Enhancing Refractory Recycling: The Role of Automated Sensor-Based Sorting Systems

Automated sensor-based sorting provides multiple benefits over traditional manual methods [3–6]. Manual sorting is constrained by the physical limitations of human operators, who can only handle large particles (i.e., >80 mm) to maintain an economical throughput (Figure 1). This restriction inherently caps the volume of material that can be processed within a given time. In contrast, automated systems equipped with advanced sensors and rapid ejection mechanisms can process smaller particles ranging from 5–80 mm, significantly expanding the accessible volume of source materials.

Furthermore, automated sorting provides detailed specifications of each particle. By utilising technologies such as laser-induced breakdown spectroscopy (LIBS) for elemental chemistry analysis and hyperspectral imaging (HSI) for mineral phase composition, the sorting process ensures a consistently high-quality output with valuable data that can be leveraged in future product developments. This reduces the incidence of downcycling, where materials are reused in lower value applications than originally intended, as well as the environmental implications and costs of landfilling.

The preliminary design of the mobile automated sensor-based sorting equipment, capable of sorting up to 10 tonnes/hour of spent refractories, is shown in Figure 2. The feedstock will be loaded into a hopper by a wheel-loader. Subsequently, the particles will be automatically singularised, characterised by sensors, and sorted, depending on their characteristics, either by pick and place robots or an air ejection system at the end of the conveyor.

The equipment design enables mobility and replication across different locations. As the cost of generating circular materials is substantially affected by transport distances, mobile sorting units will be used to achieve economic viability by minimising logistical expenses.

Automated sorting entails complex processes, necessitating a holistic approach that incorporates feedback loops due to upstream and downstream interdependencies. As achieving optimal outcomes requires collaborative efforts among various experts, the ReSoURCE (Refractory Sorting Using Revolutionizing Classification Equipment) Horizon Europe project was initiated [7]. The framework of this European programme provides an ideal setting to establish well-defined development agreements, significantly accelerate progress, and provides substantial benefits to all partners involved. Within this collaborative environment, understanding across different technology sectors is enhanced, leading to the development of cutting-edge, tailormade solutions. Additionally, the involvement of academic institutions ensures that best practices are applied throughout all processes. Featuring a total budget of ~€8.5 million, of which ~€7.0 million are funded by grants, enables the challenge to be tackled in an effective manner. Table I outlines the partners and their respective roles in the ReSoURCE project, illustrating the diverse contributions and expertise in this integrated initiative.

The integration of three distinct sensors in the equipment aims to redefine the standards for robust sorting (Figure 3). This multi-sensor approach not only increases the effectiveness of the sorting process but also enriches the data quality, paving the way for more sophisticated recycling operations.

Initially, a 3D camera plays a crucial role by pinpointing the exact location of each sample on the conveyor belt while also mapping the complete topology of the particle. This is vital for scanning fractured particles with irregular surfaces. The 3D camera helps identify the geometrically ideal regions on these uneven surfaces for conducting LIBS measurements. Optimally, the laser should hit the sample perpendicular to the test surface to ensure high accuracy and low noise levels. In addition, analysis of surface features on the fracture surfaces by the 3D camera contributes to the sorting process. For example, it can facilitate preliminary classification—either confirming the particle’s suitability for recycling or excluding it. In some cases, it might even refine existing sorting classes, enhancing the overall efficiency and precision of the sorting system.

Secondly, two HSI cameras play a critical role by capturing the spectral signature of each particle in the range of 400–2500 nm. This technology extends far beyond traditional image analysis, enabling precise visualisation of optical features with a spatial resolution of 384 pixels for the first camera and 1800 for the second. By interpreting reflectance curves obtained from the HSI, it is possible to gather information about the bonding system and identify specific mineral phases. Such information is invaluable for enhancing quality monitoring in refractory recycling, as it provides crucial insights into the material’s composition [8].

Moreover, HSI data is instrumental in assessing the midscale homogeneity of samples—typically spanning a few centimetres. This capability is particularly significant because it helps mitigate a major limitation of LIBS, which can only analyse small spots on the material (i.e., tested volumes usually in the range of a few hundred µm³). In determining the material heterogeneity, HSI aids directing the LIBS to the most relevant regions for targeted analysis. It also plays a crucial role in detecting nonrefractory materials, such as attached process media like slag or clinker, that might be present on the sample.

In summary, the broader imaging scope of HSI offers various advantages, including the ability to classify materials directly into specific sorting categories based on underlying models. By integrating these insights, HSI contributes to a more nuanced and effective sorting process, providing visual surface information as a basis for the subsequent selection of coordinates for the LIBS measurement.

Finally, LIBS measurements will be executed at optimal positions, as determined by the data from the 3D camera and HSI sensors (see Figure 3). LIBS plays a pivotal role by analysing the elemental chemistry of each particle [4,9]. This capability is fundamental for classifying materials based on their chemical composition, which is a primary specification that circular minerals must meet for recycling purposes.

Due to its high sensitivity to nearly all elements in the periodic table, LIBS can be utilised to further refine the sorting process. It enables the establishment of new sorting classes with more precise elemental limits. Following each sorting campaign, LIBS not only categorises materials but can also provide quantitative data regarding the levels of specific elements within each particle (Figure 4). This information is crucial for targeted applications in products and can also provide the basis for blending strategies at later stages. Additionally, LIBS is particularly useful for identifying contamination gradients or harmful substances in used bricks. By targeting the precrushing and liberation of infiltrated zones, automated sorting can isolate material volumes requiring special treatment, thereby reducing them, leading to more economical processing costs.

All data collected from the three sensors will be continuously analysed to identify ways to streamline the sensor setup. Furthermore, using the information gained from this initial setup, the goal is to develop more cost efficient equipment in the future, optimising both technological and economic aspects of the refractory recycling processes.

Currently, sensor and analysis technologies available on the market do not fully meet the specific needs of refractory recycling. For instance, existing LIBS systems are predominantly optimised for sorting metal scrap. This presents a significant challenge for refractory materials, which typically feature a highly heterogeneous microstructure. Unlike metals, where representative volumes are only a few hundred micrometres cubed, coarse ceramic refractories have larger representative volumes, sometimes exceeding 5 mm in terms of the individual raw material components. This size difference means that a random sample may not accurately represent the total chemistry of the refractory product, as indicated in Figure 5.

Furthermore, refractories often possess contamination layers or nonrepresentative layers that can exceed 1 mm due to process media, dust, and oxidation effects. Therefore, efficiently sorting these particles requires a laser capable of quickly ablating any coating material before the measurement pulse analyses the sample. Furthermore, the precision in detecting certain elements is crucial as even minimal levels of contamination can severely impact refractory performance.

To address these challenges, the ReSoURCE project is developing a novel laser source in collaboration with InnoLas (Germany). This advanced technology achieves higher energy levels suitable for testing larger volumes and incorporates tailored pulse modulation to enable both cleaning and analysis pulses within short repetition times. The two new lasers being developed increase the maximum energy level from 150 mJ to up to 430 mJ as well as increase the measurement rate up to 400 Hz, when used at lower energy levels. Additionally, a novel spectrometer design and novel optics developed by LSA GmbH (Germany) will enable higher resolution (in the range of 50–150 pm), particularly for critical elements.

Given the chemical complexity of refractories, AI-based algorithms are being employed to enhance the accuracy and efficiency of data analysis. This cutting-edge approach not only improves sorting accuracy but also ensures that recycled materials meet stringent quality standards.

The current sorting classes used in recycling operations often feature broad specifications. This is partly due to limitations in the precision of existing sorting technologies but can also result from a deliberate choice to accommodate the infrastructure capabilities at recycling plants. As a result, a wide variety of products, and consequently an even greater diversity of raw materials, are grouped into each sorting class. This diversity poses significant challenges in calibrating the sorting machinery effectively.

For the LIBS technology, an initial calibration step involves using a variety of raw materials to pinpoint individual peak positions in the spectrum. So far, just in the ReSoURCE project more than 20 primary raw materials and 100 circular mineral samples have been tested, analysed, and categorised. These identified peaks will be crucial for later stages, where they will help differentiate between various recycling classes. Furthermore, the precision of these measurements was remarkably high, with an average classification accuracy of far more than 90% being achieved for all investigated product types (Table II).

To classify materials based on spectral data gathered with the HSI, models such as pixel class majority are being trained and applied on the samples. To establish a solid ground truth within the project, absorption spectra of over 350 crushed circular mineral samples, raw materials, and possible impurities (e.g., slag and clinker) have been recorded. The first results show a promising average classification accuracy of the HSI data (Table III).

The effectiveness of these spectral features, proven by their high accuracy, has also been applied to the classification of circular minerals. An initial outline of this application demonstrates the potential for enhancing the precision and efficiency of sorting practices in refractory recycling, paving the way for more tailored and effective recycling processes.