Iron ore component detection method and system based on machine learning

Abstract

The invention discloses an iron ore composition detection method and system based on machine learning, and relates to the technical field of data processing.The method comprises the steps that iron ore feature information of target iron ore to be detected is collected, and a plurality of iron ore samples are collected; performing drying and grinding parameter optimization according to the iron ore characteristic information, drying and grinding the plurality of iron ore samples, and collecting the particle size and mass in each drying and grinding process to obtain a plurality of particle size sequence sets and a plurality of mass sequences; preparing a plurality of detection samples, and performing sample quality analysis according to the optimized drying and grinding parameters, the plurality of particle size sequence sets and the plurality of quality sequences to obtain a plurality of sample quality coefficients; and carrying out X-ray fluorescence spectrum detection on the plurality of detection samples to obtain a plurality of spectrum detection results, and carrying out component identification on the plurality of spectrum detection results according to the plurality of sample mass coefficients to obtain an iron ore component detection result. The technical problem of poor iron ore component detection accuracy in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for detecting iron ore composition based on machine learning. Background Art

[0002] Accurate determination of iron ore composition is crucial for resource assessment, smelting process optimization, and quality control. Current mainstream methods typically rely on X-ray fluorescence spectroscopy for compositional analysis, but this technique places stringent demands on sample preparation, and the accuracy of its results is highly dependent on the uniformity of the sample's particle size, moisture content, and physical stability. Existing techniques often employ fixed or empirical drying and grinding parameters during sample preparation, ignoring the significant differences in inherent properties such as hardness and porosity among different iron ore samples. This results in uneven particle size distribution and large fluctuations in residual moisture in the prepared samples, directly impacting the reliability of subsequent spectral analysis. Furthermore, traditional methods often rely on a single model to interpret spectral results during the compositional analysis phase, failing to effectively quantify and utilize the quality information inherent in the sample preparation process. This results in two major drawbacks: First, mismatched sample preparation parameters can easily introduce systematic errors, causing test results to deviate from the true composition; second, the composition identification process is disconnected from the actual sample preparation quality, and the inherent characteristics of the sample can affect the reliability of the test results. This ultimately leads to a disconnect between the iron ore sample preparation process and the compositional analysis phase, resulting in inaccurate iron ore composition detection. Summary of the Invention

[0003] The present application provides a method and system for detecting iron ore composition based on machine learning, which is used to solve the technical problems in the prior art of the disconnection between the iron ore sample preparation process and the component analysis link, and the insufficient accuracy of iron ore composition detection.

[0004] In view of the above problems, the present application provides a method and system for iron ore composition detection based on machine learning.

[0005] In a first aspect, the present application provides a method for detecting iron ore composition based on machine learning, the method comprising: Collecting iron ore characteristic information of the target iron ore to be tested for composition, and collecting multiple iron ore samples; Optimizing drying and grinding parameters based on the iron ore characteristic information to obtain optimized drying and grinding parameters, drying and grinding the multiple iron ore samples, and collecting particle size and mass during each round of drying and grinding to obtain multiple particle size sequence sets and multiple mass sequences; Based on the dried and ground particles, multiple test samples are prepared, and sample quality analysis is performed according to the optimized drying and grinding parameters, multiple particle size sequence sets, and multiple mass sequences to obtain multiple sample quality coefficients; X-ray fluorescence spectroscopy is performed on multiple test samples to obtain multiple spectral detection results, and components of the multiple spectral detection results are identified according to the multiple sample mass coefficients to obtain iron ore composition detection results.

[0006] In a second aspect, the present application provides an iron ore composition detection system based on machine learning, comprising: An information collection module is used to collect iron ore characteristic information of a target iron ore to be tested for composition, and to collect multiple iron ore samples; a grinding optimization module, configured to optimize drying and grinding parameters based on the iron ore characteristic information to obtain optimized drying and grinding parameters, dry and grind the multiple iron ore samples, and collect the particle size and mass during each round of drying and grinding to obtain multiple particle size sequence sets and multiple mass sequences; The mass analysis module is used to prepare multiple test samples based on the dried and ground particles, and to perform sample mass analysis based on optimized drying and grinding parameters, multiple particle size sequence sets, and multiple mass sequences to obtain multiple sample quality coefficients; The component detection module is used to perform X-ray fluorescence spectrum detection on multiple test samples to obtain multiple spectrum detection results, perform component identification on the multiple spectrum detection results according to the multiple sample mass coefficients, and obtain iron ore component detection results.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application proposes a method and system for iron ore composition detection based on machine learning. By integrating the inherent characteristic information of iron ore, it guides the optimization of the sample preparation process, and uses the dynamic monitoring data of the entire sample preparation process to build a quality assessment system, and then weights and integrates the output of multiple component identification models, significantly improving the accuracy, reliability and overall efficiency of iron ore composition detection. Compared with traditional methods that usually adopt fixed or empirical parameter settings for the sample pretreatment link, and subsequent component detection mostly relies on a single test, it is difficult to effectively deal with the inconsistency of the preparation process caused by differences in characteristics such as mineral hardness and porosity of iron ore samples, as well as the detection errors introduced, the technical solution provided by this application significantly overcomes the above limitations. First, based on the specific characteristic information of the target iron ore, an iterative optimization mechanism driven by a machine learning model is used to dynamically generate the optimal combination of drying and grinding parameters for each sample, ensuring that the sample is processed to the physical state most suitable for subsequent X-ray fluorescence spectroscopy detection, thereby improving the representativeness and uniformity of the sample from the source. Secondly, in the process of performing dry grinding according to the optimized parameters, the key physical indicators after each round of processing are continuously collected and recorded in real time for subsequent in-depth analysis. Based on these dynamic sequence data, the scheme calculates and integrates multiple coefficients that reflect the stability of particle size changes, the proximity of the final particle size to the target standard, and the drying effect, ultimately forming a comprehensive sample quality coefficient, which quantifies the degree of deviation of the physical properties of each prepared test sample from the ideal standard, providing a key basis for the reliability assessment of subsequent spectral results. Finally, in the component identification stage, the scheme uses the above-mentioned sample quality coefficient to quantify the confidence level of each test sample result, etc., and comprehensively forms a component identification coefficient for each test result. This dynamic and adaptive model mechanism can effectively utilize the complementary advantages of the model to offset the potential bias or error of a single model based on the actual situation of the sample's own preparation quality and processing efficiency, thereby obtaining iron ore composition detection results that are closer to the true value overall.

[0008] This application achieves the technical effect of significantly improving the accuracy of iron ore composition detection and the robustness of results, while effectively optimizing sample preparation efficiency and resource utilization, ultimately achieving high-reliability iron ore composition detection and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic diagram of a process for detecting iron ore composition based on machine learning provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of an iron ore composition detection system based on machine learning provided in an embodiment of the present application; In the accompanying drawings, the components represented by the reference numerals are described as follows: Information collection module 100, grinding optimization module 200, quality analysis module 300, component detection module 400. DETAILED DESCRIPTION

[0011] This application provides a machine learning-based iron ore composition detection method, system, intelligent terminal and storage medium to solve the technical problems in the existing technology that the iron ore sample preparation process is disconnected from the component analysis link and the iron ore composition detection accuracy is insufficient.

[0012] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0013] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0014] Example 1, as Figure 1 As shown, the present application provides a method for detecting iron ore composition based on machine learning, wherein the method includes: S10: Collecting iron ore characteristic information of the target iron ore to be tested for composition, and collecting multiple iron ore samples.

[0015] Current iron ore composition testing suffers from fundamental flaws in sample selection. Because the inherent differences in the target iron ore's mineral properties, such as hardness and porosity, are not considered, a uniform method of sample collection is used. This leads to poor consistency in subsequent sample preparation, making it impossible to establish a comparable data foundation and ultimately causing systematic deviations in test results.

[0016] Step S10 in the method provided in the embodiment of the present application includes: Collecting iron ore characteristic information of the target iron ore to be tested for composition, wherein the iron ore characteristic information includes the type of iron ore, mineral hardness, and porosity; A plurality of iron ore samples are collected from the target iron ore.

[0017] In the examples of this application, various methods were used to collect characteristic information about the target iron ore for composition analysis. Physical properties such as color, streak, luster, and magnetism were observed to preliminarily determine the type of iron ore. For example, hematite is typically reddish-brown, while magnetite is iron-black. Mineral hardness was measured using a Mohs hardness tester, and mineral porosity was measured using a porosity meter.

[0018] For different collection targets, the collection method can be optimized, for example, controlling the hammering force when collecting hematite and increasing the hammering force when collecting magnetite. Collect multiple iron ore samples from the iron ore.

[0019] By specifically collecting characteristic information such as the mineral hardness and porosity of the target iron ore, and scientifically selecting multiple representative samples based on this information, a data foundation is provided for subsequent optimization, allowing the drying and grinding process to be differentiated for samples with different physical properties, laying the foundation for constructing highly consistent test samples and significantly improving the starting point reliability of subsequent component analysis.

[0020] S20: Optimizing drying and grinding parameters based on the iron ore characteristic information to obtain optimized drying and grinding parameters, drying and grinding the multiple iron ore samples, and collecting the particle size and mass during each round of drying and grinding to obtain multiple particle size sequence sets and multiple mass sequences.

[0021] Traditional drying and grinding processes use fixed parameters (such as drying time and grinding intensity) and are unable to adapt to the varying physical properties of different iron ore samples. Differences in mineral hardness lead to a more discrete particle size distribution after fixed-intensity grinding, while variations in porosity result in varying residual moisture levels even at the same drying time. The lack of a dynamic parameter adjustment mechanism can lead to increased fine powder loss from overdrying, while insufficient drying can lead to particle agglomeration, ultimately causing the sample's particle size distribution to deviate from testing requirements. Furthermore, abnormal fluctuations in particle size and mass during multiple rounds of processing cannot be effectively captured and corrected.

[0022] Step S20 in the method provided in the embodiment of the present application includes: Randomly generating first dry grinding parameters for dry grinding the iron ore sample, wherein the first dry grinding parameters include the number and time of repeated grinding and drying of the iron ore sample; Analyzing and processing the iron ore characteristic information and the first drying and grinding parameter to obtain a first sample processing adaptability; The first sample processing adaptability is obtained by analyzing and processing the iron ore characteristic information and the first drying and grinding parameter, including: Obtaining a first sample processing time of the first drying and grinding parameter, and calculating a ratio of a preset processing time to the first sample processing time to obtain a first time processing fitness; Inputting the drying and grinding parameters and the iron ore characteristic information into an iron ore sample processing predictor, and predicting and outputting a first predicted particle size and a first quality, wherein the iron ore sample processing predictor is trained using a sample drying and grinding parameter set, a sample iron ore characteristic information set, a sample predicted particle size set, and a sample moisture content set; Obtaining the standard particle size and standard mass of particles; Calculating the similarity between the first predicted granularity and the standard granularity, and calculating the ratio of the standard quality to the first quality, and combining the first time processing fitness to calculate and obtain a first sample processing fitness; Iterative optimization of drying and grinding parameters is performed to obtain optimized drying and grinding parameters with maximum sample processing adaptability; Drying and grinding the plurality of iron ore samples using the optimized drying and grinding parameters; The particle size and mass are collected during each round of dry grinding to obtain multiple particle size sequence sets and multiple mass sequences.

[0023] In an embodiment of the present application, a random number generator is used to randomly generate the number and time of repeated grinding and drying of the iron ore sample. For example, the number of grinding times ranges from 3 to 8 times, and the drying time ranges from 5 to 20 minutes, and the random generation is performed within this range.

[0024] Obtain a first sample processing time for the first dry grinding parameter, and calculate the ratio of the preset processing time to the first sample processing time to obtain a first time processing fitness. The preset processing time is a generally common sample processing time, and is illustratively set to 20 minutes. The first time processing fitness = preset processing time ÷ first sample processing time. For example, if the first sample processing time is 25 minutes, the first time processing fitness = 20 ÷ 25 = 0.8.

[0025] A neural network was used to construct an iron ore sample processing predictor. The predictor employs a four-layer architecture: an input layer with five nodes, receiving two drying and grinding parameters and three iron ore characteristics. The hidden layer consists of two layers, each with 16 nodes, activated using the Reluctant Unit (ReLU) function. The output layer uses two nodes to output predicted particle size (using linear activation) and predicted moisture content (using a Sigmoid function). Supervised training of the predictor was performed using a set of sample drying and grinding parameters, a set of sample iron ore characteristics, a set of predicted particle size, and a set of sample moisture content. These sets were obtained by collecting and statistically analyzing grinding parameters, iron ore characteristics, particle size, and moisture content from historical grinding processes. Through supervised training, model parameters were continuously adjusted until the model converged. Training of the predictor was considered complete when the accuracy of the first predicted particle size and first mass output reached over 90% based on the input of grinding parameters and iron ore characteristics. The first predicted particle size refers to the predicted diameter of the iron ore sample particles after processing, and the unit is micrometer. The first mass refers to the mass of the iron ore sample that tends to be stable after grinding and drying, and the unit is gram.

[0026] According to the detection standards required for the detection, the standard particle size and standard mass of the particles are obtained. For example, the standard particle size is 75 microns and the standard mass is 5 grams.

[0027] Calculate the similarity between the first predicted particle size and the standard particle size: Similarity = 1 - |first predicted particle size - standard particle size| ÷ standard particle size. For example, if the first predicted particle size is 60, then Similarity = 1 - |60 - 75| ÷ 75 = 0.8. Calculate the ratio of the standard mass to the first mass: Ratio = standard mass ÷ first mass. For example, if the first mass is 4g, then Ratio = 5 ÷ 4 = 1.25. Combined with the first-time processing fitness, calculate the first sample processing fitness. Exemplarily, First Sample Fitness = (First-time Processing Fitness + Similarity + Ratio of Standard Mass to First Mass) ÷ 3. For example, if the first-time processing fitness is 0.8, the similarity is 0.8, and the ratio of Standard Mass to First Mass is 1.25, then First Sample Fitness = (0.8 + 0.8 + 1.25) ÷ 3 = 0.95. A higher sample fitness indicates better quality of the processed sample, facilitating subsequent measurements.

[0028] The particle swarm optimization algorithm was used to iteratively optimize the dry grinding parameters until the optimal dry grinding parameters with the maximum sample processing fitness were obtained. Particle swarm optimization (PSO) is an evolutionary computing technique that uses a massless particle to simulate birds in a flock, searching for the optimal solution in the search space.

[0029] Several iron ore samples were dried and ground using the optimized drying and grinding parameters that were optimized to have the maximum sample processing suitability.

[0030] During the drying and grinding process, the particle size and mass of each round of drying and grinding are collected and statistically analyzed to obtain multiple particle size sequence sets and multiple mass sequences.

[0031] In the embodiment of the present application, based on the characteristic information of the iron ore, the drying and grinding parameters are iteratively optimized through a machine learning model to achieve a closed-loop regulation of dynamic drying-grinding-monitoring. The particle size and mass sequence are collected in real time during each round of processing to form a process trajectory database. This adjustment mechanism can achieve stronger grinding intensity for high-hardness samples and extend the drying time for high-porosity samples, ensuring that all samples eventually approach the target particle size and moisture content. At the same time, the coupled optimization of process parameters and physical property data suppresses the fluctuations in sample preparation quality caused by parameter solidification in traditional methods.

[0032] S30: Based on the dried and ground particles, multiple test samples are prepared, and sample quality analysis is performed according to the optimized drying and grinding parameters, multiple particle size sequence sets, and multiple mass sequences to obtain multiple sample quality coefficients.

[0033] Existing techniques rely solely on final particle size to judge sample quality, ignoring key dynamic indicators during the preparation process: the stability of particle size changes during the grinding stage and the rationality of mass loss during the drying stage. This can lead to two types of risks: apparently compliant samples may harbor compositional denaturation caused by localized overheating or contamination introduced by excessive grinding. Traditional single quality indicators are unable to quantify these hidden defects, resulting in low-quality samples entering the testing process and affecting test results.

[0034] Step S30 in the method provided in the embodiment of the present application includes: Based on the dried and ground particles, multiple test samples were prepared; Based on the multiple particle size sequence sets, multiple particle size stability coefficients are calculated, and based on the last particle size in the multiple particle size sequences and the standard particle size, multiple basic particle size quality coefficients are calculated; Using the plurality of particle size stability coefficients respectively, a plurality of basic particle size quality coefficients are corrected and calculated to obtain a plurality of particle size quality coefficients; According to multiple quality sequences, multiple drying quality coefficients are calculated; A plurality of sample mass coefficients are calculated based on a plurality of particle size mass coefficients and a plurality of drying mass coefficients.

[0035] In the examples of the present application, a plurality of test samples were prepared based on the iron ore particles that had been dried and ground.

[0036] Based on multiple particle size sequence sets, multiple particle size stability coefficients are calculated: 1 - (maximum particle size in the sequence - minimum particle size in the sequence) ÷ maximum particle size. For example, if the maximum particle size is 90 and the minimum particle size is 60, the particle size stability coefficient is 1 - (90 - 60) ÷ 90 = 0.67. Based on the last particle size in the sequence and the standard particle size, multiple basic particle size quality coefficients are calculated: basic particle size quality coefficient is 1 - |last particle size in the sequence - standard particle size| ÷ standard particle size. For example, if the last particle size in the sequence is 60 and the standard particle size is 75, the basic particle size quality coefficient is 1 - |60 - 75| ÷ 75 = 0.8. The particle size quality coefficient is calculated as particle size stability coefficient × basic particle size quality coefficient. For example, if the particle size stability coefficient is 0.67 and the basic particle size quality coefficient is 0.8, the particle size quality coefficient is 0.67 × 0.8 = 0.536. The higher the particle size quality coefficient, the more uniform the particle size, which is beneficial for subsequent measurements.

[0037] Based on multiple mass sequences, multiple drying quality coefficients are calculated. The drying quality coefficient = 1 - |Last mass in the sequence - Standard mass| ÷ Standard mass. For example, if the last mass in the sequence is 7.5g and the standard mass is 5g, the drying quality coefficient = 1 - |7.5 - 5| ÷ 5 = 0.5. A higher drying quality coefficient indicates better drying results and more stable quality, which facilitates subsequent measurements.

[0038] Based on multiple particle size mass coefficients and multiple drying mass coefficients, multiple sample mass coefficients are calculated: Sample mass coefficient = (particle size mass coefficient + drying mass coefficient) ÷ 2 = 0.518. A higher sample mass coefficient indicates more stable sample quality, which is more conducive to subsequent measurements.

[0039] A multidimensional quality assessment model is constructed by integrating optimization parameters, particle size sequence sets, and mass sequences. A stability coefficient is calculated by analyzing particle size variations, and a basic quality coefficient is generated based on the difference between the final particle size and the target value. A drying uniformity coefficient is also calculated based on the mass sequence, ultimately resulting in a comprehensive sample quality coefficient. This coefficient accurately quantifies the risk points of sample property evolution during the preparation process, making implicit defects explicit and providing a critical quality anchor for subsequent component identification.

[0040] S40: performing X-ray fluorescence spectrum detection on a plurality of test samples to obtain a plurality of spectrum detection results, performing component identification on the plurality of spectrum detection results according to the plurality of sample mass coefficients to obtain an iron ore composition detection result.

[0041] Traditional component identification suffers from two major issues: first, spectral detection results are decoupled from sample preparation quality, leading to errors in detection of low-quality samples. Second, reliance on a single analytical model fails to account for the error characteristics of samples of varying preparation quality. For example, samples containing trace amounts of water can cause elemental detection values to drift, but a fixed model struggles to provide targeted correction. Furthermore, the model's inherent structural bias can further amplify errors, leading to inaccurate results.

[0042] Step S40 in the method provided in the embodiment of the present application includes: Performing X-ray fluorescence spectroscopy on multiple test samples to obtain multiple spectral detection results; Obtaining a maximum sample processing time, and obtaining an optimized sample processing time within the optimized drying and grinding parameters, calculating a ratio of the optimized sample processing time to the maximum sample processing time, and calculating a time recognition coefficient; Calculating a plurality of mass recognition coefficients based on the plurality of sample mass coefficients, and calculating a plurality of component recognition coefficients in combination with the time recognition coefficients; Obtain Y iron ore component identification networks, and calculate the number of multiple identification networks based on the multiple component identification coefficients, where Y is a positive integer; Among them, obtaining Y iron ore component identification networks includes: According to the iron ore composition detection sample data, a sample spectrum detection result set and an actual sample iron ore composition set are collected; Using machine learning, we construct Y iron ore composition recognition network architectures. Dividing the sample spectrum detection result set and the sample iron ore component set with replacement to obtain Y pieces of iron ore component identification training data, and supervising the training of Y iron ore component identification networks respectively to obtain Y iron ore component identification networks; Multiple iron ore component recognition networks are randomly selected, multiple spectral detection results are input, multiple network iron ore component sets are obtained through recognition output, and the iron ore composition is obtained by calculating the mean as the iron ore composition detection result.

[0043] In the embodiment of the present application, X-ray fluorescence spectroscopy is performed on multiple test samples to obtain multiple spectral detection results.

[0044] Obtain the maximum sample processing time, and obtain the optimized sample processing time within the optimized drying and grinding parameters, calculate the ratio of the optimized sample processing time to the maximum sample processing time, and calculate the time recognition coefficient. The maximum sample processing time is the maximum sample processing time pre-set according to the experimental manual, illustratively, set to 60 minutes. Time recognition coefficient = 1-optimized sample processing time ÷ maximum sample processing time. For example, if the optimized sample processing time is 30 minutes, then the time recognition coefficient = 1-30 ÷ 60 = 0.5. The smaller the time recognition coefficient, the longer the optimized sample processing time and the more thorough the sample drying.

[0045] Based on multiple sample mass coefficients, multiple mass recognition coefficients are calculated. Mass recognition coefficient = 1 - sample mass coefficient. For example, if the sample mass coefficient is 0.518, the mass recognition coefficient is 0.482. Combined with the time recognition coefficient, multiple component recognition coefficients are calculated. Component recognition coefficient = mass recognition coefficient × time recognition coefficient. For example, if the mass recognition coefficient is 0.482 and the time recognition coefficient is 0.5, the component recognition coefficient = mass recognition coefficient × time recognition coefficient = 0.482 × 0.5 = 0.241. A larger component recognition coefficient indicates better sample quality and easier detection and identification.

[0046] Obtain Y iron ore composition identification networks, where Y is a positive integer. Based on historical iron ore composition detection sample data, collect a set of sample spectral detection results and an actual set of sample iron ore compositions; Using machine learning, Y iron ore component recognition network architectures were constructed.

[0047] An iron ore composition recognition network has a three-layer structure. The input layer uses one node to input sample data, the hidden layer uses 32 nodes, and is activated by the ReLU function. The number of nodes in the output layer is equal to the number of elements in the iron ore. For example, if the iron ore generally contains 20 elements, the number of nodes in the output layer is 20, and linear activation is used. The optimizer uses Adam, and the loss function uses the MAE function.

[0048] The sample spectral detection result set and the sample iron ore composition set are divided with replacement. For example, 40% of the sample spectral detection result set and the sample iron ore composition set are randomly divided each time as one set of iron ore composition identification training data. The division is repeated Y times with replacement to obtain Y sets of iron ore composition identification training data. The obtained Y sets of iron ore composition identification training data are used to supervise the training of Y iron ore composition identification networks until the model converges, and Y iron ore composition identification networks are obtained.

[0049] Obtain Y iron ore component identification networks. Calculate the number of identification networks based on multiple component identification coefficients. Y is a positive integer: Y = component identification coefficient × 20. Round up any decimals. For example, if the component identification coefficient is 0.241, then Y = 0.241 × 20 = 5 (round up). A smaller component identification coefficient indicates better sample processing, allowing for more accurate identification even with fewer identification networks, improving recognition efficiency.

[0050] A plurality of iron ore component recognition networks are randomly selected, and a plurality of spectral detection results are input and output to obtain a plurality of network iron ore component sets. The mean of the output values is calculated, for example, the arithmetic mean of the element content is calculated to obtain the iron ore composition as the iron ore composition detection result.

[0051] By utilizing the sample quality coefficient and dynamically allocating detection resources, high-quality samples are verified in parallel by more heterogeneous models, while the number of models is reduced for low-quality samples to avoid error superposition. At the same time, the time-consuming preparation process is combined to generate a time identification coefficient to balance detection efficiency and accuracy requirements. Ultimately, through the weighted fusion of multi-model outputs, high-confidence samples dominate the results, while suppressing outliers caused by preparation defects. The embodiment of the present application achieves seamless coordination between the component identification link and the front-end preparation quality, breaks through the problem of error accumulation in traditional detection methods, and provides more accurate and reliable iron ore composition detection results.

[0052] Example 2, as Figure 2 As shown, based on the same inventive concept as the iron ore composition detection method based on machine learning provided in Example 1, an embodiment of the present invention further provides an iron ore composition detection system based on machine learning, comprising: The information collection module 100 is used to collect iron ore characteristic information of the target iron ore to be tested for composition, and collect multiple iron ore samples; a grinding optimization module 200 for optimizing drying and grinding parameters based on the iron ore characteristic information to obtain optimized drying and grinding parameters, drying and grinding the plurality of iron ore samples, and collecting particle size and mass during each round of drying and grinding to obtain a plurality of particle size sequence sets and a plurality of mass sequences; A mass analysis module 300 is configured to prepare a plurality of test samples based on the dried and ground particles, and to perform sample mass analysis based on optimized drying and grinding parameters, a plurality of particle size sequence sets, and a plurality of mass sequences to obtain a plurality of sample mass coefficients; The component detection module 400 is used to perform X-ray fluorescence spectrum detection on multiple test samples to obtain multiple spectrum detection results, perform component identification on the multiple spectrum detection results according to the multiple sample mass coefficients, and obtain iron ore component detection results.

[0053] In one embodiment, the information collection module 100 is further configured to: Collecting iron ore characteristic information of the target iron ore to be tested for composition, wherein the iron ore characteristic information includes the type of iron ore, mineral hardness, and porosity; A plurality of iron ore samples are collected from the target iron ore.

[0054] In one embodiment, the grinding optimization module 200 is further configured to: Randomly generating first dry grinding parameters for dry grinding the iron ore sample, wherein the first dry grinding parameters include the number and time of repeated grinding and drying of the iron ore sample; Analyzing and processing the iron ore characteristic information and the first drying and grinding parameter to obtain a first sample processing adaptability; The first sample processing adaptability is obtained by analyzing and processing the iron ore characteristic information and the first drying and grinding parameter, including: Obtaining a first sample processing time of the first drying and grinding parameter, and calculating a ratio of a preset processing time to the first sample processing time to obtain a first time processing fitness; Inputting the drying and grinding parameters and the iron ore characteristic information into an iron ore sample processing predictor, and predicting and outputting a first predicted particle size and a first quality, wherein the iron ore sample processing predictor is trained using a sample drying and grinding parameter set, a sample iron ore characteristic information set, a sample predicted particle size set, and a sample moisture content set; Obtaining the standard particle size and standard mass of particles; Calculating the similarity between the first predicted granularity and the standard granularity, and calculating the ratio of the standard quality to the first quality, and combining the first time processing fitness to calculate and obtain a first sample processing fitness; Iterative optimization of drying and grinding parameters is performed to obtain optimized drying and grinding parameters with maximum sample processing adaptability; Drying and grinding the plurality of iron ore samples using the optimized drying and grinding parameters; The particle size and mass are collected during each round of dry grinding to obtain multiple particle size sequence sets and multiple mass sequences.

[0055] In one embodiment, the quality analysis module 300 is further configured to: Based on the dried and ground particles, multiple test samples were prepared; Based on the multiple particle size sequence sets, multiple particle size stability coefficients are calculated, and based on the last particle size in the multiple particle size sequences and the standard particle size, multiple basic particle size quality coefficients are calculated; Using the plurality of particle size stability coefficients respectively, a plurality of basic particle size quality coefficients are corrected and calculated to obtain a plurality of particle size quality coefficients; According to multiple quality sequences, multiple drying quality coefficients are calculated; A plurality of sample mass coefficients are calculated based on a plurality of particle size mass coefficients and a plurality of drying mass coefficients.

[0056] In one embodiment, the component detection module 400 is further configured to: Performing X-ray fluorescence spectroscopy on multiple test samples to obtain multiple spectral detection results; Obtaining a maximum sample processing time, and obtaining an optimized sample processing time within the optimized drying and grinding parameters, calculating a ratio of the optimized sample processing time to the maximum sample processing time, and calculating a time recognition coefficient; Calculating a plurality of mass recognition coefficients based on the plurality of sample mass coefficients, and calculating a plurality of component recognition coefficients in combination with the time recognition coefficients; Obtain Y iron ore component identification networks, and calculate the number of multiple identification networks based on the multiple component identification coefficients, where Y is a positive integer; Among them, obtaining Y iron ore component identification networks includes: According to the iron ore composition detection sample data, a sample spectrum detection result set and an actual sample iron ore composition set are collected; Using machine learning, we construct Y iron ore composition recognition network architectures. Dividing the sample spectrum detection result set and the sample iron ore component set with replacement to obtain Y pieces of iron ore component identification training data, and supervising the training of Y iron ore component identification networks respectively to obtain Y iron ore component identification networks; Multiple iron ore component recognition networks are randomly selected, multiple spectral detection results are input, multiple network iron ore component sets are obtained through recognition output, and the iron ore composition is obtained by calculating the mean as the iron ore composition detection result.

[0057] In summary, the embodiments of the present application have at least the following technical effects: This application proposes a method and system for iron ore composition detection based on machine learning. By integrating the inherent characteristic information of iron ore, it guides the optimization of the sample preparation process, and uses the dynamic monitoring data of the entire sample preparation process to build a quality assessment system, and then weights and integrates the output of multiple component identification models, significantly improving the accuracy, reliability and overall efficiency of iron ore composition detection. Compared with traditional methods that usually adopt fixed or empirical parameter settings for the sample pretreatment link, and subsequent component detection mostly relies on a single test, it is difficult to effectively deal with the inconsistency of the preparation process caused by differences in characteristics such as mineral hardness and porosity of iron ore samples, as well as the detection errors introduced, the technical solution provided by this application significantly overcomes the above limitations. First, based on the specific characteristic information of the target iron ore, an iterative optimization mechanism driven by a machine learning model is used to dynamically generate the optimal combination of drying and grinding parameters for each sample, ensuring that the sample is processed to the physical state most suitable for subsequent X-ray fluorescence spectroscopy detection, thereby improving the representativeness and uniformity of the sample from the source. Secondly, in the process of performing dry grinding according to the optimized parameters, the key physical indicators after each round of processing are continuously collected and recorded in real time for subsequent in-depth analysis. Based on these dynamic sequence data, the scheme calculates and integrates multiple coefficients that reflect the stability of particle size changes, the proximity of the final particle size to the target standard, and the drying effect, ultimately forming a comprehensive sample quality coefficient, which quantifies the degree of deviation of the physical properties of each prepared test sample from the ideal standard, providing a key basis for the reliability assessment of subsequent spectral results. Finally, in the component identification stage, the scheme uses the above-mentioned sample quality coefficient to quantify the confidence level of each test sample result, etc., and comprehensively forms a component identification coefficient for each test result. This dynamic and adaptive model mechanism can effectively utilize the complementary advantages of the model to offset the potential bias or error of a single model based on the actual situation of the sample's own preparation quality and processing efficiency, thereby obtaining iron ore composition detection results that are closer to the true value overall.

[0058] This application achieves the technical effect of significantly improving the accuracy of iron ore composition detection and the robustness of results, while effectively optimizing sample preparation efficiency and resource utilization, ultimately achieving high-reliability iron ore composition detection and analysis.

[0059] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0060] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

[0061] This specification and drawings are merely illustrative of the present application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, this application is intended to include such modifications and variations as fall within the scope of this application and its equivalents.

Claims

1. A method for detecting iron ore composition based on machine learning, characterized in that: The method comprises: Collecting iron ore characteristic information of the target iron ore to be tested for composition, and collecting multiple iron ore samples; Optimizing drying and grinding parameters based on the iron ore characteristic information to obtain optimized drying and grinding parameters, drying and grinding the multiple iron ore samples, and collecting particle size and mass during each round of drying and grinding to obtain multiple particle size sequence sets and multiple mass sequences; Based on the dried and ground particles, multiple test samples are prepared, and sample quality analysis is performed according to the optimized drying and grinding parameters, multiple particle size sequence sets, and multiple mass sequences to obtain multiple sample quality coefficients; X-ray fluorescence spectroscopy is performed on multiple test samples to obtain multiple spectral detection results, and components of the multiple spectral detection results are identified according to the multiple sample mass coefficients to obtain iron ore composition detection results.

2. The iron ore composition detection method based on machine learning according to claim 1, characterized in that: Collect iron ore characteristic information of the target iron ore to be tested for composition, and collect multiple iron ore samples, including: Collecting iron ore characteristic information of the target iron ore to be tested for composition, wherein the iron ore characteristic information includes the type of iron ore, mineral hardness, and porosity; A plurality of iron ore samples are collected from the target iron ore.

3. The iron ore composition detection method based on machine learning according to claim 1, characterized in that: According to the iron ore characteristic information, drying and grinding parameters are optimized to obtain optimized drying and grinding parameters, including: Randomly generating first dry grinding parameters for dry grinding the iron ore sample, wherein the first dry grinding parameters include the number and time of repeated grinding and drying of the iron ore sample; Analyzing and processing the iron ore characteristic information and the first drying and grinding parameter to obtain a first sample processing adaptability; Iterative optimization of drying and grinding parameters was performed to obtain the optimized drying and grinding parameters with maximum sample processing fitness.

4. The iron ore composition detection method based on machine learning according to claim 3, characterized in that: Analyzing and processing the iron ore characteristic information and the first drying and grinding parameter to obtain a first sample processing adaptability includes: Obtaining a first sample processing time of the first drying and grinding parameter, and calculating a ratio of a preset processing time to the first sample processing time to obtain a first time processing fitness; Inputting the drying and grinding parameters and the iron ore characteristic information into an iron ore sample processing predictor, and predicting and outputting a first predicted particle size and a first quality, wherein the iron ore sample processing predictor is trained using a sample drying and grinding parameter set, a sample iron ore characteristic information set, a sample predicted particle size set, and a sample moisture content set; Obtaining the standard particle size and standard mass of particles; The similarity between the first predicted granularity and the standard granularity is calculated, and the ratio of the standard quality to the first quality is calculated. The first sample processing fitness is obtained by calculation in combination with the first time processing fitness.

5. The iron ore composition detection method based on machine learning according to claim 1, characterized in that: The multiple iron ore samples are dried and ground, and the particle size and mass during each round of drying and grinding are collected to obtain multiple particle size sequence sets and multiple mass sequences, including: Drying and grinding the plurality of iron ore samples using the optimized drying and grinding parameters; The particle size and mass are collected during each round of dry grinding to obtain multiple particle size sequence sets and multiple mass sequences.

6. The iron ore composition detection method based on machine learning according to claim 1, characterized in that: Based on the dried and ground particles, multiple test samples were prepared. Sample quality analysis was performed based on optimized drying and grinding parameters, multiple particle size sequence sets, and multiple mass sequences to obtain multiple sample quality coefficients, including: Based on the dried and ground particles, multiple test samples were prepared; Based on the multiple particle size sequence sets, multiple particle size stability coefficients are calculated, and based on the last particle size in the multiple particle size sequences and the standard particle size, multiple basic particle size quality coefficients are calculated; Using the plurality of particle size stability coefficients respectively, a plurality of basic particle size quality coefficients are corrected and calculated to obtain a plurality of particle size quality coefficients; According to multiple quality sequences, multiple drying quality coefficients are calculated; A plurality of sample mass coefficients are calculated based on a plurality of particle size mass coefficients and a plurality of drying mass coefficients.

7. The iron ore composition detection method based on machine learning according to claim 1, characterized in that: Performing X-ray fluorescence spectroscopy on a plurality of test samples to obtain a plurality of spectral detection results, and performing component identification on the plurality of spectral detection results according to the plurality of sample mass coefficients to obtain an iron ore component detection result, including: Performing X-ray fluorescence spectroscopy on multiple test samples to obtain multiple spectral detection results; Obtaining a maximum sample processing time, and obtaining an optimized sample processing time within the optimized drying and grinding parameters, calculating a ratio of the optimized sample processing time to the maximum sample processing time, and calculating a time recognition coefficient; Calculating a plurality of mass recognition coefficients based on the plurality of sample mass coefficients, and calculating a plurality of component recognition coefficients in combination with the time recognition coefficients; Obtain Y iron ore component identification networks, and calculate the number of multiple identification networks based on the multiple component identification coefficients, where Y is a positive integer; Multiple iron ore component recognition networks are randomly selected, multiple spectral detection results are input, multiple network iron ore component sets are obtained through recognition output, and the iron ore composition is obtained by calculating the mean as the iron ore composition detection result.

8. The iron ore composition detection method based on machine learning according to claim 7, characterized in that: Obtain Y iron ore composition identification networks, including: According to the iron ore composition detection sample data, a sample spectrum detection result set and an actual sample iron ore composition set are collected; Using machine learning, we construct Y iron ore composition recognition network architectures. The sample spectrum detection result set and the sample iron ore component set are divided with replacement to obtain Y pieces of iron ore component identification training data, and Y iron ore component identification networks are supervised and trained respectively to obtain Y iron ore component identification networks.

9. A machine learning-based iron ore composition detection system, characterized in that: A system for implementing the iron ore composition detection method based on machine learning according to any one of claims 1 to 8, comprising: An information collection module is used to collect iron ore characteristic information of a target iron ore to be tested for composition, and to collect multiple iron ore samples; a grinding optimization module, configured to optimize drying and grinding parameters based on the iron ore characteristic information to obtain optimized drying and grinding parameters, dry and grind the multiple iron ore samples, and collect the particle size and mass during each round of drying and grinding to obtain multiple particle size sequence sets and multiple mass sequences; The mass analysis module is used to prepare multiple test samples based on the dried and ground particles, and to perform sample mass analysis based on optimized drying and grinding parameters, multiple particle size sequence sets, and multiple mass sequences to obtain multiple sample quality coefficients; The component detection module is used to perform X-ray fluorescence spectrum detection on multiple test samples to obtain multiple spectrum detection results, perform component identification on the multiple spectrum detection results according to the multiple sample mass coefficients, and obtain iron ore component detection results.

Images