We propose electrical impedance spectroscopy enhanced by machine learning, in particular support vector machines, as a non-invasive, in-situ method for detecting microplastics in water and as an alternative to slow, expensive laboratory measurements. The primary measurand is the complex-valued electrical impedance of a water-filled measuring cell. We carried out stationary measurements on numerous water samples contaminated with different plastic concentrations in a cylindrical measuring cell and in the frequency range from 20 Hz to 2 MHz. The effects of various influence quantities, such as the concentration of organic material (1.0 %, 3.0 %, 5.0 %) or the salinity of water (0.5 %, 1.0 %, 3.5 %), were also investigated. Measurements at 2 MHz with water flows carrying microplastic particles of different materials and sizes served to investigate the dynamic capabilities of the measurement method. The impedance spectra (stationary measurements) or the measured impedances (dynamic measurements) were then evaluated by specific SVMs. The classification task consisted of distinguishing different plastic materials and particle sizes. In the stationary case, the application of the SVM resulted in assignment accuracies of over 98 %. In the dynamic case, the classification accuracies exceeded 91 % for the mere classification of particle sizes and 85 % for the classification of plastic particles by size and material.
Monitoring the nitrate concentration in soil is crucial to guide the use of nitrate-based fertilizers. This study presents the characteristics of an impedance sensor used to estimate the nitrate concentration in soil based on the sensitivity of the soil dielectric constant to ion conductivity and on electrical double layer effects at electrodes. The impedance of synthetic sandy soil samples with nitrate nitrogen concentrations ranging from 0 to 15 mg/L was measured at frequencies between 20 Hz and 5 kHz and noticeable conductance and susceptance effects were observed. Long short-term memory (LSTM), a variant of recurrent artificial neural networks (RNN), was investigated with respect to its suitability to extract nitrate concentrations from the measured impedance spectra and additional physical properties of the soils, such as mass density and water content. Both random forest and LSTM were tested as feature selection methods. Then, numerous LSTMs were trained to estimate the nitrate concentrations in the soils. To increase estimation accuracy, hyperparameters were optimized with Bayesian optimization. The resulting optimal regression model showed coefficients of determination between true and predicted nitrate concentrations as high as 0.95. Thus, it could be demonstrated that the system has the potential to monitor nitrate concentrations in soils in real time and in situ.
24 February 2021 in Journal of Sensors and Sensor Systems
The global economy consumes an estimated 4×1010 t of sand per year, with only 2×1010 t of sand being reproduced by natural sedimentation (Götze and Göbbels, 2017; Peduzzi, 2014). Among other things, sand is also used as a basic material for the production of molds and cores in the foundry industry. The consumption and the economic as well as ecological savings' potential in this area of application can be appreciated by way of an example: the environmental certificate of a single, albeit big German foundry (5160 employees) can be consulted, which states that 39 820 t of sand for casting molds had to be purchased in 2017 (Denes, 2018). In order to avoid having to dispose of the used sand in landfills and to reduce the use of new sand, it is therefore advantageous to renew the used sand in a so-called regeneration process and reuse it as a substitute for new sand in the production of molds and cores. It would be very advantageous if the condition of molding materials (sand–binder systems) in regenerator units could be monitored in real time because of the economic and ecological advantages of a monitored and optimized regeneration process. This work presents the results of investigations in this direction. The objects of investigation in this work are typical molding materials in the foundry industry, e.g., quartz sand, chromite sand, and bentonite as a binder, which are measured impedimetrically with the help of a plate capacitor measuring cell which is connected to an LCR meter (Agilent E4980A). The impedance of the filled capacitor is measured in a frequency range from 1.2 kHz to 1 MHz, containing 123 frequency points. The aim of this research is to work out if the mentioned substances can be measured with the presented measuring method and classified on the basis of impedance characteristics and thus whether impedance spectroscopy can be considered for process monitoring in the molding industry. It is shown that the condition monitoring can possibly be based on impedance spectroscopy because the resulting curves are characteristic of the material used. New and used sands as well as two-component mixtures of sands and binders showed a systematic behavior, which allows the sand or the composition of the mixture to be identified (classified) in the future. The examination of the scatter of the measurement results shows that the impedance data obtained with this method can be measured reproducibly. A descriptive model for multi-component systems is developed in order to be able to interpret the impedance scatter results and their representation in Nyquist plots. From this model, the filling density of the measurement cell and the density of conduction paths can be extracted as essential influence quantities.
The detection of microplastics in water requires a series of processes (sample collection, purification, and preparation) until a sample can be analyzed in the laboratory. To shorten this process chain, we are investigating whether electrical impedance spectroscopy (EIS) enhanced by a classifier based on support vector machine (SVM) can be applied to the problem of microplastics detection. Results with suspensions of polypropylene (PP) and polyolefin (PO) in deionized water proved promising: The relative permittivities extracted from the measured impedances agree with literature data. The subsequent classification of measured impedances by SVM shows that the three classes “no plastic” (below the detection limit of 1 g plastic per filling), “PP” and “PO” can be distinguished securely independent of the background medium water. Mixtures of PO and PP were not examined, i.e. either PO or PP was filled into the measuring cell. An SVM regression performed after the SVM classification yields the microplastic concentration of the respective sample. Further tests with varying salinity and content of organic or biological material in the water confirmed the good results. We conclude that EIS in combination with machine learning (MLEIS) seems to be a promising approach for in situ detection of microplastics and certainly warrants further research activities.
11 October 2022 in Journal of Sensors and Sensor Systems
This work deals with the impedimetric monitoring of used-sand regeneration in the foundry industry. During the regeneration of used sand, a quartz sand similar to new sand is produced from already used molding and core sand, which especially serves to produce new cores. We explore whether the regeneration progress can be assessed in situ based on measured impedance spectra and their features. The impedances of plate capacitors filled with different typical used-sand mixtures, consisting of quartz sand, coal dust, and bentonite, were repeatedly measured in a frequency range from 500 Hz to 1 MHz. The reproducibility of the measurements proved to be sufficient for practical applications. The mean impedances were plotted in Nyquist diagrams. From these plots, systematic impedance–composition correlations could be determined for two of the three component systems. Conclusions about the regeneration state could be drawn from the impedance curves by introducing various features. These were the median, the mean, and the standard deviation of the frequency-dependent resistance and reactance. With these indicators, it was then possible to establish one-to-one relations between the material composition of the molding materials and the impedance measurement. In field measurements on regenerated used sands, this observation was confirmed, and the regeneration progress was observed on the basis of the impedimetrically determined characteristic data.
Molding sand mixtures used in the foundry industry consist of various sands (quartz sands, chromite sands, etc.) and additives such as bentonite. The optimum control of the processes involved in using the mixtures and in their regeneration after the casting requires an efficient in-line monitoring method that is not available today. We are investigating whether such a method can be based on electrical impedance spectroscopy (EIS). To establish a database, we have characterized various sand mixtures by EIS in the frequency range from 0.5 kHz to 1 MHz under laboratory conditions. Attempts at classifying the different molding sand mixtures by support vector machines (SVM) show encouraging results. Already high assignment accuracies (above 90%) could even be improved with suitable feature selection (sequential feature selection). At the same time, the standard uncertainty of the SVM results is low, i.e., data assigned to a class by the presented SVMs have a high probability of being assigned correctly. The application of EIS with subsequent evaluation by machine learning (machine-learning-enhanced EIS, MLEIS) in the field of bulk material monitoring in the foundry industry appears possible.