A method of ore color separation

Abstract

The application discloses a kind of ore color selection methods, it is related to ore color selection technical field, method includes through visible light camera and near-infrared camera acquisition ore multimodal optical characteristics, in combination with environmental light fluctuation interference, determine the light feature mapping value reflected by friction characteristics;Based on the light feature mapping value reflected by friction characteristics, in combination with ore morphology and vibrating feeder state, determine the calibration coefficient of material dropping timing;Based on the calibration coefficient of material dropping timing, in combination with high-pressure pneumatic injection valve group operating state and historical blowing feedback coefficient, determine the pulse correction value of injection valve, realize impurity accurate separation;Periodically statistics batch injection valve pulse correction value overall deviation, reverse calibration near-infrared weight coefficient.The application fuses visible light and near-infrared characteristics, eliminates environmental light interference, realizes material level accurate identification;Through dynamic calibration material dropping timing and injection valve pulse, adapt to ore movement difference and injection valve aging, and have closed loop self iteration function, significantly improve separation precision and continuous operation stability.

Description

Technical Field

[0001] This invention relates to the field of ore color sorting technology, specifically to an ore color sorting method. Background Technology

[0002] Ore color sorting is a key process in mineral processing to improve concentrate grade and reduce subsequent processing costs. Its core principle is to remove impurities by utilizing differences in the optical properties of the ores. Currently, mainstream color sorting technologies in the industry generally employ a framework of visible light imaging combined with fixed color thresholds and fixed spray valve parameters. This framework can meet basic needs when processing ores with simple compositions and significant color differences. However, as the demand for sorting complex ores increases, the limitations of existing technologies are becoming increasingly apparent.

[0003] Currently, existing technologies rely solely on visible light cameras to capture the grayscale values ​​of ore surface color as a basis for judgment. This only reflects the apparent color of the ore, making it difficult to distinguish between similar-colored but different-composition ores (such as white quartz and white fluorite) and dissimilar-colored ores with darker colors but acceptable composition (such as concentrate contaminated by mud and water). Furthermore, ambient light fluctuations can distort the grayscale values. Secondly, existing technologies use a uniform theoretical material drop time as the triggering basis for the spray valve, failing to consider the differences in the friction coefficient of individual ore particles and their irregular shapes. The equipment is affected by factors such as vibration feed frequency fluctuations, which often result in the problem of missing the correct airflow and mistakenly blowing the wrong airflow, leading to high leakage rates and carry-over ratios. In addition, the existing technology uses fixed-width air jet pulses, which does not take into account the effects of differences in ore movement speed, air pressure attenuation due to aging of the spray valve, and fluctuations in ore density. At the same time, there is no feedback calibration mechanism for the performance of the equipment during operation. As the lens becomes dusty, the chute wears out, and the spray valve ages, the sorting accuracy will decrease month by month, requiring weekly shutdowns for manual recalibration of parameters, which seriously affects the continuity of production. Summary of the Invention

[0004] The purpose of this invention is to provide a method for color sorting of ores, which solves the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution, applicable to an ore color sorting device comprising a main inlet shell, a beneficiation shell, and an electrical control box, wherein the method includes:

[0006] The ore to be inspected enters from the feed hopper, and the multimodal optical features of the ore are collected by a visible light camera and a near-infrared camera. Combined with the ambient light fluctuation interference in the detection area, the true photosensitive feature mapping value of the ore to be inspected is determined. The photosensitive feature mapping value is used to distinguish the differences between the apparent color and the internal material properties of the ore.

[0007] Based on the frictional characteristics of the ore material reflected by the light-sensing feature mapping value, and combined with the ore morphology characteristics and the real-time operating status of the vibrating feeder, the ore dropping time calibration coefficient is determined. The dropping time calibration coefficient is used to characterize the degree of deviation between the actual sliding speed and the theoretical speed of a single ore.

[0008] Based on the ore movement characteristics reflected by the material feeding timing calibration coefficient, and combined with the real-time operating status of the high-pressure pneumatic spray valve group and the historical blow-off feedback coefficient, the spray valve pulse correction value corresponding to the impurities to be removed is determined. The spray valve pulse correction value is used to drive the high-pressure pneumatic spray valve group to accurately separate impurities.

[0009] Periodically analyze the overall deviation of the batch spray valve pulse correction value and reverse-calibrate the near-infrared weighting coefficient used in the calculation of the light-sensing feature mapping value.

[0010] Optionally, the process of obtaining the light-sensing feature mapping value includes:

[0011] The average visible light gray value of the ore to be inspected, the near-infrared gray values ​​collected by at least two of the near-infrared cameras, the real-time ambient light brightness of the detection area, and the standard average gray value of the standard background plate under the near-infrared channel are obtained.

[0012] The ratio of near-infrared resolution to visible light resolution is used as the near-infrared weighting coefficient to perform weighted fusion of multi-channel near-infrared grayscale values, thereby obtaining the near-infrared material characteristic coefficient.

[0013] The average gray value of visible light is used as the basic recognition benchmark, coupled with the near-infrared material characteristic coefficient, and then dynamically calibrated through the ambient light interference coefficient to obtain the light-sensing feature mapping value.

[0014] Optionally, the process of obtaining the near-infrared material characteristic coefficients includes:

[0015] The near-infrared gray values ​​of the ore to be inspected are acquired by the upper and lower near-infrared cameras, and then multiplied by the near-infrared weighting coefficients respectively and summed to obtain the near-infrared feature weighted sum.

[0016] The weighted sum of the near-infrared features is multiplied by the total number of near-infrared channels and the standard average gray value. The first ratio is then normalized to obtain the near-infrared material feature coefficient in the range of 0-1. The closer the near-infrared material feature coefficient is to 1, the higher the matching degree between the ore material and the standard concentrate.

[0017] Optionally, the process of obtaining the material feeding timing calibration coefficient includes:

[0018] The light-sensing feature ratio obtained is the ratio of the light-sensing feature mapping value of the ore to be inspected to the average light-sensing feature value of the standard concentrate of the same type, the second ratio of the estimated ore mass value to the product of the standard density value and the projected area, and the vibration frequency ratio of the real-time frequency of the vibrating feeder to the calibrated frequency.

[0019] The ratio of light-sensing characteristics is used as a reference for friction characteristic correction. It is multiplied and fused with the theoretical slip time and the ratio of inertia coefficient and vibration frequency corresponding to the ore morphology to obtain the material drop timing calibration coefficient. The actual ratio of light-sensing characteristics has a positive adjustment effect on the degree of friction characteristic correction.

[0020] Optionally, the process of obtaining the second ratio includes:

[0021] The projected area of ​​the ore to be inspected is obtained by image segmentation from the visible light camera, and the ore quality estimate is calculated by combining the standard density value and shape factor of the ore type.

[0022] The square root of the product of the estimated ore mass, the standard density, and the projected area is taken to obtain the inertia coefficient corresponding to the ore shape. The larger the inertia coefficient, the higher the density of the ore and the smaller the sliding resistance.

[0023] Optionally, the process of obtaining the injection valve pulse correction value includes:

[0024] The following parameters are used: the material drop timing calibration coefficient divided by the theoretical material drop time speed correction coefficient; the ratio of the actual trajectory offset of the previous light-sensing feature and similar particle size ore to the standard average trajectory offset as the historical blow-off feedback coefficient; and the ratio of the actual air pressure of the spray valve to the rated air pressure as the spray valve efficiency coefficient.

[0025] The standard jet pulse width of ore of the same particle size is multiplied and fused with the velocity correction coefficient, the historical blow-off feedback coefficient, and the jet valve efficiency coefficient to obtain the jet valve pulse correction value.

[0026] Optionally, the process of obtaining the historical blow-off feedback coefficient includes:

[0027] The photoelectric sensor at the waste hopper inlet collects the actual trajectory offset of the previous ore with similar light-sensing characteristics and particle size to the current ore.

[0028] The historical blow-off feedback coefficient is obtained by comparing the actual trajectory offset with the standard average trajectory offset of the ore under the standard pulse. When the historical blow-off feedback coefficient is greater than 1, it indicates that the previous blow-off force was insufficient and the pulse width of this time needs to be increased.

[0029] Optionally, the process of reverse calibration of the near-infrared weighting coefficients includes:

[0030] After each preset number of sorting actions are completed, the average value of all the spray valve pulse correction values ​​in that batch is calculated, and the deviation rate between the average value and the calibrated standard pulse value is calculated.

[0031] When the deviation rate exceeds the preset threshold, the near-infrared weighting coefficient is adjusted in reverse according to the deviation ratio to correct the calculation logic of the light-sensing feature mapping value.

[0032] Optionally, the method further includes:

[0033] When the light-sensing feature mapping value of the ore to be inspected falls within the preset qualified range, it is judged as qualified concentrate, and the ore slides naturally down the inclined chute to the finished product hopper.

[0034] When the light-sensing feature mapping value of the ore to be inspected exceeds the preset qualified range, it is determined to be an impurity to be removed. The trigger time of the spray valve is determined according to the material falling sequence calibration coefficient. The spray valve of the corresponding channel of the high-pressure pneumatic spray valve group is driven to open according to the spray valve pulse correction value, blowing the impurities over the material distribution guide baffle, and driving the inclined chute to rotate through the flow plate and be sent into the waste hopper.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] I. This invention addresses the limitations of existing technologies that rely solely on visible light recognition. First, it weights and normalizes near-infrared multi-channel grayscale values ​​according to mineral type differentiation to obtain near-infrared material characteristic coefficients. This couples visible light grayscale values, which originally only reflected color, with near-infrared spectral characteristics that reflect molecular-level composition. This retains the high-resolution color recognition capability of visible light while achieving accurate material-level verification, thus solving the problem of misjudgment of minerals of the same color but different textures, or different colors but the same texture. Then, it dynamically calibrates the fused features through an ambient light interference coefficient, eliminating interference from ambient light fluctuations, surface reflections, and watermarks, allowing it to adapt to changes in ambient light without frequent shutdowns for calibration.

[0037] Second, this invention addresses the error problem of fixed material feeding timing in existing technologies. First, it correlates the friction coefficient between the ore and the chute by using the ratio of the optical characteristics of the actual optical characteristics to those of the standard concentrate, thus correcting the difference in sliding speed between ores of different materials. Second, it corrects the difference in sliding resistance of irregularly shaped ores by taking the square root of the second ratio of the product of the ore mass and the standard density value and the projected area. Third, it corrects the overall conveying speed drift caused by the fluctuation of the feeder frequency by using the reciprocal of the vibration frequency ratio.

[0038] Third, this invention addresses the problems of poor robustness and lack of closed-loop control in existing fixed pulse technologies. First, it matches the ore movement speed and airflow action time by using a speed correction coefficient between the material drop time and the theoretical time, avoiding insufficient blowing force due to excessive speed or wasted airflow due to excessively slow speed. Next, it achieves feedforward feedback control by using the ratio of the blowing offset of the same type of ore in the previous test, dynamically adapting to the blowing force requirements of ores with different densities. Finally, it compensates for the air pressure decay caused by the aging of the spray valve by using the reciprocal of the actual air pressure of the spray valve, adapting to the performance changes of the spray valve throughout its entire life cycle without the need for manual calibration.

[0039] In addition, by statistically analyzing the deviation of batch pulse values, the infrared weighting coefficient of the front end is calibrated in reverse, realizing a closed-loop self-iteration of the entire process of detection, decision-making and execution. This improves the continuous running time of the equipment without manual calibration, reduces maintenance costs, and reduces the accuracy decay caused by long-term operation. Attached Figure Description

[0040] Figure 1 This is a flowchart of the color sorting method for this ore;

[0041] Figure 2 This is a schematic diagram of the overall structure of the apparatus for the ore color sorting method of the present invention;

[0042] Figure 3 This is a cross-section of the apparatus for the ore color sorting method of the present invention. Figure 1 ;

[0043] Figure 4 For the present invention Figure 3 Enlarged schematic diagram of the structure at point A;

[0044] Figure 5 This is a cross-section of the apparatus for the ore color sorting method of the present invention. Figure 2 .

[0045] In the diagram: 1-Main ore feed shell, 2-Ore dressing shell, 3-Electrical control box, 4-Feed hopper, 5-Vibrating feeder, 6-Visible light camera, 9-Inclined chute, 10-Near-infrared camera, 11-High-pressure pneumatic spray valve assembly, 12-Distribution guide baffle, 13-Flow plate, 14-Finished product hopper, 15-Waste hopper. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] This embodiment provides a method for ore color sorting. The overall idea is as follows: by fusing visible light and near-infrared multimodal visual features, ambient light interference is intelligently decoupled to achieve accurate identification of the ore's surface color and internal material; furthermore, based on the ore's friction characteristics and morphological information retrieved from visual features, the material feeding sequence is dynamically calibrated; finally, by combining the real-time status of the spray valve and historical blowing effects, the spray valve pulse is adaptively corrected, and a closed-loop self-iterative mechanism is established for the entire process, thereby solving the problems of low sorting accuracy and poor long-term operational stability under complex ore types and dynamic working conditions.

[0048] This ore color sorting method is primarily designed for use in the ore sorting stage of mining processing. For example... Figures 2 to 5 As shown, this method is applicable to an ore color sorting device comprising a main inlet shell 1, a beneficiation shell 2, and an electrical control box 3. The main inlet shell 1 is a vertical main structure, on which are installed a feed hopper 4, a vibrating feeder 5, a visible light camera 6, an inclined chute 9, a finished product hopper 14, and a waste hopper 15, primarily responsible for ore feeding, optical detection, and material guidance. The beneficiation shell 2 is obliquely positioned inside the main inlet shell 1, on which are installed an electrical control box 3, a near-infrared camera 10, a high-pressure pneumatic spray valve assembly 11, a material distribution guide baffle 12, and a flow channel 13, primarily responsible for multispectral imaging, intelligent decision-making, and precise separation.

[0049] In one exemplary embodiment, the ore color sorting method can be configured as an industrial computer system integrating machine vision and intelligent control. The ore to be inspected enters from the top feed hopper 4, is evenly arranged by the vibrating feeder 5, and falls. When the ore passes through the inspection area, the visible light camera 6 and the near-infrared camera 10 simultaneously trigger image acquisition. The industrial computer runs the method of this embodiment to process the images in real time and controls the high-pressure pneumatic spray valve assembly 11 to perform precise separation.

[0050] Example 1:

[0051] like Figure 1 As shown, this embodiment of the invention provides an ore color sorting method applicable to an ore color sorting device comprising an inlet main shell 1, a beneficiation shell 2, and an electrical control box 3. The method includes:

[0052] Step S1: The ore to be inspected enters from the feed hopper 4. The visible light camera 6 and the near-infrared camera 10 collect the multimodal optical features of the ore. Combined with the ambient light fluctuation interference in the detection area, the true light-sensing feature mapping value of the ore to be inspected is determined.

[0053] Step S2: Based on the friction characteristics of the ore material reflected by the light-sensing feature mapping value, combined with the ore morphology characteristics and the real-time operating status of the vibrating feeder 5, determine the ore feeding timing calibration coefficient.

[0054] Step S3: Based on the ore movement characteristics reflected by the material dropping timing calibration coefficient, and combined with the real-time operating status of the high-pressure pneumatic spray valve group 11 and the historical blow-off feedback coefficient, determine the spray valve pulse correction value corresponding to the impurities to be removed.

[0055] Step S4: Periodically calculate the overall deviation of the batch spray valve pulse correction value and reverse-calibrate the near-infrared weighting coefficient used in the calculation of the light-sensing feature mapping value.

[0056] Example 2:

[0057] like Figure 2 As shown in Example 1, this example provides a detailed explanation of the process for obtaining the photosensitive feature mapping value in step S1. The core of this example lies in utilizing multimodal visual perception to intelligently eliminate interference from ambient light in industrial settings, thereby enabling material identification by color.

[0058] It is understandable that single visible light imaging is highly susceptible to fluctuations in ambient light (such as light leakage from workshop doors and windows, and aging of supplementary lighting) and the effects of water stains and dust on the ore surface, leading to grayscale distortion. Near-infrared spectroscopy, however, can penetrate the surface and reflect the lattice vibration information of minerals within the ore, offering a natural advantage in material differentiation. Therefore, in this embodiment, near-infrared features are used as a check code for visible light features, and through coupling calculations, a photosensitive feature mapping value reflecting the essence of the ore is obtained.

[0059] In one exemplary embodiment, such as Figures 1 to 4 As shown, the process of obtaining the light-sensing feature mapping value includes:

[0060] Step S1-1: Obtain the average visible light gray value of the ore to be inspected, the near-infrared gray values ​​collected by at least two near-infrared cameras 10, the real-time ambient light brightness of the detection area, and the standard average gray value of the standard background plate under the near-infrared channel.

[0061] Specifically, when the ore to be inspected slides down the inclined chute 9 to the inspection window, the visible light camera 6 above captures a visible light image of the ore. After image segmentation and grayscale processing, the average grayscale value of the visible light in the ore area is obtained, denoted as L. vis Simultaneously, dual near-infrared cameras 10 (n channels in total, n=2 in this embodiment) installed on the ore dressing shell 2 collect the near-infrared grayscale values ​​of the ore, denoted as L. ir,i Furthermore, the brightness value of the detection area is acquired in real time using a photoelectric sensor, and its ratio to that under calibrated conditions is calculated to obtain the ambient light interference coefficient R. amb And read the standard average grayscale value L of the standard background plate in the near-infrared channel, which is pre-stored in the control box 3. bg,ir .

[0062] Step S1-2: Use the ratio of near-infrared discrimination to visible light discrimination as the near-infrared weighting coefficient k. ir We perform weighted fusion of multi-channel near-infrared grayscale values ​​to obtain near-infrared material characteristic coefficients.

[0063] In this embodiment, k ir The determination logic is crucial. For the mineral type to be sorted, the dispersion (discrimination) of the characteristic values ​​of concentrate and impurity samples is calculated under both visible and near-infrared channels, denoted as the near-infrared discrimination σ. ir and visible light discrimination σ vis The greater the discrimination, the less overlap in characteristics between the concentrate and impurities in that channel, and the stronger the identification ability. Then k ir =σ ir / σ vis .

[0064] For example, the near-infrared discrimination σ of a certain fluorite mine ir =35.2, visible light resolution σ vis =12.8, indicating that near-infrared light's ability to identify this mineral type is 2.75 times that of visible light, therefore k ir =2.75, to increase the weighting of near-infrared features. The near-infrared weighting coefficient k ir With the near-infrared grayscale values ​​L of each channel ir,i Multiplying and summing yields the weighted sum of near-infrared features. .

[0065] Where, k ir With L ir,i Multiplication is a hierarchical verification logic based on visible light features and corrected by near-infrared features: visible light imaging has high resolution and fast acquisition speed, so it first uses L... ir,i Obtain the apparent characteristics of the ore to complete rapid coarse screening; near-infrared spectroscopy is used as an auxiliary verification feature, multiplied by k. ir After adjusting its contribution magnitude, it is coupled with visible light features for calculation, and material-level corrections are made to the apparent features:

[0066] For example, when the visible light grayscale of a certain ore falls within the acceptable range, but its near-infrared characteristics indicate that it is an impurity, the weighted near-infrared normalization coefficient will deviate significantly from 1, and will be inconsistent with L. ir,i The result after multiplication will exceed the acceptable range, thus achieving accurate identification of heterogeneous samples of the same color.

[0067] Step S1-3: Calculate the average gray value L of visible light. vis As a basic identification benchmark, it is compared with the near-infrared material characteristic coefficient R. amb The system is coupled and then dynamically calibrated using the ambient light interference coefficient to obtain the light-sensing feature mapping value.

[0068] Specifically, the formula for calculating the light-sensing feature mapping value is as follows:

[0069] ;

[0070] In the formula: L real L represents the true light-sensing characteristics mapped onto the surface of the ore. vis L represents the average gray value of visible light. ir,i Here, k represents the near-infrared grayscale value of the ore region acquired by the i-th near-infrared supplementary lighting channel, n is the total number of near-infrared channels, and k is the near-infrared grayscale value of the ore region acquired by the i-th near-infrared supplementary lighting channel. ir L is the near-infrared weighting coefficient. bg,ir R represents the standard average grayscale value of the standard background plate under the near-infrared channel. amb The ambient light interference coefficient;

[0071] in, The calculation method is to integrate two types of features, which is essentially a weighted coupling of apparent color features and material composition features.

[0072] It should be noted that the reason for not valuing visible light is that if visible light were also weighted at the same time, it would lead to ambiguity in the coupling logic between the two types of features, increasing computational complexity.

[0073] also, Divide by It involves normalizing the near-infrared grayscale values ​​to obtain a material characteristic coefficient in the range of 0-1. When the ore is the target concentrate, the coefficient is close to 1; when it is an impurity, the coefficient will deviate significantly from 1.

[0074] The average gray value L of visible light vis Multiplying the material characteristic coefficient by the material is equivalent to performing a material verification on the apparent color in visible light: if the apparent color meets the concentrate threshold, but the near-infrared material coefficient deviates from the standard value, the result after multiplication will exceed the qualified range, thus identifying impurities of the same color but different quality; conversely, if the apparent color is darker but the material coefficient meets the standard, the result will still fall within the qualified range, avoiding the mistaken rejection of concentrates of different colors but the same quality.

[0075] Furthermore, utilizing Divide by R amb The core is to eliminate the interference of ambient light fluctuations in the industrial environment on the test results, based on the radiometric calibration principle of optical detection:

[0076] Industrial ore processing sites often suffer from problems such as light leakage from doors and windows, aging lighting fixtures, and dust obscuring light sources. This causes fluctuations in the actual light intensity of the testing area throughout the day, directly resulting in the camera capturing grayscale values ​​that are either too high or too low. Consequently, this leads to inconsistent test results for the same ore at different times. ambR is the ratio of "real-time ambient light brightness / device calibration standard brightness": when the ambient light becomes brighter. amb >1, dividing by this coefficient will correct the excessively high grayscale value to the level of standard brightness; when the ambient light becomes darker, R amb <1, the correction will calibrate the low grayscale value.

[0077] L real As the primary criterion for determining whether ore is qualified, it first obtains the qualification level (L) of the target ore type through a calibration dataset. real Interval: [L] min ,L max This interval represents all concentrate samples L from the calibration phase. real The 95% confidence interval for the value is determined by the following logic:

[0078] L min ≤L real ≤L max If so, it is determined to be a qualified concentrate;

[0079] L real <L min or L real >L max If it is, then it is determined to be an impurity to be removed.

[0080] It is understandable that the light-sensing feature mapping value L obtained through this embodiment... real This achieves a weighted coupling between apparent color features and material composition features. When the visible light grayscale L of a certain heterogeneous ore of the same color... vis If the value falls within the acceptable range, but its near-infrared material characteristic coefficient deviates from 1 due to material incompatibility, the product L real Those exceeding the acceptable range will be accurately identified as impurities. Conversely, for acceptable concentrates of different colors and similar properties, although L... vis It is slightly dark, but the near-infrared calibration item will compensate for it to ensure that it is not mistakenly rejected.

[0081] Example 3:

[0082] like Figure 3 As shown in Example 2, this example provides a detailed explanation of the process for obtaining the material dropping timing calibration coefficient in step S2. The core of this example lies in intelligently correcting the differences in material dropping time between different ores by inverting the physical motion characteristics of the ore through visual features.

[0083] It is understandable that the friction coefficients between ores of different materials and shapes and the inclined chute 9 vary significantly. In addition, the frequency fluctuations of the vibrating feeder 5 result in different actual flight times (drop times) for each ore from the detection point to the spray valve execution point. If a uniform theoretical time is used to trigger the spray valve, it will inevitably cause the problem of missing the spray when it should be sprayed and mistakenly spraying the spray when it should not be sprayed.

[0084] In one exemplary embodiment, such as Figure 1 As shown, the process of obtaining the material feeding timing calibration coefficient includes:

[0085] Step S2-1: Obtain the light sensitivity feature ratio of the light sensitivity feature mapping value of the ore to be inspected divided by the average light sensitivity feature value of the standard concentrate of the same type of ore.

[0086] Specifically, the average optical sensitivity characteristic value L of the standard concentrate for this mineral type is read from the memory of the electrical control box 3. std,avg Combined with the ore L to be tested obtained in steps S1-3 real Calculate the ratio of light-sensing features This ratio is strongly positively correlated with the coefficient of friction of the ore material: the closer the ratio is to 1, the more consistent the ore material is with the standard concentrate, and the frictional resistance is close to the calibration state; the greater the deviation of the ratio from 1, the greater the material difference and the more significant the change in sliding resistance.

[0087] Step S2-2: Obtain the second ratio of the estimated ore mass value divided by the product of the standard density value and the projected area.

[0088] In this embodiment, the projected area S of the ore to be inspected on the chute plane is first obtained using the image segmentation algorithm of the visible light camera 6. ore Combining the standard projected area and standard density value ρ of this mineral type std The ore quality estimate M is calculated by multiplying it by a preset shape factor (an empirical value between 0.7 and 1.3). est Then, calculate the inertia coefficient corresponding to the ore shape, which is the square root of the second ratio: The larger the inertia coefficient, the denser and more spherical the ore is, the less it is affected by air resistance and unevenness of the contact surface during sliding, and the faster it slides.

[0089] According to the sliding dynamics formula, the acceleration *a* is positively correlated with the square root of "mass / contact area" (the relationship between acceleration and the offsetting effect of frictional resistance is non-linear), while the sliding time *t* is related to... Since they are positively correlated, it is necessary to take the square root of the mass ratio to ensure dimensional matching of the physical quantities (the final correction factor is a dimensionless value).

[0090] Step S2-3: Obtain the ratio f of the real-time frequency of the vibrating feeder 5 to the calibrated frequency. vid

[0091] Step S2-4: Use the ratio of light-sensing characteristics as a reference for frictional characteristic correction, and multiply and fuse it with the theoretical slip time, as well as the ratio of inertia coefficient and vibration frequency corresponding to the ore morphology, to obtain the material drop timing calibration coefficient.

[0092] Thus, the material feeding timing calibration coefficient has been obtained, and the specific calculation formula is as follows:

[0093] ;

[0094] In the formula:

[0095] T cal T represents the calibration coefficient for the calibrated material feeding timing. theo L is the theoretical sliding time for ore of the same particle size, calculated based on the chute inclination angle and preset velocity. std,avg M represents the average optical sensitivity characteristic value of standard concentrates of the same mineral type, and M represents the mean of the previously calibrated dataset. est ρ is an estimated value for the ore mass. std S represents the standard density value for the corresponding mineral type. ore f is the projected area of ​​the ore on the plane of the sluice. vid The ratio of the vibration frequencies of the vibrating feeder 5 reflects the influence of the feeding vibration on the slip velocity.

[0096] Among them, the frequency ratio f of the vibrating feeder vid Used to correct fluctuations in the overall conveying speed of a vibrating chute. The working principle of the vibrating feeder 6 is to cause the ore to move forward in a jumping motion through high-frequency micro-vibration. The higher the vibration frequency, the faster the overall conveying speed of the ore and the shorter the slippage time. vid f is the ratio of the real-time vibration frequency to the calibrated frequency: when the actual vibration frequency is higher than the calibrated value, vid >1, division can shorten the theoretical material feeding time; when the frequency is lower than the calibrated value, f vid <1, division will extend the theoretical material feeding time to match the positive correlation between vibration frequency and conveying speed. Frequency data is obtained directly from the built-in frequency converter of the vibrating feeder 6, without the need for additional sensors.

[0097] It is understandable that this embodiment uses visual feature L real Indirectly reflects the coefficient of friction, through the morphological parameter S ore and M est It reflects the magnitude of inertia and, combined with real-time correction of vibration frequency, achieves personalized and accurate prediction of the drop time for each ore piece without the need for additional physical sensors. The final actual trigger time of the spray valve is: T trigger =T detect +T theo ×T cal , among which, T trigger The final absolute time for the injection valve trigger is sent directly to the injection valve controller, T. detect This is the absolute timestamp of when the ore is detected by the camera, directly given by the camera's frame synchronization signal.

[0098] Example 4:

[0099] like Figure 2 As shown in Example 3, this example provides a detailed explanation of the process for obtaining the injection valve pulse correction value in step S3. The core of this example lies in combining the actuator state with historical blow-off effects to achieve adaptive feedforward-feedback control of the injection valve action.

[0100] Understandably, the high-pressure pneumatic spray valve assembly 11 will experience pressure decay and valve core wear after long-term operation. At the same time, ores of different densities and velocities have drastically different requirements for spray momentum. A fixed-width jet pulse cannot adapt to these dynamic changes.

[0101] In one exemplary embodiment, such as Figure 1 , Figure 3 and Figure 5 As shown, the process of obtaining the injection valve pulse correction value includes:

[0102] Step S3-1: Obtain the speed correction coefficient, i.e., the material feeding timing calibration coefficient T. cal Divide by the theoretical material feeding time T theo ratio This coefficient reflects the actual speed of ore movement: when the ratio is greater than 1, the ore slides more slowly, and the required blowing time can be appropriately shortened; when the ratio is less than 1, the ore slides faster, and the blowing time needs to be extended to compensate for the impulse.

[0103] Step S3-2: Obtain the historical blow-off feedback coefficient. Using a photoelectric sensor installed at the inlet of waste hopper 15, collect the actual trajectory offset D of the previous historical ore with similar optical characteristics and particle size to the current ore. offset,last Obtain the standard average trajectory offset D of this type of ore under a standard pulse. offset,std,avg Calculate the feedback coefficient. Specifically, if the ratio is greater than 1, it indicates that the previous blowing force was too strong, which may have caused misblowing or wasted airflow, and the pulse width should be reduced this time; if the ratio is less than 1, it indicates that the previous blowing force was insufficient and failed to blow impurities into the waste hopper 15, and the pulse width should be increased this time. This forms a feedforward-feedback closed-loop control based on historical blowing effects.

[0104] Step S3-3: Obtain the injection valve efficiency coefficient η valve This is the ratio of the actual air pressure of the spray valve to its rated air pressure. This coefficient is used to compensate for aging of the spray valve and fluctuations in air circuit pressure.

[0105] Step S3-4: Standard jet pulse width P for ore of the same particle size base The pulse correction value for the injection valve is obtained by multiplying and fusing the above three coefficients:

[0106] ;

[0107] In the formula:

[0108] P adj The corrected pulse value for the injection valve directly controls the injection valve opening duration, P. base The standard jet pulse width for ore with the corresponding particle size, factory calibrated value, D offset,last D represents the actual trajectory offset after the last instance of the same type of ore being blown away. offset,std,avg η represents the standard average trajectory offset of the corresponding ore under a standard pulse. valve This is the efficiency coefficient of the injection valve.

[0109] Where, η valve This is the ratio of the actual air pressure to the rated air pressure of the injection valve, correcting for airflow force deviations caused by valve aging and air circuit pressure fluctuations: the lower the air pressure, the longer the pulse width is required to provide sufficient impulse. valve When the actual air pressure is 80% of the rated pressure, the pulse width needs to be multiplied by 1.25 to compensate for insufficient air pressure.

[0110] At this point, the final injection valve pulse correction value P was obtained. adj When the current ore is determined to contain impurities to be removed, the industrial computer uses T... trigger At a given time, a time interval P is sent to the solenoid valve of the corresponding channel in the high-pressure pneumatic injection valve assembly 11. adj A high-level pulse signal drives the spray valve to open.

[0111] Understandably, this embodiment matches material movement through speed correction, corrects blowing force deviation through historical feedback, and compensates for actuator attenuation through efficiency coefficient. The combination of these three factors enables precise and dynamic adjustment of the spray valve pulse width, ensuring that impurities are accurately blown over the material distribution guide baffle 12 and smoothly fall into the waste hopper 15 through the guiding action of the flow plate 13, while qualified concentrate naturally slides down the inclined chute 9 to the finished product hopper 14.

[0112] Example 5:

[0113] like Figure 2 As shown in Example 4, this example provides a detailed explanation of the process of reverse calibration of the near-infrared weighting coefficients in step S4. The core of this example lies in establishing a closed-loop self-iterative mechanism from the execution end (spray valve) to the sensing end (vision) to achieve intelligent operation and maintenance of the equipment.

[0114] It is understandable that as the equipment operates for a long time, the detection window may accumulate dust, and the supplementary light may weaken, leading to changes in the light-sensing feature mapping value L. real This results in a systematic deviation. This deviation will ultimately manifest as the injection valve pulse correction value P.adj The statistical average value deviates from the standard value consistently and over a long period. This can be achieved by monitoring P. adj The macro trend can be used to infer and correct the front-end perception parameters.

[0115] In one exemplary embodiment, the process of reverse calibration of the near-infrared weighting coefficients includes:

[0116] Step S4-1: After completing a preset number of sorting operations (e.g., 1000 times), the processor in the control box 3 calculates the correction value P of all spray valve pulses used to remove impurities in that batch. adj And calculate its average value P. adj,avg .

[0117] Step S4-2: Calculate the average value P adj,avg The standard pulse value P calibrated for this mineral type std The deviation rate.

[0118] Step S4-3: When the deviation rate exceeds a preset threshold (e.g., 10%), the system determines that a system deviation has occurred in the front-end light-sensing feature recognition. At this time, the near-infrared weighting coefficient k is adjusted in reverse according to the deviation ratio. ir The calculation logic for correcting the light-sensing feature mapping value is then implemented.

[0119] Specifically, if P adj,avg >P std This indicates that the actuator requires a larger pulse to complete the sorting. This usually means that the front-end identification mistakenly identifies ore that should be impurities as concentrate (missed blowing), or misidentifies concentrate as impurities (false blowing), causing the spray valve to over-act. To correct this bias, update the near-infrared weighting coefficients:

[0120] .

[0121] This completes the online reverse calibration of the near-infrared weighting coefficients. If P adj,avg If it is too large, then Less than 1, k ir,new This will decrease, thereby reducing the weight of near-infrared features in the calculation of light-sensing feature mapping values ​​and adjusting the sensitivity of material recognition.

[0122] Understandably, this embodiment uses the overall deviation of macroscopic statistical microscopic execution parameters (spray valve pulses) to deduce the systematic drift of the sensing layer and automatically correct the core weight coefficients. This enables the color sorting equipment to have self-sensing and self-calibration capabilities, eliminating the need for frequent manual shutdowns for calibration and significantly improving the equipment's continuous operation stability and long-term ability to maintain sorting accuracy in complex industrial environments.

[0123] Example 6:

[0124] Based on all the above embodiments, this embodiment summarizes the complete workflow of the ore color sorting method to clearly illustrate how the components work together.

[0125] like Figures 1 to 5 As shown, the workflow of the ore color sorting method of the present invention is as follows:

[0126] The ore to be inspected enters through the feed hopper 4, is evenly spread by the vibrating feeder 5, and then falls.

[0127] When the ore passes through the detection area, the visible light camera 6 and the near-infrared camera 10 are synchronously triggered to acquire multimodal images. The industrial computer runs the method described in Example 2, combining ambient light information to calculate the photosensitive feature mapping value L of the ore to be inspected. real The system is based on L real Value determination:

[0128] Scenario 1 (Qualified Concentrate): If L real If the ore falls within the preset acceptable range, it is determined to be concentrate. The high-pressure pneumatic spray valve assembly 11 does not operate, and the ore slides naturally down the inclined chute 9, eventually entering the finished product hopper 14 for collection.

[0129] Scenario 2 (Impurities to be removed): If L real If the value exceeds the preset acceptable range, it is determined to be an impurity. In this case, the system executes the method of Example 3, according to L... real Based on ore morphology, vibrating feeder frequency, etc., the material feeding timing calibration coefficient T is calculated. cal And determine the precise valve trigger time T. trigger Simultaneously, by implementing the method of Example 4, and combining historical blow-off feedback with the real-time status of the injection valve, the optimal injection valve pulse correction value P is calculated. adj When T is reached trigger At a certain moment, the industrial computer sends a time duration P to the corresponding channel of the high-pressure pneumatic injection valve assembly 11. adj When the signal level is received, the spray valve opens, and the high-pressure airflow precisely blows the impurities away from their original trajectory, causing them to pass over the material distribution guide baffle 12 and, driven and guided by the flow plate 13, finally fall into the waste hopper 15.

[0130] During continuous operation of the equipment, the method of Example 5 is executed in parallel in the system background. Every preset number of times (e.g., 1000 valve actions), the processor counts the P values ​​for that batch. adj The mean value is calculated and compared with the standard value. If a systematic deviation exceeds a threshold, the near-infrared weighting coefficient k is automatically adjusted in reverse. ir This corrects the calculation model of the light-sensing feature mapping value in Example 2, thereby realizing intelligent closed-loop self-iteration throughout the entire process.

[0131] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for color sorting of ores, characterized in that, A color sorting device for ores comprising a main inlet shell (1), a beneficiation shell (2), and an electrical control box (3), the method comprising: The ore to be inspected enters from the feed hopper (4), and the multimodal optical features of the ore are collected by the visible light camera (6) and the near-infrared camera (10). Combined with the ambient light fluctuation interference in the detection area, the true light-sensing feature mapping value of the ore to be inspected is determined. The light-sensing feature mapping value is used to distinguish the difference between the apparent color and the internal material properties of the ore. Based on the friction characteristics of the ore material reflected by the light-sensing feature mapping value, combined with the ore morphology characteristics and the real-time operating status of the vibrating feeder (5), the ore dropping time calibration coefficient is determined. The dropping time calibration coefficient is used to characterize the degree of deviation between the actual sliding speed and the theoretical speed of a single ore. Based on the ore movement characteristics reflected by the material dropping timing calibration coefficient, and combined with the real-time operating status of the high-pressure pneumatic spray valve group (11) and the historical blow-off feedback coefficient, the spray valve pulse correction value corresponding to the impurities to be removed is determined. The spray valve pulse correction value is used to drive the high-pressure pneumatic spray valve group (11) to accurately separate impurities. Periodically analyze the overall deviation of the batch spray valve pulse correction value and reverse-calibrate the near-infrared weighting coefficient used in the calculation of the light-sensing feature mapping value.

2. The ore color sorting method according to claim 1, characterized in that, The process of obtaining the light-sensing feature mapping value includes: The average visible light gray value of the ore to be inspected, the near-infrared gray values ​​collected by at least two of the near-infrared cameras (10), the real-time ambient light brightness of the detection area and the standard average gray value of the standard background plate under the near-infrared channel are obtained. The ratio of near-infrared resolution to visible light resolution is used as the near-infrared weighting coefficient to perform weighted fusion of multi-channel near-infrared grayscale values, thereby obtaining the near-infrared material characteristic coefficient. The average gray value of visible light is used as the basic recognition benchmark, coupled with the near-infrared material characteristic coefficient, and then dynamically calibrated through the ambient light interference coefficient to obtain the light-sensing feature mapping value.

3. The ore color sorting method according to claim 2, characterized in that, The process of obtaining the near-infrared material characteristic coefficients includes: The near-infrared gray values ​​of the ore to be inspected are obtained by the upper and lower dual near-infrared cameras (10), and then multiplied by the near-infrared weighting coefficients respectively and summed to obtain the near-infrared feature weighted sum. The weighted sum of the near-infrared features is multiplied by the total number of near-infrared channels and the standard average gray value. The first ratio is then normalized to obtain the near-infrared material feature coefficient in the range of 0-1. The closer the near-infrared material feature coefficient is to 1, the higher the matching degree between the ore material and the standard concentrate.

4. The ore color sorting method according to claim 3, characterized in that, The process of obtaining the material feeding timing calibration coefficient includes: The light-sensing feature ratio obtained by dividing the light-sensing feature mapping value of the ore to be inspected by the average light-sensing feature value of the standard concentrate of the same type, the second ratio obtained by dividing the estimated ore quality value by the product of the standard density value and the projected area, and the vibration frequency ratio of the real-time frequency of the vibrating feeder (5) to the calibrated frequency; The ratio of light-sensing characteristics is used as a reference for friction characteristic correction. It is multiplied and fused with the theoretical slip time and the ratio of inertia coefficient and vibration frequency corresponding to the ore morphology to obtain the material drop timing calibration coefficient. The actual ratio of light-sensing characteristics has a positive adjustment effect on the degree of friction characteristic correction.

5. The ore color sorting method according to claim 4, characterized in that, The process of obtaining the second ratio includes: The projected area of ​​the ore to be inspected is obtained by image segmentation of the visible light camera (6), and the ore quality estimate is calculated by combining the standard density value and shape factor of the ore type. The square root of the product of the estimated ore mass, the standard density, and the projected area is taken to obtain the inertia coefficient corresponding to the ore shape. The larger the inertia coefficient, the higher the density of the ore and the smaller the sliding resistance.

6. The ore color sorting method according to claim 1, characterized in that, The process of obtaining the injection valve pulse correction value includes: The following parameters are used: the material drop timing calibration coefficient divided by the theoretical material drop time speed correction coefficient; the ratio of the actual trajectory offset of the previous light-sensing feature and similar particle size ore to the standard average trajectory offset as the historical blow-off feedback coefficient; and the ratio of the actual air pressure of the spray valve to the rated air pressure as the spray valve efficiency coefficient. The standard jet pulse width of ore of the same particle size is multiplied and fused with the velocity correction coefficient, the historical blow-off feedback coefficient, and the jet valve efficiency coefficient to obtain the jet valve pulse correction value.

7. The ore color sorting method according to claim 6, characterized in that, The process of obtaining the historical blow-off feedback coefficient includes: The photoelectric sensor at the inlet of the waste hopper (15) collects the actual trajectory offset of the previous ore with similar light-sensing characteristics and particle size to the current ore. The historical blow-off feedback coefficient is obtained by comparing the actual trajectory offset with the standard average trajectory offset of the ore under the standard pulse. When the historical blow-off feedback coefficient is greater than 1, it indicates that the previous blow-off force was insufficient and the pulse width of this time needs to be increased.

8. The ore color sorting method according to claim 1, characterized in that, The process of reverse calibration of near-infrared weighting coefficients includes: After each preset number of sorting actions are completed, the average value of all the spray valve pulse correction values ​​in that batch is calculated, and the deviation rate between the average value and the calibrated standard pulse value is calculated. When the deviation rate exceeds the preset threshold, the near-infrared weighting coefficient is adjusted in reverse according to the deviation ratio to correct the calculation logic of the light-sensing feature mapping value.

9. The ore color sorting method according to claim 1, characterized in that, The method further includes: When the light-sensing feature mapping value of the ore to be inspected falls within the preset qualified range, it is judged as qualified concentrate, and the ore naturally slides down the inclined chute (9) to the finished product hopper (14). When the light-sensing feature mapping value of the ore to be inspected exceeds the preset qualified range, it is determined to be an impurity to be removed. The trigger time of the spray valve is determined according to the material falling time calibration coefficient. The spray valve of the corresponding channel of the high-pressure pneumatic spray valve group (11) is driven to open according to the spray valve pulse correction value, and the impurities are blown over the material distribution guide baffle (12). The flow through the flow plate (13) drives the inclined chute (9) to rotate and is sent into the waste hopper (15).

Images