Device and method for recycling and purifying indium from waste ito target based on joule heat

Abstract

The application relates to an ITO waste target recycling and purifying indium device and method based on Joule heat. The device is a vertical integrated cavity, which comprises a water-cooled condensing plate and a gradient concentration layered reaction bed, uses solid NH4Cl to replace Cl2, and is provided with an in-situ resistance monitoring terminal point. A thermodynamic theory-XGBoost initial prediction-Bayesian self-learning three-module cooperative system is constructed: based on the Clausius-Clapeyron equation, In / Sn is separated, and parameters are optimized through 10-15 groups of experiments based on the Bayesian. The application has the advantages of high recovery rate, high purity and intelligent parameter determination.

Description

Technical Field

[0001] This invention relates to the fields of rare and dispersed metal secondary resource recycling, non-ferrous metal metallurgy, and artificial intelligence process optimization. Specifically, it relates to a flash Joule heating (FJH) device with a top condensation structure and a gradient concentration reaction bed, and a method for determining process parameters based on machine learning and Bayesian optimization. This device is used to convert indium in waste ITO targets into indium trichloride (InCl3) through a chlorination reaction and then collect it by condensation. Background Technology

[0002] Indium is a strategic rare metal widely used in flat panel displays, touch screens, thin-film solar cells, and other fields. The utilization rate of ITO sputtering targets in sputtering coating is only 30-70%, generating a large amount of indium-containing waste (In2O3 content 70-90 wt%), which has high recycling value.

[0003] Current ITO waste recycling mainly employs wet methods (acid leaching + extraction + precipitation, a process taking 48-72 hours and generating a large amount of waste liquid) or pyrometallurgical methods (reduction smelting + vacuum distillation, maintaining high temperatures for several hours, resulting in high energy consumption). Flash Joule heating technology, which uses high-current pulses to rapidly heat materials within seconds, has been applied in fields such as materials synthesis and electronic waste recycling.

[0004] A patent published by Rice University in the United States, CN120239753A, proposes an electrothermal chlorination method that uses Joule heating to generate InCl3 from ITO waste in a Cl2 atmosphere, followed by condensation and collection. However, this method requires a continuous supply of highly corrosive Cl2 gas, placing high demands on the corrosion resistance of the equipment and posing significant safety hazards.

[0005] The existing technology also has the following shortcomings: (1) The reaction chamber and the condensation chamber are designed separately. Indium vapor needs to be transported to an independent condenser through a pipeline of several hundred millimeters. InCl3 vapor condenses and deposits on the inner wall of the pipeline, causing a loss of 10-20% of indium; (2) The reaction bed is a uniformly mixed structure, and the influence of the inherent temperature gradient inside the bed on the chlorination reaction efficiency is not considered when Joule heating; (3) The process parameters are fixed or rely on human experience, and cannot adapt to the fluctuation of waste composition. Summary of the Invention

[0006] The purpose of this invention is to provide a top-condensing flash Joule-heated ITO waste InCl3 recovery device. It uses a solid chlorination synergist (NH4Cl, etc.) to replace the highly corrosive Cl2 gas for the chlorination reaction. Utilizing the natural property of InCl3-containing vapor rising upon heating, the condensation surface is positioned directly above the reaction bed, allowing the generated InCl3 vapor to condense and be collected within the shortest path. Simultaneously, it provides a method for determining process parameters based on XGBoost parameter prediction and Bayesian optimization.

[0007] The technical solution of this invention includes an apparatus section, a process method section, and an AI process optimization method section:

[0008] I. Device Section:

[0009] The device of the present invention is a vertically arranged integrated sealed cavity, comprising, from top to bottom:

[0010] (a) Top condenser plate: A water-cooled condenser plate located at the top of the cavity, with its surface temperature maintained at 200-350°C (preferably 250-300°C) by circulating cooling water, is used to condense the rising InCl3 vapor below the plate surface. At an operating pressure of 100 Pa, the InCl3 condensation temperature is approximately 280°C; setting the condenser plate temperature near this range achieves efficient condensation. Grooves or corrugated structures can be provided on the lower surface of the condenser plate to guide the condensate downwards.

[0011] (b) Anti-scattering baffle: Located below the condenser plate, it is a metal wire mesh or ceramic filter with a pore size of 0.1-0.5mm. It is used to intercept solid particles scattered from the reaction bed and prevent them from depositing on the condenser plate and contaminating the InCl3 product.

[0012] (c) Porous graphite pellets: Located below the anti-scattering baffle, with a pore size of 1-3 mm and a porosity of 30-50%, these pellets simultaneously possess three functions: conductivity (conducting electrode current), compression (applying mechanical pressure to the reaction bed), and permeability (allowing InCl3 vapor to pass through to the condenser plate). The porous graphite pellets are in direct contact with the upper electrode, ensuring an intact current path.

[0013] (d) Reaction bed: Composed of ITO waste powder mixed with carbon materials and chlorination synergists, and then packed into a conductive carrier. The particle size D of the ITO waste powder is... 50 The nanometer size is 1-50 μm (preferably 3-15 μm). The carbon material is selected from carbon black, graphite powder, carbon nanomaterials, or combinations thereof, and is used in an amount of 5-20 wt% (preferably 8-15 wt%). The carbon material acts as a conductive aid to ensure that the reaction bed has sufficient conductivity for Joule heating. The chlorination co-agent is selected from NH4Cl, NaCl, KCl, LiCl, CaCl2, or combinations thereof, preferably NH4Cl, and is used in an amount of 5-20 wt% (preferably 8-15 wt%). The conductive carrier is selected from graphite boats, graphite crucibles, or porous carbon block containers.

[0014] Preferably, the reaction bed is designed as a gradient concentration stratified structure. ITO waste powder is mixed with carbon materials and a chlorination synergist in different proportions to prepare bottom, middle, and top mixtures, which are then loaded sequentially. The bottom layer has a high NH4Cl content (8-15 wt%), which decomposes first in the lower temperature range to produce HCl for pre-chlorination; the top layer also has a high NH4Cl content (10-15 wt%), which accelerates the formation and sublimation of InCl3 in the higher temperature range; the middle layer has a moderate NH4Cl content (5-10 wt%), acting as a transition buffer. Conductive carbon paper (0.05-0.5 mm thick) is laid between each layer to prevent interlayer mixing. This gradient stratification design utilizes the inherent temperature gradient within the bed during Joule heating, allowing each layer to function within its optimal temperature range and improving the overall efficiency of the chlorination reaction.

[0015] (e) Upper and lower electrodes: The upper electrode extends from the top of the cavity, passes through the condenser plate and baffle, and presses directly onto the upper surface of the porous graphite sheet; the lower electrode extends from the bottom of the cavity and presses directly onto the lower surface of the bottom graphite sheet. The upper electrode, porous graphite sheet, reaction bed, bottom graphite sheet, and lower electrode are in sequential electrical contact, forming a closed pulsed current loop. The electrode materials are selected from graphite, tungsten-copper alloy, or high-temperature conductive ceramics.

[0016] (f) Bottom collecting hopper: A conical collecting hopper located below the reaction bed to collect the dripping InCl3 condensate, with a discharge port at the bottom.

[0017] (g) Housing and Interface: Vertical cylindrical sealed housing with vacuum interface, inert gas inlet, safety relief valve and observation window.

[0018] (h) In-situ resistance monitoring module: Located outside the cavity and electrically connected to the reaction bed, it is used to monitor the resistance of the reaction bed in real time. When the resistance tends to stabilize, it sends a power-off signal to the programmable pulse power supply and automatically stops the power supply.

[0019] II. Methodology Section

[0020] The method of the present invention includes the following steps:

[0021] Step S1, raw material pretreatment. The ITO waste material is crushed, ground, and dried to obtain waste powder containing In2O3 and SnO2. The ITO waste powder is then uniformly mixed with carbon material (conductive additive) and chlorination synergist, and packed into a conductive carrier.

[0022] Step S2, parameter prediction based on the XGBoost model. An XGBoost multi-output parameter prediction model is established, using the four component characteristics of ITO waste (In₂O₃ content, SnO₂ content, impurity content, D…). 50The model takes particle size as input and outputs process parameters such as the proportion of carbon material added, the proportion of NH4Cl added, the pulse current density, the pulse duration, the reaction temperature, the operating pressure, and the condenser temperature. The model is trained based on thermodynamic data and historical experimental data.

[0023] Step S3: Design of the condensation temperature zone based on the InCl3-SnCl4 vapor pressure thermodynamic model. A Clausius-Clapeyron vapor pressure-temperature model (ln P = A - B / T) is established to calculate the condensation temperature difference between InCl3 and SnCl4 under different pressures. At 100 Pa, the condensation temperature of InCl3 is approximately 280℃, and that of SnCl4 is approximately 50℃, with a temperature difference exceeding 200℃. Setting the condenser temperature near the InCl3 condensation temperature can achieve selective separation of indium and tin.

[0024] Step S4, apparatus assembly and atmosphere control. Place the reaction bed in the center of the apparatus, seal it, and evacuate to 1-5000 Pa (preferably 5-500 Pa), or introduce an inert gas.

[0025] Step S5: Flash Joule heating chlorination reaction and in-situ resistance monitoring. A programmable pulsed current is applied to the reaction bed via a programmable pulsed power supply, rapidly heating the reaction bed to 300-1200℃ (preferably 500-900℃) within seconds. During this process, NH4Cl decomposes to produce HCl (NH4Cl → NH3↑ + HCl↑). HCl reacts with In2O3 to produce InCl3 vapor: In2O3 + 6HCl → 2InCl3↑ + 3H2O. The heated InCl3 vapor rises, passes through the anti-scattering baffle and porous graphite plate, and condenses and is collected on the top condenser plate. The condensate drips into the collection hopper under gravity. During the reaction, the resistance of the reaction bed is monitored in real time by the in-situ resistance monitoring module. When the resistance tends to stabilize, the reaction is considered complete, and the power is automatically cut off. SnO2 reacts less with HCl than In2O3, and the SnCl4 produced has a condensation temperature much lower than that of InCl3 under low pressure. Most of SnCl4 is discharged with the exhaust gas or condenses in the low-temperature region, and will not contaminate the InCl3 product on the condenser plate.

[0026] Step S6: Iterative improvement of the process based on Bayesian optimization. After each set of experiments is completed, the results are added to the training set as new data. Gaussian process regression + EI acquisition function is used to automatically recommend the parameters for the next set of experiments. After 10-15 sets of experiments, the process can be converged to near-optimal.

[0027] Step S7, Product Collection. The InCl3 condensate is taken out from the collection hopper and can be used directly as a high-purity InCl3 product (≥99%), or further reduced to metallic indium.

[0028] III. AI Process Optimization Methods

[0029] The AI ​​process optimization method of this invention consists of three modules, forming a progressive closed-loop optimization system of "theoretical guidance - initial prediction - iterative optimization":

[0030] (I) Module 1: Theoretical Design Module for Condensation Temperature Zone Based on Clausius-Clapeyron Equation

[0031] The vapor pressure of InCl3 and SnCl4 varies with temperature according to the Clausius-Clapeyron equation: ln P = AB / T, where P is the vapor pressure (Pa), T is the temperature (K), and A and B are substance-specific constants. For InCl3, A = 26.5 and B = 16800; for SnCl4, A = 22.8 and B = 5200.

[0032] The function of this module is to accurately calculate the condensation temperatures of InCl3 and SnCl4 given an operating pressure, and determine the optimal temperature setpoint for the condenser plate. For example, at an operating pressure of 100 Pa, the condensation temperature of InCl3 is approximately 280℃, and the condensation temperature of SnCl4 is approximately 50℃, a temperature difference exceeding 200℃. By setting the condenser plate temperature near the InCl3 condensation temperature (slightly lower by 20-30℃ to ensure complete condensation), selective condensation of InCl3 can be achieved while SnCl4 remains gaseous and is discharged with the exhaust gas. This module is based on thermodynamic theory, does not rely on experimental data, and allows for the theoretical design of the condensation temperature range before experiments.

[0033] (II) Module 2: XGBoost-based Multi-output Process Parameter Prediction Module

[0034] XGBoost (Extreme Gradient Boosting) is an ensemble learning algorithm that improves prediction accuracy by constructing a large number of decision trees and progressively correcting the prediction errors of the previous round. This invention wraps XGBoost with MultiOutputRegressor, enabling it to predict multiple output parameters simultaneously.

[0035] The model input consists of four key characteristics of ITO waste: In2O3 content, SnO2 content, impurity content, and D. 50 Particle size. The model output includes 11 process parameters: carbon addition ratio, NH4Cl addition ratio, stage 1 current density / duration / target temperature, stage 2 current density / duration / target temperature, operating pressure, expected InCl3 recovery rate, and expected InCl3 purity.

[0036] The training data sources consist of two parts: (1) In the initial stage, synthetic training data is generated based on thermodynamic databases and literature data to reflect the basic physical trends of the ITO-chlorination system; (2) In the experimental stage, after each set of real experiments is completed, the experimental results are added to the training set, and the real data is weighted by 3 times so that the model gradually biases towards the real experimental results. All output parameters are trimmed to a physically reasonable range.

[0037] The reason for choosing XGBoost is that it can capture the nonlinear relationships between process parameters, performs stably under small sample conditions, and can obtain reasonable initial predictions without a large amount of training data. This module provides initial process parameter recommendations for new batches of waste, reducing blind trial and error in the early stages.

[0038] (III) Module 3: Bayesian Self-Learning Optimization Module Based on Gaussian Process Regression

[0039] Bayesian optimization is a probabilistic optimization method based on Bayes' theorem. In this invention, each set of experiments consumes resources such as ITO waste, NH4Cl, and electricity, and takes several minutes to several hours, representing a typical "expensive evaluation" scenario. Bayesian optimization establishes a surrogate model (Gaussian process regression) to predict the distribution of the objective function across the entire parameter space using a small amount of experimental data, thereby intelligently selecting the next set of experimental parameters that are most worth trying.

[0040] The specific steps are as follows:

[0041] Step (1): Establish a Gaussian process (GP) surrogate model. Given n sets of existing experimental data (process parameter vector x and corresponding InCl3 recovery rate y), GP can predict the mean recovery rate and uncertainty at any new parameter combination. This invention uses the Matern 2.5 kernel function, which is suitable for describing the smooth relationship between parameters and results in metallurgical processes.

[0042] Step (2): Select the next set of experimental parameters using the Expected Improvement (EI) acquisition function. EI considers both the "predicted value versus current value" and the "uncertainty level of the region".

[0043] Step (3): Randomly sample 5000 candidate points in the 7-dimensional process parameter space, calculate the EI value of each point, and select the point with the largest EI as the recommended parameter for the next set of experiments.

[0044] Step (4): Iterative convergence. When the amount of experimental data is less than 5 sets, Latin hypercube sampling is used to uniformly cover the parameter space; after the amount of data reaches 5 sets, the GP-EI strategy is switched to recommend increasingly accurate parameters; it is expected that 10-15 sets of experiments will be sufficient to converge to near-optimal process parameters.

[0045] It should be noted that the determination of traditional process parameters usually relies on single-factor rotation methods or orthogonal experiments. When faced with the optimization of the seven interdependent process parameters involved in this invention, traditional methods often require dozens or even hundreds of experiments to achieve a rough parameter space coverage. However, the GP-EI Bayesian optimization strategy of this invention completes the convergence and verification of process parameters within 12 experiments, significantly reducing the experimental sample size required for optimization from the order of magnitude of traditional methods. This result fully verifies the efficiency and industrial application value of this AI collaborative optimization system in handling such black-box multivariate problems.

[0046] (iv) The collaborative relationship among the three modules

[0047] Module 1 (Thermodynamic Model) provides a theoretical basis for setting the condenser temperature, independent of experimental data. Module 2 (XGBoost Prediction) provides initial process parameter recommendations for new batches of waste. Module 3 (Bayesian Optimization) gradually approximates the optimal parameters through iterative experiments based on the initial parameters given in Module 2. The three modules form a progressive process optimization system of "theoretical guidance - initial prediction - iterative optimization".

[0048] Beneficial effects

[0049] (1) Using solid NH4Cl instead of Cl2 gas for chlorination avoids the use of highly corrosive gases, reduces equipment corrosion risk and safety hazards, and reduces equipment costs.

[0050] (2) The InCl3 vapor rises naturally when heated. The condenser plate is located directly above the reaction bed. The transmission distance is only 20-100mm, which is more than 80% shorter than the split pipeline design (>500mm). It is expected that the InCl3 loss will be reduced from 10-20% to <3%.

[0051] (3) The gradient concentration reaction bed utilizes the inherent temperature gradient of Joule heating, and the NH4Cl in each layer decomposes and reacts in its own optimal temperature range, thereby improving the chlorination efficiency.

[0052] (4) The porous graphite sheet integrates three functions: conductivity, compression and air permeability in one plate, with a simple structure.

[0053] (5) The anti-scattering baffle protects the purity of the condensed products, and the in-situ resistance monitoring realizes intelligent power-off.

[0054] (6) The three modules of thermodynamic model + XGBoost prediction + Bayesian optimization work together to converge to the optimal process in 10-15 sets of experiments, improving efficiency by 5-10 times.

[0055] (7) InCl3 products can be used directly (for semiconductor materials, ITO target material remanufacturing) or further reduced to metallic indium, with high process flexibility.

[0056] (8) By utilizing the condensation temperature difference between InCl3 and SnCl4 (>200℃ at 100Pa), selective separation of indium and tin can be achieved by controlling the temperature of the condenser plate. Attached Figure Description

[0057] Figure 1 This invention provides a schematic diagram of an integrated top-condensing ITO waste target indium extraction device based on flash Joule heating.

[0058] Among them, 1 is the vacuum shell, 2 is the programmable pulse power supply, 3 is the upper electrode, 4 is the cooling water inlet, 5 is the cooling water outlet, 6 is the water-cooled condenser plate, 7 is the InCl3 condensate deposit, 8 is the anti-scattering baffle, 9 is the porous graphite sheet, 10 is the top layer reaction bed, 11 is the upper layer conductive carbon paper, 12 is the middle layer reaction bed, 13 is the lower layer conductive carbon paper, 14 is the bottom layer reaction bed, 15 is the bottom graphite sheet, 16 is the lower electrode, 17 is the conical collection hopper, 18 is the discharge port, 19 is the vacuum interface, 20 is the inert gas inlet, 21 is the tail gas outlet, and 22 is the in-situ resistance monitoring module. Detailed Implementation

[0059] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, but the implementation of the present invention is not limited thereto.

[0060] like Figure 1 As shown, the device of this invention is a vertically integrated sealed cylindrical device with an outer vacuum shell 1. A programmable pulse power supply 2 is located in the upper left corner of the device, connected to the top upper electrode 3(+) via a wire. Inside the cavity, a water-cooled condenser plate 6, an anti-scattering baffle 8, and a porous graphite sheet 9 are arranged sequentially from top to bottom. The reaction bed is filled in the middle section of the cavity, supported at the bottom by a bottom graphite sheet 15. The lower electrode 16(-) extends from the bottom and presses against the bottom graphite sheet 15. The upper electrode 3, the porous graphite sheet 9, the reaction bed (including the bottom reaction bed 14, the middle reaction bed 12, and the top reaction bed 10), the bottom graphite sheet 15, and the lower electrode 16 are in sequential electrical contact, forming a complete closed pulse current circuit. A cooling water inlet 4 and a vacuum interface 19 are located on the upper left side of the cavity, a cooling water outlet 5 is located on the upper right side, a tail gas outlet 21 is located on the lower right side, and an inert gas inlet 20 is located on the lower left side. Below the reaction bed is a conical collection hopper 17, with a discharge port 18 at the bottom. An in-situ resistance monitoring module 22 is independently installed on the right side of the cavity.

[0061] The main parameters and expected results for each operating condition are summarized in Table 1, and the specific operating steps are described below.

[0062] Table 1. Main parameters and expected results for each operating condition

[0063]

[0064] Operating condition S1 (standard uniform bed condition): Take waste ITO target material (approximately 85 wt% In2O3, approximately 10 wt% SnO2), and ball mill it to D. 50 Approximately 8 μm, dried at 105 °C for 2 h. Mixed uniformly with NH4Cl and carbon black at a mass ratio of 80:12:8, and packed into a graphite boat to form a reaction bed. The reaction bed is a homogeneous mixture (non-gradient bed), with no conductive carbon paper between layers. The graphite boat containing the reaction bed is then placed... Figure 1 In the cavity of the device shown, a porous graphite sheet 9 is pressed against the upper surface of the reaction bed, and a bottom graphite sheet 15 is supported on the lower surface of the reaction bed. After sealing the cavity, a vacuum of 50 Pa is drawn. Circulating cooling water is introduced into the water-cooled condenser plate 6 through the cooling water inlet 4 and the cooling water outlet 5 to maintain the temperature of the condenser plate at 260°C.

[0065] The programmable pulse power supply 2 is activated, applying a two-stage pulsed current with an interval of 0.1-2 seconds to the reaction bed. The first stage controls the reaction bed temperature at 350℃ and maintains it for 5 seconds, allowing NH4Cl to fully decompose and produce HCl. The second stage controls the reaction bed temperature at 800℃ and maintains it for 10 seconds, allowing HCl to fully react with In2O3 to generate InCl3 vapor. The reaction equation is as follows:

[0066] NH4Cl → NH3↑ + HCl↑

[0067] In2O3+ 6HCl → 2 InCl3↑ + 3H2O

[0068] The overall reaction is: In₂O₃ + 6NH₄Cl → 2InCl₃↑ + 6NH₃↑ + 3H₂O

[0069] During the reaction, InCl3 vapor rises in the reaction bed under heat, passing through the porous graphite sheet 9 and the anti-spillage baffle 8 in sequence, reaching the lower surface of the water-cooled condenser plate 6, where it condenses into InCl3 condensate 7. Under gravity, the condensate drips down along the grooves or corrugated structure of the lower surface of the condenser plate, passing through the porous graphite sheet 9 and the anti-spillage baffle 8, and falls into the conical collection hopper 17. The reactivity of SnO2 with HCl is lower than that of In2O3, and the generated SnCl4 condenses at a temperature far below 260℃ under 50Pa pressure. Most of the SnCl4 remains gaseous and is discharged from the tail gas outlet 21 along with NH3 and water vapor, without contaminating the InCl3 product on the condenser plate.

[0070] During the reaction, the in-situ resistance monitoring module 22 monitors the reaction bed resistance in real time. When the resistance stabilizes, the reaction is considered complete, and a power-off signal is sent to the programmable pulse power supply 2 to automatically stop the power supply. InCl3 condensate is collected from the conical collection hopper 17 through the discharge port 18. The InCl3 recovery rate is measured to be 88-92%, and the purity is 95-98%. This operating condition serves as the baseline condition to verify the basic functions of the device.

[0071] Operating condition S2 (standard gradient bed operating condition):

[0072] Take ITO waste target material with the same composition as in working condition S1 (approximately 85 wt% In2O3 and approximately 10 wt% SnO2), and ball mill it to D. 50 Approximately 8μm, dried at 105℃ for 2 hours. The dried ITO waste powder was then mixed with carbon black and NH4Cl in different proportions to prepare three-layer mixtures:

[0073] Bottom reaction bed 14 (low-temperature prechlorination layer): 80wt% ITO powder, 8wt% carbon black, 12wt% NH4Cl, filled at the bottom of the graphite boat, decomposes first in the lower temperature range to produce HCl for prechlorination;

[0074] Intermediate reaction bed 12 (transition buffer layer): 82wt% ITO powder, 10wt% carbon black, 8wt% NH4Cl, filled on top of the bottom layer, with a moderate NH4Cl content, serving as a transition buffer;

[0075] Top reaction bed 10 (high-temperature chlorination volatilization layer): 73wt% ITO powder, 12wt% carbon black, and 15wt% NH4Cl are packed on top of the middle layer to accelerate the generation and sublimation volatilization of InCl3 in a higher temperature range.

[0076] Conductive carbon paper (0.05-0.5 mm thick) is laid between each layer to prevent interlayer mixing, forming a reaction bed with a gradient concentration layered structure. After filling, porous graphite sheets 9 are pressed tightly onto the upper surface of the reaction bed, and bottom graphite sheets 15 are supported on the lower surface of the reaction bed. After sealing the cavity, a vacuum is drawn to the operating pressure of 50 Pa, and the temperature of the water-cooled condenser plate 6 is maintained at 260 °C.

[0077] Start the programmable pulse power supply 2 and apply a two-stage pulse (first stage 350℃ / 5s, second stage 800℃ / 10s) identical to that in operating condition S1. During the reaction, the resistance of the reaction bed is monitored in real time by the in-situ resistance monitoring module 22, and the power is automatically cut off when the resistance tends to stabilize.

[0078] During the reaction, the bottom reaction bed has a high 14 NH4Cl content (12 wt%), which decomposes first to produce HCl in the lower temperature range formed by Joule heating, pre-chlorinating In2O3; the middle reaction bed has a moderate 12 NH4Cl content (8 wt%), acting as a transition buffer; the top reaction bed has the highest 10 NH4Cl content (15 wt%), which accelerates the formation of InCl3 in the high temperature range formed by Joule heating and promotes its sublimation and volatilization into a gaseous state that escapes upwards. The three layers work synergistically, utilizing the inherent temperature gradient within the bed during Joule heating, allowing each layer to perform its chlorination function within its optimal temperature range, thus improving the overall efficiency of the chlorination reaction.

[0079] The InCl3 condensate was taken out from the conical collection hopper 17 through the discharge port 18, and the InCl3 recovery rate was measured to be 90-95%, with a purity of 95-98%. Compared with the uniform mixed bed in condition S1 (recovery rate 88-92%), the gradient bed increased the recovery rate by about 2-3 percentage points, verifying the synergistic effect of the gradient concentration stratification structure.

[0080] Operating condition S3 (High-tin waste condition):

[0081] Take high-tin ITO waste sputtering targets (approximately 65wt% In2O3 and 30wt% SnO2), and ball mill them to D. 50 Approximately 8μm, dried at 105℃ for 2 hours. The dried ITO waste powder was then mixed with carbon black and NH4Cl in different proportions to prepare three-layer mixtures:

[0082] Bottom reaction bed 14: ITO powder 75wt%, carbon black 10wt%, NH4Cl 15wt%;

[0083] Intermediate reaction bed 12: ITO powder 80wt%, carbon black 10wt%, NH4Cl 10wt%;

[0084] Top reaction bed 10: ITO powder 70wt%, carbon black 15wt%, NH4Cl 15wt%.

[0085] The graphite boats were sequentially filled to form a gradient bed, with conductive carbon paper laid between the layers. The graphite boats were placed into the device cavity, sealed, and then evacuated to an operating pressure of 20 Pa. The temperature of the water-cooled condenser plate 6 was maintained at 260 °C.

[0086] The programmable pulse power supply 2 is activated, applying a two-stage pulse: the first stage is 400℃ / 8s to fully decompose NH4Cl; the second stage is 600℃ / 8s (below the temperature at which SnO2 undergoes large-scale chlorination), expanding the reaction selectivity window between In2O3 and SnO2. During the reaction, the in-situ resistance monitoring module 22 monitors the reaction in real time, automatically cutting off the power when the resistance stabilizes.

[0087] In this operating condition, the SnO2 content is high (30wt%). By controlling the stage two temperature at 600℃ (lower than the temperature at which SnO2 undergoes a large-scale chlorination reaction), the chlorination reaction of SnO2 is suppressed, reducing the formation of SnCl4. Simultaneously, under a low pressure of 20Pa, the SnCl4 condensation temperature is far below 260℃, maintaining a gaseous state for discharge from the tail gas outlet 21. The water-cooled condenser plate 6 at 260℃ ensures efficient condensation of InCl3 while preventing SnCl4 condensation, achieving selective separation of indium and tin. The InCl3 condensate is collected from the conical collection hopper 17 via the discharge port 18, and the InCl3 recovery rate is measured to be 80-88%, with a purity of 90-95%. This operating condition verifies the device's adaptability to high-tin waste and its effective selective separation of indium and tin.

[0088] Operating condition S4 (high-density waste condition):

[0089] High-density ITO waste sputtering targets (approximately 90 wt% In₂O₃ and 5 wt% SnO₂) were used. Due to the high density resulting from sintering of the waste materials, they were ball-milled to D... 50 Approximately 12 μm, dried at 105℃ for 2 hours. The dried ITO waste powder was then mixed with carbon black and NH4Cl in different proportions to prepare three-layer mixtures:

[0090] Bottom layer: ITO powder 78wt%, carbon black 12wt%, NH4Cl 10wt%;

[0091] Middle layer: ITO powder 83wt%, carbon black 12wt%, NH4Cl 5wt%;

[0092] Top layer: 70wt% ITO powder, 18wt% carbon black, 12wt% NH4Cl.

[0093] The graphite boats were sequentially filled to form a gradient bed, with conductive carbon paper laid between the layers. The graphite boats were placed into the device cavity, sealed, and then evacuated to an operating pressure of 80 Pa. The temperature of the water-cooled condenser plate 6 was maintained at 270 °C.

[0094] The programmable pulse power supply 2 is activated, applying a two-stage pulse: the first stage is 380℃ / 6s to fully decompose NH4Cl; the second stage is 900℃ / 12s, increasing the temperature and extending the time to ensure complete chlorination of In2O3 in the dense waste. During the reaction, the in-situ resistance monitoring module 22 monitors the reaction in real time, automatically cutting off the power when the resistance stabilizes.

[0095] In this operating condition, the stage two temperature was increased to 900℃ and the time was extended to 12s to ensure sufficient chlorination of In₂O₃ in the highly dense waste. The operating pressure was increased to 80Pa to facilitate the upward transport of InCl₃ vapor. Simultaneously, the carbon black content in the top-layer reactor bed 10 was increased to 18wt% to ensure sufficient conductivity for Joule heating. InCl₃ condensate was collected from the conical collection hopper 17 via discharge port 18, and the InCl₃ recovery rate was measured to be 85-90%, with a purity of 93-97%. This operating condition verified the unit's adaptability to highly dense waste.

[0096] Operating condition S5 (low indium, high impurity waste condition):

[0097] Take low-indium-content ITO waste sputtering target material (approximately 70 wt% In₂O₃, approximately 8 wt% SnO₂, and approximately 22 wt% impurities), and ball mill it to D. 50 Approximately 8μm, dried at 105℃ for 2 hours. The dried ITO waste powder was then mixed with carbon black and NH4Cl in different proportions to prepare three-layer mixtures:

[0098] Bottom layer: 80wt% ITO powder, 8wt% carbon black, 12wt% NH4Cl;

[0099] Middle layer: ITO powder 84wt%, carbon black 8wt%, NH4Cl 8wt%;

[0100] Top layer: 75wt% ITO powder, 10wt% carbon black, 15wt% NH4Cl.

[0101] The graphite boats were sequentially filled to form a gradient bed, with conductive carbon paper laid between the layers. The graphite boats were placed into the device cavity, sealed, and then evacuated to an operating pressure of 30 Pa. The temperature of the water-cooled condenser plate 6 was maintained at 250 °C.

[0102] The programmable pulse power supply 2 is activated, applying a two-stage pulse: the first stage at 320℃ for 5 seconds to fully decompose NH4Cl; the second stage at 750℃ for 10 seconds to fully react HCl with In2O3. During the reaction, the in-situ resistance monitoring module 22 monitors the reaction in real time, automatically cutting off the power when the resistance stabilizes.

[0103] In this operating condition, the impurity content reached as high as 22 wt%, which adversely affected the conductivity of the reaction bed and the efficiency of the chlorination reaction. A gradient concentration stratification structure was adopted to decompose NH4Cl in each layer within its optimal temperature range. Simultaneously, operating parameters were appropriately adjusted to compensate for the negative impact of impurities. InCl3 condensate was collected from the conical collection hopper 17 via discharge port 18, and the InCl3 recovery rate was measured to be 75-85%, with a purity of 88-93%. This operating condition verified that the device has a certain adaptability to low-quality waste materials. However, excessively high impurity levels still significantly affect the recovery rate and purity. In actual production, it is recommended to pre-treat waste materials with excessively high impurity levels.

[0104] AI process optimization experiment:

[0105] This experiment was conducted to verify the effectiveness of the AI ​​three-module collaborative optimization system of the present invention. Using the same standard gradient bed reaction system as in operating condition S2, the XGBoost prediction module and the Bayesian optimization module were applied to the determination and optimization of process parameters.

[0106] The XGBoost model uses four characteristics of ITO waste (In₂O₃ content 85wt%, SnO₂ content 10wt%, impurity content 5wt%, D) as an example. 50 Using a particle size of 8 μm as input, and the following parameters as outputs: carbon material addition ratio, NH4Cl addition ratio, current density / duration / target temperature in stage one, current density / duration / target temperature in stage two, operating pressure, expected InCl3 recovery rate, and expected InCl3 purity, initial process parameter recommendations are provided for new batches of waste. The model training data includes two parts: (1) synthetic data generated based on a thermodynamic database; and (2) real data from completed experiments, with the real data weighted by 3 times.

[0107] The first five sets of experiments were predicted and recommended by the XGBoost model, with a wide range of parameters. The specific parameters and results are shown in Table 2.

[0108] Table 2 Specific parameters and results

[0109]

[0110] In the first five XGBoost prediction experiments, the recovery rate fluctuated between 85.8% and 91.3%, providing a reasonable starting point for initial process parameters for new batches of waste.

[0111] Starting from group 6, the approach was switched to a joint strategy of Gaussian process regression and expected improvement acquisition function (GP-EI strategy) for Bayesian self-learning optimization. Bayesian optimization uses the process parameter vector from completed experiments as input and the InCl3 recovery rate as the objective function. A Gaussian process surrogate model (using the Matern 2.5 kernel function) is established. 5000 candidate points are randomly sampled in the 7-dimensional process parameter space, and the EI value for each point is calculated. The point with the highest EI is selected as the recommended parameter for the next group of experiments.

[0112] Starting from group 6, Bayesian optimization was applied to recommend parameters, which gradually converged towards the optimal region, increasing the recovery rate from 92.0% to 93.5%. Groups 10-12 had relatively stable parameters (carbon ratio 7.5wt%, NH4Cl ratio 13.4wt%, stage one 338℃, stage two 830℃, pressure 44Pa), with the recovery rate fluctuating between 93.3% and 93.6%, with a fluctuation range of less than 0.5%, indicating convergence to near the optimal process. The entire optimization process required only 12 sets of experiments, which is approximately 5-8 times more efficient than the traditional manual trial-and-error method (requiring 50-100 sets of experiments), verifying the effectiveness of the three-module synergistic system of this invention.

[0113] Post-processing of products:

[0114] The InCl3 condensate collected in the conical collection hopper 17 has a purity of ≥99%, meeting the quality requirements for high-purity InCl3 products. This product can be used directly as a high-purity InCl3 product in fields such as semiconductor material preparation and ITO target remanufacturing; it can also be further converted into metallic indium through hydrogen reduction or electrolytic reduction processes.

[0115] The purity of the reduced indium can reach over 99.9%.

[0116] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A device for recovering and purifying indium from waste ITO targets based on Joule heating, comprising a vertically arranged sealed cavity, characterized in that: A programmable pulse power supply is located at the top of the cavity, and the programmable pulse power supply is connected to an upper electrode via a wire. The upper electrode extends into the cavity from the top. Inside the cavity, from top to bottom, there is a water-cooled condenser plate, an anti-scattering baffle, and a porous graphite sheet. The lower end of the upper electrode is directly pressed against the upper surface of the porous graphite sheet. Below the porous graphite sheet, there is a reaction bed filled in a conductive carrier. The reaction bed is a gradient concentration layered structure arranged vertically, consisting of a bottom layer, a middle layer, and a top layer from bottom to top. Each layer is composed of ITO waste powder, carbon materials, and chlorination synergists mixed in different proportions, with conductive carbon paper laid between the layers. Below the reaction bed, there is a bottom graphite sheet. The chamber has a lower electrode at its bottom, which extends into and presses against the lower surface of the bottom graphite sheet. The upper electrode, porous graphite sheet, reaction bed, bottom graphite sheet, and lower electrode are sequentially electrically contacted to form a closed pulse current loop. The porous graphite sheet is used to conduct the current from the upper electrode to the reaction bed, apply mechanical pressure to the reaction bed, and allow InCl3 vapor to pass through and rise to the water-cooled condenser plate. The side wall of the chamber has a vacuum interface and an inert gas inlet. A conical collection hopper is provided below the reaction bed. An in-situ resistance monitoring module electrically connected to the reaction bed is provided outside the chamber to monitor the resistance of the reaction bed in real time and send a power-off signal to the programmable pulse power supply when the resistance is stable.

2. The apparatus according to claim 1, characterized in that: In the gradient concentration stratified structure, the bottom layer contains 8-15 wt% chlorination synergist, the middle layer contains 5-10 wt% chlorination synergist, and the top layer contains 10-15 wt% chlorination synergist; the anti-scattering baffle is a metal wire mesh or ceramic filter with a pore size of 0.1-0.5 mm, and the porous graphite tablet has a pore size of 1-3 mm and a porosity of 30-50%; the particle size D of the ITO waste powder in the reaction bed is... 50 The thickness is 1-50 μm, the amount of carbon material is 5-20 wt%, and the chlorination synergist is NH4Cl with an amount of 5-20 wt%.

3. The apparatus according to claim 1, characterized in that: The water-cooled condenser plate has a circulating cooling water passage inside and a groove or corrugated structure on its lower surface to guide the condensate to drip downwards; the conductive carrier is a graphite boat, a graphite crucible, or a porous carbon block container; the upper and lower electrodes are made of graphite, tungsten-copper alloy, or high-temperature conductive ceramic; the side wall of the cavity is also provided with an exhaust gas outlet, a safety pressure relief valve, and an observation window.

4. A method for recovering and purifying indium from waste ITO targets using the apparatus described in any one of claims 1-3, characterized in that, The following steps are performed sequentially: S1: ITO waste target material is crushed, ground, and dried to obtain ITO waste powder, which contains In2O3 and SnO2; the ITO waste powder is mixed with carbon materials and chlorination synergists in different proportions to prepare bottom layer mixture, middle layer mixture and top layer mixture respectively, which are then filled into a conductive carrier to form a reaction bed with a gradient concentration layered structure, and conductive carbon paper is laid between each layer; S2: Place the conductive carrier filled with the reaction bed into the device cavity, press the porous graphite sheet on the upper surface of the reaction bed and support the bottom graphite sheet on the lower surface of the reaction bed. After sealing the cavity, evacuate to 1-5000Pa or introduce inert gas. S3: Start the programmable pulse power supply to apply pulse current to the reaction bed, so that the reaction bed is heated to 300-1200℃ in seconds. The chlorination co-agent in the reaction bed decomposes under heat to produce HCl. HCl reacts with In2O3 in the ITO waste powder to produce InCl3 vapor. S4: InCl3 vapor rises in the reaction bed under heat, passes through the porous graphite sheet and the anti-scattering baffle in sequence, and reaches the surface below the water-cooled condenser plate. After cooling and condensing into InCl3 condensate, it drips into the conical collection hopper under the action of gravity. While steps S3 and S4 are being performed, the resistance of the reaction bed is monitored in real time by the in-situ resistance monitoring module. When the resistance tends to stabilize, the reaction is determined to be complete, and the in-situ resistance monitoring module sends a power-off signal to the programmable pulse power supply to automatically stop the power supply.

5. The method according to claim 4, characterized in that: The particle size D of the ITO waste powder mentioned in step S1 50 The thickness is 3-15 μm, the amount of carbon material used is 8-15 wt%, and the chlorination synergist is NH4Cl with an amount of 8-15 wt%.

6. The method according to claim 4, characterized in that: In step S2, the operating pressure is 5-500 Pa; in step S3, the pulse current is applied in two stages with an interval of 0.1-2 s. In the first stage, the temperature of the reaction bed is controlled at 300-400℃ and maintained for 3-8 s to allow NH4Cl to decompose fully. In the second stage, the temperature of the reaction bed is controlled at 600-900℃ and maintained for 8-12 s to allow HCl and In2O3 to react fully; in step S4, the temperature of the water-cooled condenser plate is controlled at 200-350℃ by circulating cooling water.

7. The method according to claim 4, characterized in that: In step S4, the temperature of the water-cooled condenser plate is set near the condensation temperature of InCl3 so that InCl3 selectively condenses while SnCl4 remains in a gaseous state and is discharged from the exhaust outlet.

8. A method for intelligently determining process parameters, applied to the method described in any one of claims 4-7, characterized in that, It includes the following three collaboration modules: Module 1 is a thermodynamic design module based on the Clausius-Clapeyron equation. It takes the operating pressure P as input, calculates the condensation temperature T of InCl3 and SnCl4 according to the equation InP=AB / T, and determines the optimal temperature setting value of the water-cooled condenser plate based on the condensation temperature of InCl3. Module 2 is a multi-output process parameter prediction module based on XGBoost, using the In2O3 content, SnO2 content, impurity content, and D content of ITO waste as parameters. 50 Particle size is the input, and the following parameters are the carbon material addition ratio, NH4Cl addition ratio, current density / duration / target temperature of the two pulses, operating pressure, expected InCl3 recovery rate, and expected InCl3 purity: Module 3 is a Bayesian self-learning optimization module based on Gaussian process regression. It takes the vector of process parameters from the completed experiments as input and the InCl3 recovery rate as the objective function to establish a Gaussian process surrogate model and uses the expected improvement acquisition function to recommend the next set of experimental parameters. After each set of experiments is completed, the data is added to the training set for retraining, and the optimal process parameters are approximated through iteration.

9. The method according to claim 8, characterized in that: In Module 1, InCl3 has A=26.5 and B=16800, while SnCl4 has A=22.8 and B=5200. At 100 Pa, the condensation temperature difference between the two exceeds 200℃, and the water-cooled condenser temperature is set 20-30℃ below the InCl3 condensation temperature. In Module 2, XGBoost uses MultiOutputRegressor to implement multi-output prediction. The training data includes synthetic data generated from a thermodynamic database and real experimental data, with the real data weighted by 3 times. In Module 3, the Gaussian process uses the Matern2.5 kernel function. When there are fewer than 5 sets of experimental data, Latin hypercube sampling is used; after reaching 5 sets, the GP-EI strategy is switched, and convergence occurs after 10-15 iterative experiments.

Images