The 2017 SQALAR software tool from EVK enables quantitative values of chemical parameters to be determined from spatially resolved images, acquired in real-time, of a product stream. This offers many possibilities to control or optimise processes. One such example is described in detail next.
Introduction
The quality improvement of Refuse Derived Fuel (RDF) material streams is an important part of improving energy efficiency, meeting environmental standards and following regulations –such as EU directive (2010/75/EU) on incineration and combustion of refuse materials.
Hyperspectral imaging (HSI) offers a solution that can analyse the quality of RDF streams, of varied complex composition, to extract relevant parameters, such as the calorific content, to optimise combustion processes. This is achieved by combining an HSI camera with relevant classifying and sorting technology provided by EVK.
Through many years’ experience in embedded vision systems, EVK has refined and evolved the technology to the point that it can offer all-in-one solutions for process control and optimization. In-house developed cameras, classification engines and sorting engines fit seamlessly together, with software, to deliver high performance – this is further guaranteed through access to EVK’s vast experience in quantitative and qualitative analysis of bulk material streams.
SQALAR Software enables the HELIOS Classification System to be parametrized to do real-time classification decisions. This ability delivers real improvements to process optimizations and it contributes to a good return-on-investment.
Devices and Procedures
The optical sensor for determining quality parameters of RDFs is a HELIOS NIR G2 CLASS infrared hyperspectral imaging camera. Data analysis and modelling are done in SQALAR software from EVK.
Reference measurements of humidity and calorific content are determined form weight loss after drying and from calorimetry, respectively.
Results
The evaluation of humidity measurements in SQALAR is shown in figure 1. Shown to the left of the image in figure1, is a selection of input spatially resolved hyperspectral images, from which the user can select reference points, or areas. The respective spectra of these reference points, or areas, are shown in the middle; these are used to for creating the measurement model shown below them. The model includes groupings of reference spectra, reference measurement values, a correlation line and statistical quality parameters. Shown on the right-hand side, are the output images of the model showing false colour images that represent the quantity of the modelled material stream parameter – which in this case is relative humidity.
The model shown in figure 1 shows a precision (RMSEC) of 1.52% relative humidity and a coefficient of determination of 0.927. With this high precision and accuracy, it is certainly possible to control and steer a material stream.
Shown in figure 2 is a calorific energy quantitative model, which is represent in the same way as described above for figure 1. The output image showing the result of calorific measurements is shown on the right-hand side. The precision (RMSEC) achieved for this model is 1.68kJ/g (which is 3.7 % of the maximum value), with a correlation of 0.903.
The large spread in output values of the model is representative of the inhomogeneity of the material stream. This is transferred to the model through the spread in reference values, obtained from spot sampling of the material stream, that are then used as an input for the model. As can be seen in figure 1 and 2, the spot sampling values are spread out in the range of 0.5 – 20 % and 22.7 – 44.5 kJ/g, respectively. In fact, the scattering within a single reference sample is therefore within the magnitude, of the entire averaged measuring range, of the reference measurements.
Discussion
A fundamental issue is the error-of-measurement associated with obtaining reference values – and this issue ultimately limits output data quality. This is the case in both point sampling measurements and when applied to quality monitoring. If point sampling is not representative of the bulk material stream (this is very hard to establish a-priori in the absence of spatial and time information), then no matter how precise the measurement is, the values obtained will not constitute a useful description of the material stream. It is only though our method that the inconsistency of point sampling can really be observed.
Summary and outlook
The statistical power of real-time measurements of process parameters, over the entire material stream, is far superior to high precision laboratory analysis which, by default, depend entirely on point sampling (which is often not representative). This is especially the case when dealing with inhomogeneous material streams. Only a system with real-time, spatially resolved, monitoring of the entire material stream, can handle a high level of inhomogeneity successfully. This statistical power allows processes to be optimised, that would otherwise not be possible – when using standard laboratory sampling methods.
Forerunners in this technology recognise how it provides self-contained and complete solutions to many diverse and complex applications that the market now demands. The powers of this technology are its adaptability, for instance to many different market sectors; and its ability to handle complexity and deliver precision through statistics.