FIELD

[0001] The present disclosure is related to the analysis of inhomogeneous feed material to provide compositional information for process control and optimization. In particular, the present disclosure is related to systems and methods of analyzing feed in a smelting process.

BACKGROUND

[0002] In industry, feed materials are often solid, granular, and are homogenously mixed but heterogeneous in composition. In order to determine the bulk elemental or mineralogical composition of the feedstock, a sample is currently homogenized by solvation or digestion and the resulting product is analyzed. Alternatively, the solid sample can be analyzed on a grain-by-grain basis. Current methods cannot therefore determine the average bulk composition of a solid granular feed material in real-time.

[0003] Examples of such heterogenous feeds are electronic wastes, metal and non-metallic wastes, contaminated soils, and concentrates in mineral processing.

[0004] One example of a heterogeneous feed in mineral processing is a copper sulfide concentrate which is smelted around the world. Smelting is a popular pyrometallurgical process used to extract base metals, such as copper, nickel, zinc, etc. from finely pulverized mineral concentrates. For example, copper flash furnace smelting produces copper matte (sulfide) with an estimated copper enrichment of up to 75%.

[0005] The conversion of concentrated powders, which may contain chalcopyrite (CuFeS), bornite (CusFeS^, covellite (CuS), chalcocite (CU2S), and pyrite (FeSz), as well as non-sulphur containing minerals, can occur when the sulphur-bearing minerals ignite after entering a flash furnace and mix with oxygen- enriched air and flux, such as silica (SiC ). Copper concentrates can also be smelted directly in different types of furnaces, but their feeding is similar. The main expected reactions occur in the reaction shaft for chalcopyrite, bornite, and pyrite, and produce a mixture of sulfides and oxides. After falling to the bottom of the furnace, two distinct layers form. They are called the matte and slag, and consist of metal sulfides and oxides, respectively, with the goal of reducing iron in the matte and reducing the overall sulphur burden.

[0006] Conventionally, process operators can adjust system parameters in order to optimize the conditions within the furnace according to the bulk composition of the mineral ore concentrate being fed into the processing system, typically by adjusting the oxygen to feed ratio based on the (Cu + Fe)/S ratio.

[0007] To determine the composition of the mineral ore powder in a lab, spectroscopy is often used. Two common and useful flame-based analytical techniques to determine the elemental composition of solid and liquid samples are flame emission (FES) and atomic absorption (AAS) spectroscopy. Both techniques analyze a sample by first nebulizing a solution into a very hot flame by entraining droplets in the fuel/oxidant gas flow. The flame desolvates and atomizes the fine droplets. This thermally excites some of the atoms. The number of excited atoms depends on the flame temperature, the transition probability, and the concentration of the analyte. FES instruments then measure the resulting emission intensity from the relaxation of these excited species at one or many wavelengths depending on the optical layout. After calibrating the instrument at a fixed temperature, the emission intensity is then related to the analyte's concentration.

[0008] The emission spectra produced from the atomic emission spectroscopy are usually analyzed one at a time to identify the composition of the sample. The composition of the sample is typically obtained from a multi-parameter fit using known molecular and atomic constants and sophisticated fitting software, such as pGopher (A Program for Simulating Rotational, Vibrational and Electronic Spectra, C. M. Western, Journal of Quantitative Spectroscopy and Radiative Transfer, 186 221-242 (2017) doi: 10.1016/1.iasrt.2016.04.010, incorporated herein by reference). Once the composition of the mineral ore powder is determined, the process conditions may be adjusted in order to optimize the processing of the feed, for example, to maximize the quality or quantity of the matte output.

[0009] Smelting processes often run continuously and can, in principle, be fully automated by controlling process conditions. However, given the heterogeneity of the feedstock and blending of concentrates, the composition of the feedstock can vary as it is fed into the system. As well, the mineralogy of the feed can impact the combustion of the sulfides and this is not taken into account in current process optimization.

SUMMARY OF THE INVENTION

[0010] In one aspect, the specification relates to adjustment system for optimizing processing of feedstock by a processing system, the adjustment system comprising: a processor, a network interface, and a memory storing computer executable instructions, wherein when executed in the processor, the computer executable instructions cause the processor to: determine a chemical composition of the feedstock by: obtaining an emission spectrum, the emission spectrum having been generated based on captured light emissions from a sample of the feedstock following atomic excitation of the sample; and computing, using a trained neural network, a chemical composition of the feedstock from the emission spectrum; identify optimal parameters for processing the feedstock based on the computed chemical composition of the sample of the feedstock; and instruct the processing system to modify operating parameters of the processing of the feedstock to the identified optimal parameters.

[0011] In a second aspect, the specification relates to a method of optimizing processing of feedstock by a processing system, the method comprising: determining a chemical composition of the feedstock by: obtaining an emission spectrum, the emission spectrum having been generated based on captured light emissions from a sample of the feedstock following atomic excitation of the sample; and computing, using a trained neural network, a chemical composition of the feedstock from the emission spectrum; identifying optimal parameters for processing the feedstock based on the computed chemical composition of the feedstock; and instructing the processing system to modify operating parameters of the processing of the feedstock to the identified optimal parameters.

[0012] In a third aspect, the specification relates to a computer-readable medium storing instructions that, when executed by a processor of a system, causes the system to: determine a chemical composition of feedstock by: obtaining an emission spectrum, the emission spectrum having been generated based on captured light emissions from a sample of the feedstock following atomic excitation of the sample; and computing, using a trained neural network, a chemical composition of the feedstock from the emission spectrum; identify optimal parameters for processing the feedstock based on the computed chemical composition of the feedstock; and instruct a processing system to modify operating parameters of the processing of the feedstock to the identified optimal parameters.

[0013] In a forth aspect, the specification relates to a system for optimizing processing of feedstock, the system comprising: a processing system configured to process the feedstock under operating parameters; and an adjustment system coupled to the processing system, the adjustment system comprising: a processor, a network interface, and a memory storing computer executable instructions, wherein when executed in the processor, the computer executable instructions cause the processor to: determine a chemical composition of the feedstock by: obtaining an emission spectrum, the emission spectrum having been generated based on captured light emissions from a sample of the feedstock following atomic excitation of the sample; and computing, using a trained neural network, a chemical composition of the feedstock from the emission spectrum; identify optimal parameters for processing the feedstock based on the computed chemical composition of the sample; and instruct the processing system to modify the operating parameters of the processing system to the identified optimal parameters.

[0014] In a fifth aspect, the specification relates to a burner assembly for use with a spectroscopic device for determining the chemical, elemental and/or mineralogical composition of a sample of feedstock, the burner assembly comprising: a burner coupled to a source of an oxidant and a source of a fuel, the burner configured to generate a flame from the oxidant and the fuel to atomically excite the sample and produce light emissions for generating emission spectrum; an injection device with a sample inlet and a sample outlet, the sample inlet positioned above the sample outlet and configured to receive the sample of feedstock for delivery into the flame generated by the burner.

BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Reference will now be made, by way of example, to the accompanying drawings which show example embodiments of the present application, and in which:

[0016] Figure 1 is a schematic illustration of an example adjustment system for optimizing processing of feedstock in line with a processing system in accordance with the specification;

[0017] Figure 2 is a schematic illustration of a first example of the adjustment system of Figure 1 in isolation;

[0018] Figure 3 is a schematic illustration of a processing device of the adjustment system of Figure 2 in use with input and output devices;

[0019] Figure 4 is an image of the first example adjustment system of Figure 2 in use when copper powder is introduced into a flame;

[0020] Figure 5 is a normalized emission spectrum collected 3 cm from the flame of Figure 4; [0021] Figure 6 illustrates overlaid copper powder emission spectra, collected 1 cm from the flame of Figure 4;

[0022] Figure 7 illustrates a portion of the copper powder emission spectra of Figure 6 overlaid with theoretical transitions for CuO;

[0023] Figure 8 is an image of the example adjustment system of Figure 2 in use when iron powder is introduced into the flame;

[0024] Figure 9 is a normalized emission spectrum collected 3 cm from the flame of Figure 8;

[0025] Figure 10 illustrates overlaid iron powder emission spectra with expected emission lines from FeO and elemental Fe;

[0026] Figure 11 is a graphical representation of the composition of iron/copper sulfide mineral samples as determined via ICP-OES and ELTRA (Sulfur);

[0027] Figure 12 illustrates results of the training of an artificial neural network (ANN) according to the elemental composition of sulfide reference minerals with R.MSE for testing data;

[0028] Figure 13 illustrate results of ANN training based on MLA mineralogy with R.MSE displayed for testing data;

[0029] Figure 14 illustrates overlaid flame emission spectra (normalized) of 12 concentrate samples;

[0030] Figure 15 illustrates results of ANN training on samples analyzed with the adjustment system of Figure 2 with R.MSE displayed for testing data;

[0031] Figure 16 shows a schematic illustration of a second example adjustment system with a burner assembly in accordance with the specification;

[0032] Figure 17 are images of the second example adjustment system of Figure 16 operating under non-ideal (left) and ideal (right) conditions; [0033] Figure 18 are images of the second example adjustment system of Figure 16 before powder injection (top) and during passive powder injection (bottom);

[0034] Figure 19 illustrates normalized emission spectra results from testing of the adjustment system of Figure 16 with oxygen/acetylene flame using the vibration feeder;

[0035] Figure 20 illustrates output from ANN predictive model for elemental composition;

[0036] Figure 21 are images of a neutral nitrous oxide-acetylene flame using the second example adjustment system of Figure 16 without oxygen injection (top), with powder injection without oxygen (bottom), and the resulting emission spectra (inset);

[0037] Figure 22 are images of a neutral nitrous oxide-acetylene flame using the second example adjustment system of Figure 16 with oxygen injection (top), with powder injection aided with oxygen (bottom), and the resulting emission spectra (inset);

[0038] Figure 23 are images of a rich nitrous oxide-acetylene flame using the second example adjustment system of Figure 16 with oxygen injection (top), with powder injection aided with oxygen (bottom), and the resulting spectra (inset);

[0039] Figure 24 is a flow diagram illustrating a method for optimizing processing of feed material by a processing system;

[0040] Figure 25 illustrates an example graphical user interface that may be used to control the mass flow controller in the adjustment system of Figure 2 or Figure 16;

[0041] Figure 26 illustrates an example graphical user interface that may be used to control the spectrometer component in the adjustment system of Figure 2 or Figure 16;

[0042] Figure 27 is a top perspective view of an alternate burner assembly. [0043] Figure 28 is a side view of the alternate burner assembly of Figure 27;

[0044] Figure 29 is a cross-sectional view of the alternate burner assembly along A-A in Figure 28;

[0045] Figure 30 is an enlarged view of portion B of the alternate burner assembly of Figure 29;

[0046] Figure 31 is an enlarged view of portion C of a part of Figure 30;

[0047] Figure 32 is an enlarged view of portion D of a part of Figure 31;

[0048] Figure 33 is an image of the alternate burner assembly of Figure 30 operating under ideal conditions without active injection or sheath gas;

[0049] Figure 34 illustrates results of ANN training on elemental composition samples analyzed with the alternate burner assembly of Figure 27;

[0050] Figure 35 illustrates results of ANN training on mineralogical composition samples analyzed with the alternate burner assembly of Figure 7,'

[0051] Figure 36 illustrates normalized emission spectra results from testing of the alternate burner assembly of Figure 27 with reference minerals; and

[0052] Figure 37 illustrates normalized emission spectra results from testing of the alternate burner assembly of Figure 27 with mixtures of concentrated copper ores.

[0053] Similar reference numerals may have been used in different figures to denote similar components.

DESCRIPTION OF EXAMPLE EMBODIMENTS

[0054] The mineral ore feedstock composition is often continuously changing during processing/smelting operations, but the specific changes to the feed composition cannot be directly measured in-situ. Even when using sophisticated fitting software, such as pGopher, hundreds or thousands of emission spectra cannot be analyzed in real-time. Thus, process feedback is delayed and any adjustments made may no longer be optimal for the mineral feedstock that is subsequently being processed. In order to better optimize process conditions, realtime compositional analysis and feedforward automation is required. This is especially true given that nearly 65% of all copper smelting processes utilize flash/continuous methods.

[0055] To overcome the above-noted challenges, the specification relates to an adjustment system 10 and a method 500 for optimizing the processing of feed material in real-time by, or in combination with, a processing system 100 using a trained artificial neural network.

[0056] Figure 1 illustrates an example of the adjustment system 10 in combination, in use, and in-line with the processing system 100. In the present implementations, the processing system 100 may be a smelting system. The smelting system may include a known smelting furnace, such as an Outokumu flash smelting furnace, an Inco flash furnace, or a Kivcet flash furnace. In the illustrated embodiment of Figure 1, the processing system 100 includes a flash furnace that receives a continuous stream of feedstock or inhomogeneous solid material, such as mineral ore feedstock. The mineral ore feedstock may comprise chalcopyrite (CuFeS), bornite (CusFeS^, covellite (CuS), chalcocite (CU2S), and/or pyrite (FeSz), from a feedstock source 102. Typically, the flash furnace also receives heated and/or oxygen-enriched air and flux, which mixes and reacts with the mineral ore feedstock. Further oxygen may be added to the flash furnace as needed to adjust the oxygen concentration of the reaction. After falling to the bottom of the furnace, the metal oxides further react with the flux to form a slag layer, and the metal sulfides form a matte layer. The slag and matte are collected separately, the former to be discarded and the latter collected for further treatment. Waste off-gas is also produced as a by-product.

[0057] In the case of the flash smelting process then, the parameters that may be adjusted to optimize the smelting process include the pressure and/or temperature of the pre-heated air that is injected into the flash furnace, the level of oxygen enrichment of the pre-heated air, and the additional oxygen enrichment that may subsequently be added. Other parameters may alternatively or additionally be involved depending on the requirements of the particular smelting process.

[0058] In order to optimize the processing of feed material (such as inhomogeneous solid material) in real-time, the adjustment system 10 may be coupled to, and optionally in parallel with, the processing system 100 and the feedstock source 102 as shown in Figure 1. In that manner, as the mineral ore feedstock travels from the feedstock source 102 to the processing system 100 for processing, a sample of the feedstock/inhomogeneous solid material may be diverted to the adjustment system 10. As will be discussed in further detail below, the adjustment system 10 is configured to analyze the solid feedstock sample in real-time using a trained artificial neural network and then to adjust the processing parameters of the processing system 100 to ideal parameters in response to the analysis in real-time. The adjustment system 10 may analyze the solid feedstock sample with the trained artificial neural network and adjust the processing parameters while (or at the same time as) the processing system 100 is processing the bulk of the feedstock.

First Embodiment

[0059] One example of the adjustment system 10 is shown in isolation (and in greater detail) in Figure 2. The adjustment system 10 of Figure 2 is shown including a processing device 12, a spectrometer component 14, a mass flow controller 16, a burner 18, an injection device 20, and a parameter controller 22. Each component will be discussed in turn.

[0060] Figure 3 illustrates the processing device 12 in further detail. The processing device 12 may include a processor 120, such as a microprocessor, a graphics processing unit (GPU), a tensor processing unit (TPU), an applicationspecific integrated circuit (ASIC), a field-programmable gate array (FPGA), a dedicated logic circuitry, or combinations thereof. The processing device 12 may optionally include one or more input/output (I/O) interfaces 122, to enable interfacing with one or more optional input devices, such as the spectrometer component 14, and/or optional output devices, such as the mass flow controller 16 and the parameter controller 22. The processing device 12 may include one or more network interfaces 124 for wired or wireless communication with other processing systems. The network interface(s) may include wired links (e.g., Ethernet cable) and/or wireless links (e.g., one or more antennas) for intra-network and/or internetwork communications.

[0061] The processing device 12 may also include one or more storage unit(s) 126, which may include a mass storage unit such as a solid-state drive, a hard disk drive, a magnetic disk drive and/or an optical disk drive. In some example embodiments, the storage unit(s) 126 may include data 132 of optimized operating parameters for various feedstock based on the elemental and/or mineral composition of the feedstock and other aspects necessary for optimization of processing the feed material. In the case of smelting operations, the optimized operation parameter data 132 may include oxygen enrichment or then need to add supplemental heating. Other parameters may alternatively or additionally be involved depending on the requirements of the particular processing or smelting process. Each optimal parameter may be different depending on the corresponding composition of the feedstock. The data may be set out in a database or expressed as equations. Each optimal parameter is set based on a predetermined knowledge of the conditions or parameters necessary to optimize the processing of the feedstock, such as for example, to maximize the matte output and to minimize the slag output.

[0062] While the processing device 12 may include the storage unit(s) 126 having optimized operation parameter data 132, in other examples this optimized operation parameter data 132 may be received from one or more remote storage unit(s) that can be accessed remotely via a wireless or a wired network.

[0063] The processing device 12 may include one or more non-transitory memories 128, which may include a volatile or a non-volatile memory (e.g., a flash memory, a random access memory (RAM), and/or a read-only memory (ROM)). The non-transitory memory(ies) 128 may store instructions for execution by the processor 120, such as to carry out example methods of optimizing processing of feed material as in the example embodiments. The memory(ies) 128 may store other software (e.g., instructions for execution by the processor 120), such as an operating system and other applications/functions. In some embodiments, one or more datasets and/or modules may be provided by an external memory (e.g., an external drive in wired or wireless communication with the processing device 12) or may be provided by a transitory or non-transitory computer-readable medium. Examples of non-transitory computer-readable media include a RAM, a ROM, an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, a CD-ROM, or other portable memory storage. In an embodiment, memory(ies) 128 stores an analysis module 130, which may be a software system that includes machine-readable instructions that may be stored locally in the memory(ies) 128, or remotely, and are executable by the processor 120.

[0064] The processor 120 is, thus, configured to execute the computer executable instructions to cause the processor 120 to determine or predict the chemical composition of the feedstock by first obtaining an emission spectrum of the solid feedstock sample. The emission spectrum may have been stored in the memory 128, or received from the spectrometer component 14. In either case, the emission spectrum was generated based on captured light emissions from the solid sample of feedstock following atomic excitation of the sample feedstock.

[0065] The processor 120 is further configured to execute the computer executable instructions to cause the processor 120 to compute or predict the chemical composition of the sample feedstock from the emission spectrum using the analysis module 130. The generated emission spectrum is extremely complex and contains atomic and molecular emission lines, which are proportional to the concentration of constituent species in the sample. It is very difficult to interpret such emission spectrum. Thus, the analysis module 130 includes a trained (artificial) neural network, and the trained (artificial) neural network is capable of determining the composition of numerous feedstock samples in real-time. The training of the neural network will be discussed in greater detail later below.

[0066] After the chemical composition of the emission spectrum is determined, the processor 120 is then configured to execute the computer executable instructions to cause the processor 120 to identify optimal parameters for processing the feedstock based on the computed chemical composition of the feedstock. The processor 120 may access the optimized parameter data 132 in the storage 126 to identify the optimal parameters that correspond with the determined chemical composition. The processor 120 is then configured to execute the computer executable instructions to cause the processor 120 to send instructions to the processing system 100 to modify its current operating parameters (as it is processing the feedstock) to the identified optimal parameters. Since the trained (artificial) neural network in the analysis module 130 can determine the composition of a feedstock sample in real-time, the optimal parameters of the operating system 100 can also be determined in real-time, and the operating parameters of the operating system 100 can then also be automatically adjusted to the optimal parameters in real-time as the bulk of the same solid feedstock is being processed.

[0067] The adjustment system 10 may include the spectrometer component 14, which is coupled to processing device 12. The spectrometer component 14 is configured to capture light emissions from the sample feedstock following atomic excitation of the sample feedstock. In the present case, the processing device 12 may obtain the emission spectrum from the spectrometer component 14 in real time. To collect the optical emission from the sample feedstock, the depicted spectrometer component 14 may comprise an armoured fiber optic cable 24 that is mounted on an adjustable rail (not shown) and directed to a spectrometer 26, such as a CCD spectrometer. In the present case, the spectrometer 26 (such as Avantes- ULS-3648, 300 lines/mm, blazed 500 nm) may be equipped with a 10 mm entrance slit to achieve a resolution of 0.6 nm. A ceramic tube 28 may be placed over the end of the fiber optic cable 24 to limit the field of view to an (approximately) 1 cm region of the flame. The fiber optic cable 24 may be positioned to collect light being emitted from a 1 cm region of the flame at different intervals along the flame. The present spectrometer unit was calibrated based on the blackbody emission of a tungsten halogen lamp.

[0068] The adjustment system 10 may also include the burner 18, which may be coupled to a source of an oxidant 30 (such as oxygen) and a source of a fuel 32 (such as acetylene). The burner 18 is configured to generate a flame from the oxidant and the fuel to atomically excite the sample feedstock and produce the light emissions for generating the emission spectrum. In the depicted embodiment of Figure 2, the burner 18 is a single nozzle burner, such as an oxyacetylene burner. The single nozzle burner is adapted to generate a long, sharp flame. For example, if a National Torch OX-O with an orifice of 0.81" is used, it may produce a 5-8" flame. Injecting the crushed feed material into the high-temperature flame results in vaporization and atomization of the solid sample, thereby eliminating the need for sample preparation. The thermal energy then results in an optical emission that is captured by the fiber optic cable 24 and measured using the spectrometer component 14. The spectrometer outputs a characteristic emission spectrum unique to the sample being analyzed. The chemical composition of the sample feedstock is then computed from the emission spectrum as described above.

[0069] The mass flow controller 16 is coupled between the processing device 12 and the burner 18, the mass flow controller 16 being configured to receive instructions from the processing device 12 and to modify flow of the oxidant and the fuel to the burner 18. In the depicted embodiment of Figure 2, the mass flow controller 16 is a calibrated Aalborg GFC that ensures that acetylene and oxygen are supplied at a constant flow according to their setpoint. This setpoint can either be controlled manually or digitally with a graphical user interface (GUI) utilizing an Arduino microcontroller 34. An example of such a GUI is illustrated in Figure 25, which indicates the general function of each button for MFC control under Tab 1. The illustrated tab gives the user control over the fuel and/or oxidant flow rates to achieve optimal flame conditions for flame emission by quickly adjusting flow rates of one, or both Arduino-controlled MFCs. Figure 26 illustrates Tab 2 of the GUI shown in Figure 25. Once the desired flame is achieved using Tab 1, a user can switch to Tab 2 to prepare the spectrometer component 14 for data collection and processing while still being able to access emergency shutdown. General functions and settings have been labelled in Figure 26.

[0070] The adjustment system 10 may further include the injection device 20. In the depicted embodiment, the injection device comprises a feed tube 36 positioned perpendicular to the flame. The feed tube 36 has a sample inlet 36a and a sample outlet 36b and is shown orientated vertically, with the sample inlet 36a positioned above the sample outlet 36b. It may, in some applications, be a 2-foot ceramic tube positioned above the flame. The (solid) sample feedstock may be introduced by means of gravity into the flame's reaction zone through the feed tube 36 using a vibrational feeder and/or a powder injector (not shown) to provide a constant feed rate. The height of the feed tube 36 increases the speed at which the sample feedstock may be introduced into the flame and helps to allow for more of the sample feedstock to fall into the inner core of the flame.

[0071] The parameter controller 22 may be coupled between the processing device 12 and the processing system 100, where the parameter controller 22 is configured to receive instructions from the processing device 12 and to automatically modify the operating parameters of the processing system 100 to the optimal parameters. As noted above, when the processing system 100 is a smelting system, optimal parameters may be parameters relating to one or more of the pressure of the pre-heated air that is injected into the flash furnace, the temperature of the pre-heated air, the level of oxygen enrichment of the preheated air, and the amount of additional oxygen that is injected into the flash furnace.

[0072] Experimental Testing

[0073] Copper: To test whether the adjustment system 10 of Figure 2 is capable of producing measurable emission spectrum, pure copper powder was introduced into the oxidizing acetylene/oxygen flame while approximately 100 emission spectra (1 ms integration time/10 averages) was collected at 1 cm intervals, up to 9 cm away from the tip. The injection of copper powder resulted in a very bright, white/green emission for the entire length of the flame.

[0074] Figure 4 is an image of the adjustment system 10 of Figure 2 in use when copper powder was introduced into the flame. The resulting emission spectra were first processed by removing any weak or saturated spectra, applying the instrument correction factor and then normalizing to a region of the spectrum that contained no emission lines. Spectra obtained 3 cm from the torch tip provided the highest signal. Figure 5 is a normalized emission spectrum that was collected 3 cm from the flame of Figure 4.

[0075] Those spectra with sufficient signal (i.e. without saturation) displayed numerous discrete emission lines, see Figure 6. Several atomic lines from alkali metals are indicated along with atomic lines from copper. Copper oxide lines are also observed. These lines were compared to the NIST spectra database and atomic lines were attributed to Na, Li, and K as well as Cu. Several features were also attributed to copper oxide (Cu(II)O) and are confirmed in the next section. The presence of Cu° and Cu ²⁺ suggested Cu ⁺¹ could also be present in the form of CU2O. A number of lines in the 400-550 nm region of the spectrum appeared to be molecular emissions bands but still require identification.

[0076] CuO Emission: The presence of copper oxide emission lines was confirmed by generating the electronic and vibrational energy levels from the known electronic and vibrational constants of CuO. This information was obtained from the book, "Molecular Spectra and Molecular Structure - Spectra of Diatomic Molecules", by Herzberg. The following equation generates the vibrational lines from Table 1.

[0077] Table 1 : Spectroscopic Constants of CuO

[0078] The resulting transitions from the 1 ^st , 2 ^nd , and 3 ^rd vibrational states of A to Xi and X2 were plotted according to Boltzmann distribution at 2500K. These describe the observed lines very well as seen in Figure 7.

[0079] Iron: Next, pure iron powder was introduced into the flame of the adjustment system 10 of Figure 2 while approximately 100 spectra was collected at 1 cm intervals, up to 9 cm away from the tip. The iron powder resulted in a very bright, white/orange emission for the entire length of the flame with incandescent particles exiting the flame (see Figure 8). Similar to the copper powder, the spectra obtained at 3 cm provided the highest signal. The resulting spectra were processed by removing any weak or saturated spectra, applying the instrument correction factor and then normalizing to a region of the spectrum that contained no emission lines. Figure 9 is the normalized emission spectrum collected 3 cm from the flame of Figure 8.

[0080] FeO Emission: Those spectra with sufficient signal (without saturation) displayed numerous discrete emission lines. These lines were compared to the NIST spectral database and atomic lines were attributed to Na, Li, and K as well as numerous Fe lines below 550 nm. In fact, the NIST database reports over 1300 transitions within this region. The same approach of using information from Herzberg was used to confirm the presence of FeO. Transitions from the a, bi, and b2 excited states were observed. Term symbols for these excited states are not assigned. Figure 10 illustrates overlaid iron powder emission spectra with expected emission lines from FeO and Fe°.

[0081] Table 2: Spectroscopic Constants of FeO

[0082] Training the Composition Neural Network

[0083] In order to train an artificial neural network (ANN) as a predictive model using emission spectral information, a high-quality dataset with a set of physical descriptors was first divided into training, validation, and testing subsets. Weights and biases were adjusted to improve the performance of the network. To measure the performance of the model, several metrics can be used. The most common of these are mean square error (MSE),

[0084] The gradient of C (-VC) continues to modify the weights and bias of each connection until a stopping criterion is met. Carefully selecting the stopping criteria is very important since overfitting is a common problem for ANN given the model's complexity. Validation and testing data subsets prevent this overfitting by ensuring the model can correctly predict its descriptors.

[0085] To serve as a predictive model for the present system, two ANNs were trained based on elemental assays and mineralogical assays, respectively. Custom code was written using pre-existing MATLAB architecture to iteratively determine the optimal ANN for a dataset by adjusting the size of the network and convergence criteria. When training a network, data was randomly segregated into training, validation, and testing subsets in a 90%: 5%: 5% ratio. Data was further randomized to prevent model bias toward sample spectra of similar composition. The network was optimized according to the training data while testing and validation data prevented overfitting.

[0086] Reference Mineral Samples: After successfully testing the adjustment system 10 of Figure 2 with pure elemental powders, the adjustment system 10 of Figure 2 was further tested by analyzing a series of reference minerals. These minerals, chalcopyrite (CuFeSz), bornite (CusFeS^, pyrite (FeSz), pyrrhotite (Fe(i- _X )S), and sphalerite ((Zn, Fe)S), are relevant to copper smelting processes and therefore make ideal candidates to test the system. Like the pure elemental samples, the powders were deposited into the flame and hundreds of emission spectra were collected for each. In addition to "pure" minerals, mineral powders were mixed at 25% intervals to produce 30 unique binary mechanical mixtures which were also analyzed. In total, 8750 spectra were collected and 7220 spectra remained after eliminating weak or saturating spectra. To train a predictive model, the dataset was randomized and separated into groups with 6498, 361, and 361 emission spectra to serve as the training, validation, and testing subsets. Two different predictive models were prepared to predict the elemental and mineralogical composition of the mineral samples by changing the number of output nodes.

[0087] For the elemental predictive model, ICP-OES elemental analysis was first conducted on the samples to obtain the training dataset. To better understand the elemental composition of each reference mineral and to produce labels for ANN training, samples were analyzed via ICP-OES for Cu, Fe, Zn, Ca, Si, and Al. The concentration of sulfur was also determined by thermogravimetric analysis (ELTRA). Surprisingly, there was a relatively large amount of silicon and zinc in the bornite and pyrrhotite samples which indicated that our reference minerals contained extraneous minerals. Figure 11 provides a graphical representation of the composition of iron/copper sulfide mineral samples as determined via ICP-OES and ELTRA (Sulfur).

[0088] For the mineral predictive model, mineral liberation analysis (MLA) was conducted on the samples to obtain the training dataset. An analysis technique called Mineral Liberation Analysis was selected to determine the mineralogical composition of the reference mineral samples. MLA uses spatially resolved energy- dispersive X-ray spectroscopy obtained through a scanning electron microscope (SEM) to determine the mineralogy of a polished sample by classifying unique phases. An energy-dispersive X-ray spectrum was collected from each phase to allow the user to identify the mineralogy of the phase. Five samples were prepared in-house for MLA by mounting and polishing samples in epoxy before being carbon- coating and analyzed using an FEI MLA 650 FEG ESEM instrument. A custom mineralogical database was generated by classifying the mineralogy for each unique phase in the sample above 0.1% surface area. After generating a mineralogical database, the modal mineralogy of the pure mineral samples was calculated by determining the total surface area of each mineral phase and then calculating the mass percentage of each phase from the density. Any mineral that did not meet the minimum matching threshold of 70% to any mineral in the database was classified as an unknown.

[0089] All samples could be well described with 17 minerals (over 97% total mass). Not surprisingly, MLA revealed that the reference minerals were, in fact, not pure. All samples contained mostly copper, iron, and zinc (sphalerite sample) containing sulfide minerals but also contains significant quantities of silicates (quartz, orthoclase) and carbonates (dolomite). The bornite sample contained 4 separate non-sulfide minerals above 1% which amounted to almost 15% of the mass of the sample.

[0090] For the elemental predictive model, a network was trained to predict the elemental composition of the 35 samples with 2272 input nodes (emission spectrum), 50 hidden nodes, and 5 output nodes (elemental composition). To prevent overfitting, the network was limited to 10 validation failures before training was stopped and a new iteration started. The network with the best performance was then selected as the best network. The graphical output in Figure 12 shows that the network was able to predict the composition of our mineral samples with a high degree of accuracy. Average correlation values (R ² ) and root mean squared error (R.MSE) for training, validation, and testing subsets were excellent at 0.26%, 1.31% and 1.03%, respectively. Elemental R.MSE also shows good accuracy for testing subsets (see Figure 12).

[0091] For the mineralogy predictive model, a second ANN was trained in an identical manner to predict the mineralogy of 35 mineral samples. The network was trained to output a prediction for the 12 most abundant minerals in the sample set (all above 2.5% in at least one sample) and the graphical output is shown in Figure 13.

[0092] R.MSE ranged from 0.4% to 3.8%. The neural network was good at predicting major components but struggled with minerals below a mass fraction of 5%. It also struggled with minerals with large amounts of alkali metals (K, Na, Li) since the emission lines from these elements were removed due to spectrometer pixel saturation.

[0093] Compared to the elemental model, the mineralogical model could not achieve the same level of predictive power. To ensure the results from the MLA software are valid, the calculated assay was compared to benchmark inductively coupled plasma (ICP) analysis. Overall, there was good agreement with ICP results except for a few major elements in chalcopyrite, pyrite, and sphalerite. The discrepancy of Fe in the pyrite sample could be the software attributing a phase of pyrrhotite to pyrite given their elemental composition is identical. Similarly, Zn and S in the sphalerite sample could have been misattributed based on the total iron contained within each phase. Minerals such as pyrrhotite and sphalerite pose a problem for this type of analysis because their chemical composition can exist in a continuum instead of discrete mineral ratios.

[0094] The elemental neural network model was then further selected to illustrate the machine learning approach to solid feedstock analysis as described below.

[0095] Testing the Trained Composition Neural Network [0096] Industrial Samples: 13 concentrated mineral powder samples were used. 12 of the samples were retrieved from their concentrate warehouse while the 13th was a sample of revert material. The revert sample was comprised of materials that were subjected to the flash smelting furnace but did not end up in the matte phase (dust or slag). Hundreds of emission spectra were obtained using the adjustment system 10 of Figure 2 and processed in the same manner described above. As expected, each sample produced a unique emission spectrum and bore similarities to those spectra obtained from the reference mineral spectra. Figure 14 show overlaid flame emission spectra (normalized) of the 12 concentrate samples.

[0097] To train a neural network from the spectra collected from the concentrate powders, a compositional label was created for each spectrum based on the results of the ICP-OES analysis.

[0098] The result of the network training can be seen in Figure 15. The trained composition neural network is now able to predict the elemental composition of an unknown sample containing similar minerals as the training sets with an accuracy ranging between 0.13-0.33%. These results show that elemental labels on the industrial samples can train a network that is far better than the previous testing on standard mineral samples.

Second Embodiment

[0099] Another example of the adjustment system 10 is shown in isolation (and in greater detail) in Figure 16. The adjustment system 10 of Figure 16 includes the processing device 12, the spectrometer component 14, the mass flow controller 16, and the parameter controller 22 as described above with regards to the adjustment system 10 shown in Figure 2. The adjustment system 10 of Figure 16, however, also includes a burner assembly 38, which comprises an alternate burner 40, and an alternate injection device 44 coupled to the alternate burner 40.

[00100] In the present embodiment, the alternate burner 40 is a ring burner that generates a cylindrical flame. The alternate burner 40 has a fuel/oxidant inlet 41 coupled to the source of the oxidant and the fuel. The ring burner may further comprise multiple apertures 42, such as eight apertures, arranged in a ring formation around an orifice 43. In some applications, each aperture 42 may be 1/32 inch in diameter. In other applications, the apertures 42 may be a different size. Unlike the burner 18, the alternate burner 40 is shown orientated vertically, such that the generated flame is directed downwards from the orifice 43. The apertures 42 are also shown to be orientated perpendicular to the longitudinal axis of the alternate burner 40 so the generated flame is cylindrical.

[00101] Similar to the injection device 20, the alternate injection device 44 also comprises an alternate feed tube 46 with a sample inlet 46a and a sample outlet 46b, where the sample inlet 46a is positioned above the sample outlet 46b. In the present embodiment, at least a portion of the alternate feed tube 46 with the sample outlet 46b is positioned coaxially with the ring burner in order to inject the feedstock coaxially into the cylindrical flame in use. In some applications, the feedstock may be injected coaxially into the cylindrical flame through the orifice 42, which may be 0.5". One advantage of the adjustment system 10 is shown in Figure 16 is that it ensures that the sample feedstock is completely introduced into the cylindrical flame.

[00102] The alternate injection device 44 may further comprise a carrier gas inlet 48 in fluid communication with the alternate feed tube 46. The carrier gas inlet 48 may be configured to receive a carrier gas for mixing with the (solid) sample feedstock in the alternate feed tube 46 before injection into the cylindrical flame. In that manner, the solid sample may be entrained by the carrier gas before injection into the cylindrical flame. The alternate injection device 44 may utilize the carrier gas to optimize flame temperature and sample feedstock introduction. Preliminary characterization of the burner was conducted using different modes of operation.

[00103] Experimental Testing

[00104] Optimization with Oxygen: Safe and optimal operating conditions for the adjustment system 10 shown in Figure 16 were determined to be 4 LPM of acetylene and oxygen at 9 and 11 psi, respectively. A neutral flame can be seen in Figure 17 operating under non-ideal (left) and ideal (right) conditions, where the inner feathers become short and sharp while the overall flame is approximately 6" in length. This is the lower limit for gas flow since any lower will result in a flashback when the alternate burner 40 is switched off. Additionally, the alternate burner 40 can and should be ignited at close to final operating conditions to prevent a sooty acetylene flame from forming.

[00105] Emission Testing: The alternate burner 40 was tested by configuring the optical fiber cable 24 to sample emission from the flame roughly 4 cm from the burner face and by passively adding sample feedstock down the alternate feed tube 46 without using the vibration feeder or powder injector. While very little emission was observed as the powder exited the burner face, the anticipated thermal emission could be seen within a few centimeters (Figure 18). This result suggests a colder section at the mouth of the burner. The spectra collected during this test were processed and normalized and fed into the neural network shown in Figure 15 to assess the quality of the data collected. The emission spectra appeared like those previously collected with the adjustment system 10 shown in Figure 2 and the predicted value from the neural network gives expected results for all elements. Figure 19 illustrates normalized emission spectra results from testing of the adjustment system of Figure 16 with oxygen/acetylene flame using the vibration feeder, and Figure 20 illustrates output from the trained composition neural network for elemental composition.

[00106] Nitrous Oxide with Active Injection : To characterize the performance of our coaxial burner with nitrous oxide (N2O) as the oxidant and the use of oxygen as a carrier gas, the alternate burner 40 was tested using 3 different operating modes by measuring the emission spectrum. Overall, it was found that nitrous oxide provided a cooler, smaller flame which resulted in a reduced emission intensity when compared to the oxygen/acetylene flame.

[00107] Neutral Flame without O2 Injection: A neutral flame, with equivalent stoichiometry, was obtained when the inner feathers of the flame became short and rounded. The oxidant and fuel flows corresponded to 3.38 LPM and 1.9 LPM of nitrous oxide and acetylene, respectively. Passively injecting a mineral powder into the flame resulted in emission spectra with less intensity than the oxygen/acetylene flame produced, notably for copper and iron atomic emission lines. This is perhaps an indicator that more molecular information is retained with the cooler flame. Figure 21 shows images of a neutral nitrous oxide-acetylene flame using the second example adjustment system of Figure 15 without oxygen injection (top), with powder injection without oxygen (bottom), and the resulting emission spectra (inset).

[00108] Neutral Flame with O2 Injection: After obtaining a neutral flame, a stream of oxygen was turned on and a mineral powder was actively injected into the flame. This resulted in emission spectra with much less intensity (see Figure 22) than the previous mode of operation. Figure 22 shows images of a neutral nitrous oxide-acetylene flame using the second example adjustment system of Figure 15 with oxygen injection (top), with powder injection aided with oxygen (bottom), and the resulting emission spectra (inset).

[00109] Rich Flame with O2 Injection: A rich flame was obtained by increasing the flow of acetylene to 3 LPM which caused the inner feathers of the flame to become long and luminous (see Figure 23). A mineral powder was actively injected into the flame together with an oxygen stream resulting in emission spectra with a similar maximum intensity to the first operating mode. The turbulent mixing of carrier gas and rich fuel/oxidant gases dispersed the sample and produces an inner section closer to a neutral flame. However, the collected spectra also exhibited more intensity variability. Figure 23 show images of a rich nitrous oxide-acetylene flame using the second example adjustment system of Figure 15 with oxygen injection (top), with powder injection aided with oxygen (bottom), and the resulting spectra (inset).

[00110] Overall, the use of the adjustment system 10 with the spectrometer component 14 and the trained composition neural network was able to predict the elemental composition of an unknown sample containing similar minerals with good accuracy and in real-time. Thus, the adjustment system 10 allows for real-time identification of optimal parameters for feedforward automation to the processing system 100. Consequently, the processing system 100 may continuously process the feedstock under optimal parameters, as its operating parameters are also continuously being modified in response to the composition of the feedstock being fed into the adjustment system 10 and the processing system 100.

Third Embodiment

[00111] Another embodiment of the burner assembly of the adjustment system 10 is shown in isolation and in greater detail in Figures 27 to 32. The alternate burner assembly 49 of Figures 27 to 32 may utilize the processing device 12, the spectrometer component 14, the mass flow controller 16, and the parameter controller 22 as described above with regards to the adjustment system 10 shown in Figure 2 and Figure 16. The alternate burner assembly 49 of Figure 27 to 32, however, includes a further alternate burner 50 and a further alternate injection device 52 coupled to the further alternate burner 50.

[00112] In the embodiment shown in Figures 27 to 32, the further alternate burner 50 and the further alternate injection device 52 may be secured to a support frame 53 with a vibrational mount 54, a powder injector 55, and a containment vessel 56. The vibrational mount 54 and the powder injector 55 may help to improve sample introduction.

[00113] A collection canister 58 may be attached to the containment vessel 56 to capture any solid waste from the combustion process. The containment vessel 56 may include an exhaust port 60 (that can be attached to a gas handling system), and a viewport 62, that can be used to inspect burner operation therein. The viewport 62 may be 1.5" in diameter. The containment vessel 56 may further include an optical port 63 which may be coupled to the fiber optic cable 24 to collect the optical emission from the sample feedstock. The optical emission would then be directed to a spectrometer 26, as described above.

[00114] The support frame 53 shown in Figures 27 to 29 sits on caster wheels 64, which allows the alternate burner assembly 49 to be moved, and leveling feet 66, which allows the alternate burner assembly 49 to be held stationary when desired.

[00115] Turning to Figure 30, the further alternate burner 50, the further alternate injection device 52, and the vibrational mount 54 may be seen in greater detail. Similar to the alternate burner 40, the further alternate burner 50 is a ring burner with the fuel/oxidant inlet 41 and multiple apertures 68 arranged in a ring formation around the orifice 70. The ring burner may comprise eight apertures 68, where each aperture 68 may be 1/32 inch in diameter. In the depicted embodiment, unlike the alternate burner 40, each of the apertures 68 are orientated at an angle relative to the longitudinal axis of the further alternate burner 50 so the generated flame is conical. The angle of the apertures 68 may be 20 to 60 degrees relative to the longitudinal axis of the further alternate burner 50. In the depicted embodiment, the angle of the apertures 68 is 26.5 degrees. In this embodiment the burner holes were angled to avoid the formation of a large colder zone close to the burner face as described above, resulting in greater emission close to the burner face and when flowing the carrier gas.

[00116] The further alternate injection device 52 may be coupled to the further alternate burner 50. Similar to the injection device 42, the further alternate injection device 52 also comprises a further alternate feed tube 72 with a sample inlet 72a and a sample outlet 72b, where the sample inlet 72a is positioned above the sample outlet 72b. Also similar to the injection device 42, at least a portion of the further alternate feed tube 72 and the sample outlet 72b are positioned coaxially with the further alternate burner 50 to inject the sample coaxially into the flame.

[00117] Also similar to the alternate burner 40, the further alternate burner 50 may additionally include the carrier gas inlet 48 in fluid communication with the further alternate feed tube 72. The carrier gas inlet 48 may be configured to receive a carrier gas for mixing with the sample feedstock in the further alternate feed tube 72 before injection into the flame. The carrier gas may be used to optimize flame temperature and sample feedstock introduction. [00118] In the present embodiment, the further alternate injection device 52 may be a Venturi injection device. As illustrated in Figures 31 and 32, the Venturi injection device may have a Venturi portion 73 comprising a first wide section 73a, a middle constricted second 73b, and a second wide section 73c. The Venturi portion 73 may be fluidly coupled to, and between, the carrier gas inlet 48 and the sample inlet 72a of the further alternate feed tube 72. The Venturi portion 73 functions by flowing the carrier gas (from the carrier gas inlet 48) through the first wide section 73a to the small or constricted section 73b, and then into the second wide section 73c. This induces a pressure drop in the second wide section 73c. In the present embodiment, the first wide section 73a has an inner diameter of 0.32". The constricted or small section 73b may be, or include, a nozzle with an aperture (that may be 0.03" in diameter, for example), to produce a Venturi effect and actively entrain and inject the feedstock coaxially through the orifice 70 (which may be 0.5" in diameter) within the conical flame in use. The diameters of the Venturi injection device may vary in other applications.

[00119] The further alternate feed tube 72 further has a sample conduit 72c that extends at an angle o from the further alternate feed tube 72 to a funnel 72d. In the depicted embodiment, the angle o is approximately 35 degrees. In other applications, the angle o may be between 25 and 45 degrees. The sample conduit 72c is further shown in fluid communication with the second wide section 73c of the further alternate feed tube 72. In that manner, the sample received by the funnel 72d may be directed into the second wide section 73c via the sample conduit 72c for mixing with the carrier gas before entering the further alternate feed tube 72 via the sample inlet 72a.

[00120] One advantage of the Venturi injection device is that, unlike the injection device 44, which passively entrains particles into the flow of carrier gas, the Venturi injection device can produce a small vacuum that actively pulls material into the carrier gas stream, which helps to improve mass transport. In the depicted embodiment of Figures 31 and 32, the further alternate feed tube 72 is also slightly narrower than the second wide section 73c, but is wider than the middle constricted second 73b. This may also help to improve mass transport.

[00121] Returning to Figure 30, the containment vessel 56 may be coupled with the further alternate burner 50 though a flanged connector 74. The flanged connector 74 is shown in Figure 30 surrounding at least a portion of the further alternate burner 50 and forming an interior sheath space 76 coaxially between the flanged connector 74 and the further alternate burner 50, and also forming a sheath gas outlet 78 coaxially around the orifice 70 of the further alternate burner 50. The flanged connector 74 is shown having a sheath gas inlet 80 in fluid communication with the interior sheath space 76 and the sheath gas outlet 78. The sheath gas inlet 80 may be configured to receive a sheath gas for ejection through the sheath gas outlet 78 coaxially around the flame.

[00122] One advantage to the use of sheath gas is the cooling it may provide to the further alternative burner 50 when in the interior sheath space 76. Additionally, the sheath gas helps to prevent build up on the burner face from recirculating material in containment vessel 56 and helps to direct material away from the optical port 63. The use of sheath gas in the manner may also help reduce the risk of exposure by the containment vessel 56 to designated substances that may be contained within feedstock.

[00123] Experimental Testing

[00124] Optimization with nitrous oxide: Safe and optimal operating conditions for the alternate burner assembly 49 shown in Figures 27 to 29 were determined to be 1.25 LPM and 3 LPM of acetylene and nitrous oxide at 9 and 11 psi, respectively. A neutral flame can be seen in Figure 33 operating under ideal (right) conditions without active injection or sheath gas, where the inner feathers become short and sharp while the overall flame is approximately 6" in length. This flow rate provides sufficient thermal heating and is the lower limit for gas flow since any lower will result in a flashback when the further alternate burner 50 is switched off. Similar to the alternate burner 40, the further alternate burner 50 can and should be ignited at close to final operating conditions to prevent a sooty acetylene flame from forming. The sheath gas flow rate and venturi injection were determined to be 10 LPM of air and 1 LPM of air, respectively.

[00125] Sample Analysis: The further alternate burner 50 was tested by configuring the optical fiber cable 24 with the optical port 63 to "look" at the flame roughly 6" from the burner face through the optical port 63 to limit the field of view to 1 cm. Sample feedstock is fed down the further alternate feed tube 72 using the vibrational mount 54, and powder injector 55, and Figure 34 and Figure 35 illustrates output from the trained composition neural network for elemental composition.

[00126] Reference Mineral Samples: After successfully testing the alternate burner assembly 49 with pure elemental powders, the alternate burner assembly 49 was tested by analyzing a series of reference minerals. These minerals, chalcopyrite (CuFeSz), bornite (CusFeS^, pyrite (FeSz), pyrrhotite (Fe(i- _X )S), and sphalerite ((Zn, Fe)S), are relevant to copper smelting processes and therefore make ideal candidates to test the system. A normalized representative emission spectrum produces from the alternate burner assembly 49 in the adjustment system 10 for each concentrate sample can be seen in Figure 36.

[00127] Custom Concentrate Samples: 12 concentrated mineral powder samples were prepared from the 13 industrial samples described above. 12 of the samples were retrieved from their concentrate warehouse while the 13th was a sample of revert material. A normalized, representative emission spectrum produced from the alternate burner assembly 49 in the adjustment system 10 for each concentrate sample can be seen in Figure 37.

[00128] To train a neural network to predict the elemental composition from the spectra collected from the reference mineral and custom concentrate powders, a compositional label was created for each spectrum based on the results of the ICP-OES analysis. [00129] The result of the network training can be seen in Figure 35. The trained composition neural network is now able to predict the elemental composition of a sample containing similar minerals as the training set with an accuracy ranging between 0.23-0.67%.

[00130] To train a neural network to predict the mineralogical composition from the spectra collected from the reference mineral and custom concentrate powders, a compositional label was created for each spectrum based on the results of the mineral liberation analysis for those mineral phases that were present in abundance greater than 5% (w/w%).

[00131] The result of the network training can be seen in Figure 35. The trained composition neural network is now able to predict the mineralogical composition of an unknown sample containing similar minerals as the training set with an accuracy ranging between 0.27-1.75%. This is a large improvement over results obtained using adjustment system 10 shown in Figure 2.

Method

[00132] Figure 24 is a flowchart illustrating an example method 500 for optimizing processing of feedstock by a processing system. The example method 500 may be performed by the adjustment system 10, the burner assemblies 38, the alternate burner assembly 49, and the processing system 100 as described above using the processing device 12, for example. In particular, the method 500 may be performed in real-time, where the analysis of the feedstock and determination of optimal operating parameters may be performed in-line sequentially and/or simultaneously with the processing of the feedstock.

[00133] At 502, the emission spectrum of a sample of the feedstock may first be generated. For example, the emission spectrum may be generated using the spectrometer component 14 as described above. To generate the emission spectrum, light emission from atomically excited feedstock is captured at 504. One manner in which to atomically excite the sample feedstock is to apply high heat, such as with a flame. The flame may be generated with a burner using oxidant and fuel to atomically excite the sample feedstock. In some applications, a single long, sharp flame may be generated. In another application, at 506, a cylindrical or conical flame may be generated.

[00134] To adjust the temperature and/or quality of the flame, the flow of the oxidant and the flow of the fuel to the burner may be modified. To further adjust the temperature of the flame, a carrier gas may be mixed with the sample feedstock at 508 just before it is injected into the flame. The flow rate of the carrier gas may also be modified when it is mixed with the sample feedstock before injection into the flame.

[00135] In the case when a cylindrical or conical flame is generated, at 510, the sample feedstock and optional carrier gas may be injected coaxially within the flame. The flame atomizes and thermally excites the sample feedstock, which release light/optical emissions when the atoms return to their ground state. Once the light/optical emission is captured, such as by the fiber optic cable 24, the optical emission is diffracted, and its wavelengths are detected, such as by the spectrometer 26, to generate the emission spectrum of the sample feedstock. Optionally, a sheath gas may be injected coaxially around the flame.

[00136] At 512, the method 500 includes determining the chemical composition of the sample feedstock from the corresponding emission spectrum. The emission spectrum may be obtained at 514 directly after its generation at 502. Alternatively, the emission spectrum may be obtained by retrieving it from memory either internally or externally. At 516, the composition of the sample feedstock is computed from the emission spectrum using a trained composition neural network.

[00137] As discussed above, the composition neural network may be trained as a predictive model in a known manner using a high-quality emission spectral dataset with a set of physical descriptors. When training the network, the emission spectra data may be randomly segregated into training, validation, and testing subsets in a 90%: 5%: 5% ratio. Data may be further randomized to prevent model bias toward sample spectra of similar composition. The trained composition neural network was optimized according to the training data while testing and validation data prevented overfitting.

[00138] At 518, based on the elemental composition of the sample feedstock as determined at 516, optimal parameters for processing the feedstock are identified. The optimal parameters may be determined by consulting an optimal parameters database, or by inputting values form the determined composition into equations to calculate the optimal parameters. The optimal parameters may be different depending on the corresponding composition of the feedstock. For example, in the case where the processing of the feedstock is smelting the feedstock in a flash furnace, its operating parameters may include one or more of the pressure of pre-heated air that is injected into the flash furnace, the temperature of the pre-heated air, the level of oxygen enrichment of the preheated air, and the amount of additional oxygen that may be injected into the flash furnace. Other parameters may alternatively or additionally be involved depending on the requirements of the particular smelting process. Thus, the optimal parameters would be the parameters that maximize the matte output and minimize the slag output for the particular feedstock currently being processed.

[00139] At 520, once the optimal parameters are identified, instructions are then sent to the processing system to modify its operating parameters to match that of the identified optimal parameters. For example, the instructions may be sent to parameter controller 22, which is coupled to the parameter system 100. At 522, the parameter controller 22 may then automatically modify the operating parameters to match the identified optimal parameters in real-time.

[00140] In the case where the processing of the feedstock is smelting the feedstock in a flash furnace (at 524), its operating parameters, including one or more of the pressure of pre-heated air that is injected into the flash furnace, the temperature of the pre-heated air, the level of oxygen enrichment of the preheated air, and the amount of additional oxygen injected into the flash furnace (at 526), may be automatically and continually modified in order to match the identified optimal parameters as the feedstock is being smelted. Operating under such conditions, the quality and/or quantity of the matte output should be maximized.

[00141] Overall, the use of the method 500 with the trained composition neural network allows the elemental composition of an unknown sample containing similar minerals to be predicted with good accuracy and in real-time. Thus, the method 500 allows for real-time identification of optimal parameters for feedforward automation when processing the feedstock. Consequently, a continuous stream of feedstock may be processed under optimal parameters, as the operating's system's operating parameters may also be continuously modified in real-time in response to the composition of the feedstock.

[00142] While the processing system 100 has been discussed in terms of a smelting system, the processing system 100 may instead be a different system which processes a feedstock whose composition can be analyzed and determined using emission spectra.

[00143] Although the present disclosure describes methods and processes with operations (e.g., steps) in a certain order, one or more operations of the methods and processes may be omitted or altered as appropriate. One or more operations may take place in an order other than that in which they are described, as appropriate.

[00144] All values and sub-ranges within disclosed ranges are also disclosed. Also, although the systems, devices and processes disclosed and shown herein may comprise a specific number of elements/components, the systems, devices and assemblies could be modified to include additional or fewer of such elements/components. For example, although any of the elements/components disclosed may be referenced as being singular, the embodiments disclosed herein could be modified to include a plurality of such elements/components. The subject matter described herein intends to cover and embrace all suitable changes in technology. [00145] When the functions are implemented in the form of a software functional unit and sold or used as an independent product, the functions may be stored in a computer-readable storage medium. Based on such an understanding, the technical solutions of this disclosure essentially, or the part contributing to the prior art, or some of the technical solutions may be implemented in a form of a software product. The software product is stored in a storage medium and includes several instructions for instructing a computer device (which may be a personal computer, a server, or a network device) to perform all or some of the steps of the methods described in the embodiments of this application. The foregoing storage medium includes any medium that can store program code, such as a universal serial bus (USB) flash drive, a removable hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, among others.

[00146] The present disclosure may be embodied in other specific forms without departing from the subject matter of the claims. The described example embodiments are to be considered in all respects as being only illustrative and not restrictive. Selected features from one or more of the above-described embodiments may be combined to create alternative embodiments not explicitly described, features suitable for such combinations being understood within the scope of this disclosure.

Table of reference numerals

10 adjustment system

12 processing device

14 spectrometer component

16 mass flow controller

18 burner

20 injector device

22 parameter controller

24 fiber optic cable

26 spectrometer

28 ceramic tube

30 oxidant source 2 fuel source 4 Arduino microcontroller 6 feed tube a sample inlet b sample outlet 8 burner assembly 0 alternate burner 1 fuel/oxidant inlet 2 aperture 3 orifice 4 alternate injection device 6 alternate feed tube a sample inlet b sample outlet 8 carrier gas inlet 9 alternate burner assembly 0 further alternate burner 2 further alternate injection device3 support frame 4 vibrational mount 5 powder injector 6 containment vessel 8 collection canister 0 exhaust port 2 viewport 3 optical port 4 caster wheels 6 leveling feet 8 apertures 0 orifice 2 further alternate feed tube a sample inlet b sample outlet c sample conduit d funnel 3 Venturi portion a first wide section b constricted middle section c second wide section 4 flanged connector 6 interior sheath space sheath gas outlet sheath gas inlet processing system feedstock source processor

I/O interface network interface storage memory analysis module optimized parameter data