain more uncertainty.In terms of plastic materials, there are also observed differences.Both polyethylene types (HDPE and LDPE) and polypropylene (PP) are easier to detect than polyethylene terephthalate (PET) or polystyrene (PS).This is a direct result from differences in the materials as HDPE, LDPE and PP have lower density than PET and PS and thus they are more likely to remain floating at the top of the container.We confirmed the differences in turbidity across all micro-plastics are significant (Kruskal-Wallis tests: p< .001).Post-hoc comparison (Dunn-Bonferroni) confirms these differences were statistically significant for majority micro-plastic pairs, except LDPE-PET.Differences between water and water mixed with microplastics are significant (Kruskal-Wallis tests: p< .001),and these differences were confirmed for all water-micro-plastic pairs using post-hoc comparison (Dunn-Bonferroni).
Detecting Plastic Type.We next demonstrate that our approach can not just detect the presence of micro-plastics but can also provide insights into the type of plastics that are present.We show this through classification experiments where we train three basic classifiers to detect plastic types.The classifiers we consider are k-Nearest Neighbors (kNN), Random Forest (RF), and AdaBoost.As our primary feature, we consider the turbidity value of the liquid.We separately also train classifiers that consider both turbidity and estimated concentration level as features to assess whether controlling the amount of liquid would impact classification performance.When only turbidity is considered, we can detect microplastic types with an overall accuracy of 88.8%.For most microplastics (HDPE, LDPE, PET, and PS), the accuracy is at least 91.57% with the sole exception being PP, which is the lightest plastic in terms of density, meaning that it mostly accumulates at the surface.When combining the turbidity feature with the concentration feature, the average accuracy for detecting the different microplastic types increases to 97.6%.Most notably, the accuracy of detecting the presence of  1. PP has the lowest density of all plastics and thus it mostly floats around the surface.As the sensor needs to be partially submerged into the liquid, the sensor only samples a small surface area and the concentration of the liquid has a significant impact on how many particles can be captured.Other plastics are heavier and thus they are either suspended in the liquid or sink, which means that the coverage of the sampling area does not influence the result.
Different Liquids.We next demonstrate that our system is not limited to water, but can also operate with other liquids (tea, coffee, milk) that are common drinks and different in color and turbidity characteristics.As the turbidity sensor is based on optical sensing, its performance depends on the capability of light to penetrate the liquid.To ensure best possible performance, we placed an additional light source (a flashlight) under the container when taking the measurements.Figure 5 shows turbidity values for different liquids.
We can observe clear differences in the turbidity levels of different drinks.We separately verified that these differences were also preserved when measurements were taken in ambient light (Figure 5b), and the flashlight merely helps to reduce noise and outliers in the measurements.The results are in line with expectations with water having the lowest turbidity, followed by tea, coffee, and milk.We used a Kruskal-Wallis test to verify that differences in turbidity across the drinks are statistically significant  < .001).Post-hoc comparison (Dunn-Bonferroni) confirms these differences were statistically significant for all drink pairs.No statistically significant differences were found when the measurements were taken with the external light included or when only ambient light was available.

DISCUSSION
Prevention.We have demonstrated that low-cost sensor technology and simple algorithms can be used to improve personal food and drink safety by detecting the presence of harmful micro-plastics in containers that are used for serving food and drinks.While our solution can be used as a passive solution to monitor plastics, it could also be integrated with cleaning or warning solutions.In the simplest case, cleaning could mean filtering the contents and using a different container, whereas a more advanced solution is to use so-called electrocoagulation where an electric current is used to coagulate plastic materials.This makes the plastics larger and easier to gather.Our solution could also be integrated into taps or customized bottle caps that are used to periodically test the contents of a container.If micro-plastics are detected, a warning can then be issued to the user.
Micro-Plastic Sizes.In our work, we focused on micro-plastics that have a diameter less than 5 mm.In nature, plastics never decompose and instead they merely break into smaller and smaller pieces until they are ingested by some organism or otherwise removed from circulation.This means that even micro-plastics break into smaller and smaller pieces with the smallest known micro-plastics being less than 1 m in diameter.Low-cost turbidity sensors may not be sufficient for detecting the smallest fragments as this would require very precise lasers and high sensitivity photo-receptors before optical methods can identify them.Thus, further work is needed to improve the scale of particles that can be detected.
On Wearable Integration.While the current prototype is practical as a standalone passive sensor or as a separate wearable that is carried around similarly to a key-chain, we foresee that there would be other types of modalities for integrating our approach.
The system could take advantage of wireless power transfer and connectivity to link with a smartwatch or a related wearable.Further research is needed to investigate better and more user friendly designs that offer a seamless user experience.
Other Use Cases.Given the cheap cost of the components, there are also many other possibilities to take advantage of a similar solution as ours.For example, heath studies have shown that turbidity levels of tap water correlate with occurrence rates of gastrointestinal illnesses [19].Our solution can also be useful for coffee or tea hobby baristas.Indeed, coffee turbidity depends on the growth terroir and thus our solution could be used to assess and evaluate the quality and authenticity of brewed coffee [6].Similarly, turbidity sensors can be used to assess the grade and amount of nutrients in tea [8].Beyond coffee and tea, turbidity can even be used to assess the fermentation process of beers [12].Our solution goes beyond food and drink safety, and can be expanded to health studies to quality control support for at-home coffee and tea baristas or homebrewers.

RELATED WORK
Detecting Plastics.The most common approach for detecting plastics, including micro-plastics, is to take advantage of spectral diffusion [3].Laboratory instruments take advantage of spectroscropy where the absorption and/or reflectance patterns of materials are analyzed and used to establish a fingerprint of the material, which then can be used to detect the precise composition of the materials.
There have been some work on adopting these methods to the field, but the general problem is that the necessary components are costly and easily become bulky as the best performance is obtained when a broad range of light frequencies are investigated.In specific contexts, these methods can be adopted in the field, e.g., drifting floaters can use optical instruments to detect floating plastics [4], and these techniques can also be integrated into aerial drones [1].
Food and Drink Safety.Optical and pervasive sensing has been suggested to prove the detection of spiking [26], by calculating differences in light reflectivity resulting from small particles inside the drink.Mobile phones were used to detect maturity estimation of meat [16].Near-Infrared Spectroscopy has been applied to analyze food components [5] and sugar level in drinks [11].While they are claimed to be portable and ready for in-situ analysis, being small in size, robust, simple to use and analyze, they remain highly costly [24] and mostly used for in-vitro studies [10].Should the claim is correct that one credit card of plastics is consumed by humans each week [9], there is a critical point to find rapid and cost-effective solutions for detecting the presence of micro-plastics in food and drinks.

CONCLUDING REMARKS
We developed a simple yet efficient solution for detecting the presence of micro-plastics in food or drink containers by repurposing off-the-shelf sensors (turbidity and IMU) into a prototype that can used to test containers by filling them with liquids and measuring the turbidity level of the liquid.Through experiments, we demonstrated that our approach can detect the presence of micro-plastics with over 91% accuracy and also classify the type of plastic with over 88% accuracy.Our approach offers an easy-to-use solution that ordinary people can use to analyze whether re-using the container is a safe choice for their health and avoid unnecessary ingestion of micro-plastics -beyond to what we are already exposed to in our everyday life.Besides facilitating food and drink safety, our solution also can be used to detect which containers are becoming degraded to a degree where they should be destroyed, rather than simply cleaned and re-used.Indeed, once the plastic structure starts to degrade, the process tends to continue and thus even if cleaning can help in the short-term, the micro-plastics are prone to returning as more particles fracture from the surface of the container.

Figure 5 :
Figure 5: Turbidity of different liquids with external light (a) and in ambient light (b).