Ultrasound imaging for quantitative evaluation of magnetic density separation

Abstract

This thesis is dedicated to an investigation of the potential and technological possibilities of an inline ultrasound system as a quality control system for wet recycling of solid waste. The main targeted recycling technology is magnetic density separation (MDS), a novel technique that was investigated and technologically matured in a project running in parallel to this work. In MDS, the easily magnetisable ferrofluid is used as the separation medium to sort different materials based on their mass densities. The MDS is very accurate compared to conventional recycling separation techniques as it is effective even when the densities are very close (< 1 weight percent), such as for different polyolefin plastics. The special attention for plastics in this work is motivated by the economical and environmental gains that are obtained from the separation of plastics from large waste flows such as automobile scrap and household waste. Due to the inherent optical opaqueness of ferrofluid, the ultrasound imaging system is the only effective method that allows accurate observation and study of the waste particles as they separate in the channel. Moreover, the intrinsic properties of ultrasound make it suitable for quantitative analyses, such as particle tracking and measurement of particle sizes, volume and density distributions as the particles flow in large quantities towards the extraction units. The main objectives for this work have been achieved. It was shown that commercial medical 2D ultrasonic imaging systems provide a good technological point of departure for the desired inline system. They are capable of generating good quality images of moving particles, provided the view of the probe onto the particles is well controlled. Moreover, it has been shown that real-time ultrasound is capable of delivering online quantitative information that is crucial to the performance of an MDS. In particular, image processing techniques have been applied on the real-time ultrasound video-streams to evaluate the particles density distribution in the channel, measure the particle velocity and to analyze their motion behaviour as they float in the ferrofluid. The limitations of the medical commercial technology are that it cannot serve as a reliable stand-alone machine and also cannot provide all types of quantitative analyses that are desirable in an industrial recycling operation. As a next step, the investigation turned towards the imaging methods themselves. For that purpose the general linear acoustic theory for waves in ferrrofluids and acoustic imaging was reviewed, first to establish the basic physics and main principles of imaging. The potential of quantitative ultrasound analysis was determined by focusing on cross-section imaging, which is the biggest challenge for accurate 2D imaging. It has been shown that probe positioning and the overall data acquisition strategy deserve due consideration, since data quality proved paramount for good quality images. For the imaging research, the most technologically promising ultrasound methods were adapted from the fields of seismology, medical ultrasound and non-destructive testing. These imaging methods were developed in either the space-time domain or the Fourier domain, as each approach proved to have its own advantages and limitations for data requirements and computational costs. The methods were implemented in Matlab and supplied with raw ultrasound data, scanned from static scenes with just a few generic test objects. These objects were generic in the sense that all possible shapes and size-dependent acoustic wave effects were captured that could be expected with ‘real’ waste particles. Two complementary data sets were used to investigate the possible benefits of having either a wider sensor array aperture during transmission (pulse-echo data) or during data reception (plane wave data). The resulting images were evaluated in terms of performance indicators, which were introduced to obtain a more objective judgement of image quality. This research showed that accurate ultrasound cross-section imaging is quite feasible if good quality data can be scanned, i.e. if the data contain the necessary acoustic information. In particular, the availability of acoustic information from both front and back surfaces was found to be the key factor for good quality data. It is also concluded that all the imaging methods tested in this work are in principle capable of delivering good image quality, provided the data are of sufficient quality. What sets them apart are the substantial differences in computational costs and the ability to process different types of data. Finally, the research conducted in this thesis has also led to the compilation of a set of recommendations for future realization of an ultrasound system, dedicated to inline quality control in recycling.