Search

CN-121982537-A - Intelligent control system for barium ore separation and screening

CN121982537ACN 121982537 ACN121982537 ACN 121982537ACN-121982537-A

Abstract

The invention discloses an intelligent control system for barium ore separation and screening, and belongs to the technical field of industrial process control. The system specifically comprises an ore data acquisition module, a multidimensional feature extraction and fusion module, a classifier engine module and a sorting control module, wherein the ore data acquisition module acquires original ore data through a multidimensional sensor, the multidimensional feature extraction and fusion module extracts and fuses features based on the original data, the classifier engine module identifies and sorts and screens barium ores to form control instructions based on the fused features by using a machine learning algorithm, and the sorting control module executes screening actions based on the control instructions. The invention provides technical support for realizing the separation and the accurate separation of barium ore and self-adaptive control by utilizing a solution combining multi-sensor information fusion and a lightweight machine learning classifier.

Inventors

  • DU LIANG
  • ZHANG MAN

Assignees

  • 重庆巴山松科技发展有限公司

Dates

Publication Date
20260505
Application Date
20260127

Claims (9)

  1. 1. The intelligent control system for barium ore separation screening is characterized by comprising an ore data acquisition module, a multidimensional feature extraction fusion module, a classifier engine module and a separation control module which are connected in sequence; The ore data acquisition module acquires ore raw data through a multi-dimensional sensor; The multidimensional feature extraction fusion module is used for extracting important features from the raw ore data by utilizing a feature extraction technology, and obtaining numerical feature vectors of the ore through feature vector construction and standardization processing; the classifier engine module uses an offline trained random forest classifier model as a classification engine to perform classification screening and result voting on the ore numerical feature vector to form a decision instruction, and the decision instruction is converted into a classification control signal; the sorting control module analyzes the signals based on the sorting control signals, and sends the signals to the hardware control mechanism through the internal control decision rule to finish barium ore separation.
  2. 2. The intelligent control system for barium ore separation and screening according to claim 1, wherein the ore data acquisition module comprises a hyperspectral linear array imaging sub-module and a miniature XRF sensing sub-module; the hyperspectral linear array imaging submodule is used for acquiring an original hyperspectral image cube; the miniature XRF sensing sub-module is used for acquiring an original X-ray fluorescence spectrum.
  3. 3. The intelligent control system for barium ore separation and screening according to claim 2, wherein the multi-dimensional feature extraction fusion module comprises a hyperspectral feature extraction sub-module and an XRF feature extraction sub-module; The hyperspectral feature extraction submodule selects a minimum rectangular region containing whole ore particles as an interested region based on the space dimension of an original hyperspectral image cube, averages the spectrum of all pixels in the region to obtain an average reflection spectrum vector R (lambda) of the particles, calculates a spectrum angle theta SAM between the average reflection spectrum vector R (lambda) and a standard barite reference spectrum, selects a band range sensitive to the diagnostic absorption feature of barite, calculates an absorption deep valley ratio R band to enhance barium ore features, and finally combines the spectrum angle theta SAM and the absorption deep valley ratio R band into a two-dimensional hyperspectral feature vector H= [ theta SAM ,R band ]; The XRF characteristic extraction submodule utilizes an SNIP algorithm to smooth and subtract background of an original X-ray fluorescence spectrum, utilizes a Gaussian fitting method to characteristic peaks of key elements to calculate net intensities of the key elements, the key elements comprise barium, strontium, iron and silicon, intensity ratios of the net intensities of the key elements are calculated, the intensity ratios comprise barium-iron ratios, barium-silicon ratios and strontium-barium ratios, and finally the ratios are combined into a three-dimensional XRF characteristic vector X= [ barium-iron ratios, barium-silicon ratios and strontium-barium ratios ].
  4. 4. The intelligent control system for barium ore separation and screening according to claim 3, wherein the multi-dimensional feature extraction and fusion module further comprises a feature standardization and fusion sub-module; The feature standardization fusion submodule is used for splicing the hyperspectral feature vector H= [ theta SAM ,R band ] and the XRF feature vector X= [ barium-iron ratio, barium-silicon ratio and strontium-barium ratio ] into a 5-dimensional vector F, the system is used for introducing an offline training set to obtain a mean value mu and a standard deviation sigma from the offline training set, and Z-score standardization is carried out on the 5-dimensional vector F by using the mean value mu and the standard deviation sigma, so that a standardized feature vector FN is generated, and a timestamp is recorded at the same time.
  5. 5. The intelligent control system for barium ore separation and screening according to claim 4, wherein the classifier engine module comprises a sample library construction sub-module, a model construction sub-module and a verification parameter sub-module; the sample library construction submodule sets a real label Y for the feature vector FN based on the feature vector FN and a time stamp, wherein the value ore is represented when Y=1, the waste stone ore is represented when Y=0, and feature data are randomly divided into a training set Dt and a verification set Dv according to the proportion of [7:3 ]; The model construction submodule randomly extracts N samples from the training set Dt based on the replacement ground, repeatedly carries out T times in the process, constructs T different self-service sample sets Done, constructs an unbeared CART decision tree for each self-service sample set Done, randomly selects m features when each node is split in the process of constructing the CART decision tree, and selects the features and the threshold value to split according to the reduction of the Indonesia, and repeats the above processes until T decision trees are generated, and combines the T decision trees to form an uneptimized original random forest model RF 1 ; The verification parameter submodule verifies the performance of the original random forest model RF 1 by utilizing the verification set Dv based on the original random forest model RF 1 and the verification set Dv, performance indexes during verification comprise recall rate and accuracy rate, internal super parameters of the original random forest model RF 1 are finely adjusted through a grid search method after verification to obtain an optimized random forest model RF boss , the structure of each decision tree in the optimized random forest model RF boss is serialized into a classification model sequence parameter table, and the classification model sequence parameter table is stored in a nonvolatile memory.
  6. 6. The intelligent control system for barium ore separation and screening according to claim 5, wherein the classifier engine module further comprises a model loading sub-module, a vector receiving sub-module and a forest traversal voting sub-module; the model loading submodule acquires a classification model sequence parameter table from a nonvolatile memory, loads the classification model sequence parameter table into a system high-speed RAM, initializes a decoding data structure and obtains a classification ready model RF boss (s); The vector receiving sub-module calls an ore data acquisition module and a multidimensional feature extraction fusion module, analyzes a data packet, extracts a 5-dimensional standardized feature vector and a time stamp, and obtains a total vector to be classified; The forest traversal voting submodule calls a classification ready model RF boss (s) to conduct online prediction on total vectors to be classified, traverses all T trees and collects prediction results of all the trees to form a voting set Vot= [ c 1 ,c 2 ...c t ], wherein c represents a class label.
  7. 7. The intelligent control system for barium ore separation and screening according to claim 6, wherein the classifier engine module further comprises a decision instruction packaging sub-module and an instruction conveying sub-module; The decision instruction packaging submodule counts the votes of the 'value ores' with the prediction result of 1 based on a voting set Vot= [ c 1 ,c 2 ...c t ] and calculates the confidence coefficient, determines a final decision by using a double-threshold decision method, and packages the final decision, the confidence coefficient and the time stamp into a standard control instruction packet; The instruction delivery submodule transmits the instruction packet to the sorting control module by using a communication protocol based on a standard control instruction packet.
  8. 8. The intelligent control system for barium ore separation and screening according to claim 7, wherein the separation control module comprises an instruction receiving sub-module, a delay calculating sub-module and a task queue management sub-module; The instruction receiving submodule extracts key fields based on a standard control instruction packet, wherein the key fields comprise a final decision, a confidence coefficient and a time stamp, the final decision comprises a final control instruction [1 does not act, 0 is blown, -1 is led to a middling trough ], and finally the key fields are repackaged into a triplet; The delay calculation operator module makes decision and judgment on final control instruction parameters in the triples, when the final control instruction is not 0, the system immediately stops, at the moment, only records an event into the operation log, when the final control instruction is 0, theoretical motion delay calculation and instruction trigger absolute time calculation are executed, and finally the instruction trigger absolute time is output; The task Queue management sub-module is used for programming new blowing instructions into a blowing task Queue according to the order of small to large according to the absolute time triggered by the instructions, calculating the time interval between the new blowing instructions and the previous blowing instructions in the Queue, calculating the space interval of corresponding ore particles based on the time interval, judging that the distance between the two ores is too close when the space interval is smaller than the set minimum interval threshold Q, canceling the later instruction by the system, guiding the decision to a middling tank by-1, simultaneously sending conflict alarm to an upper computer, clearing the outdated instruction, and finally obtaining the updated blowing task Queue.
  9. 9. The intelligent control system for barium ore separation and screening according to claim 8, wherein the separation control module further comprises an action execution sub-module; And the action execution feedback submodule executes the injection task to finish barium ore separation and screening according to the task time stamp based on updating the injection task Queue, and records the action executed at the time into a system operation log.

Description

Intelligent control system for barium ore separation and screening Technical Field The invention relates to the technical field of industrial process control, in particular to an intelligent control system for barium ore separation and screening. Background The barium ore is barite containing barium sulfate mineral as main component, as important industrial raw material, widely applied to fields of petroleum drilling, chemical industry, building materials and the like, the development of separation technology is deeply bound with the overall technical progress of the mineral processing industry, early barium ore separation relies on traditional manual separation mode, separation of valuable ore and waste rock is realized by virtue of experience judgment of appearance characteristics of color, luster and the like of ore by operators, in recent years, the rise of embedded technology provides new support for intelligent upgrading of barium ore separation technology, the rapid development of automatic technology promotes the upgrading of barium ore separation technology to the accurate direction of sensing detection and automatic separation, and sensing detection technology begins to be widely applied to the field of barium ore separation, wherein the hyperspectral detection technology can rapidly capture characteristic spectrum difference of barium ore and gangue in visible light to near infrared band by virtue of the accurate identification capability of spectrum fingerprints of minerals, and rapid qualitative identification of mineral components is realized. However, the prior art relies on a single vision sensor to detect the surface of the ore, and the surface of the ore usually contains a plurality of attachments, so that the surface of the ore is in a state of uneven granularity, and the barium ore is similar to gangue in appearance, which results in low precision and poor adaptability of the traditional separation system in separating the barium ore. Disclosure of Invention The invention aims to provide an intelligent control system for barium ore separation and screening, which is developed based on a real-time barium ore classification assembly line, wherein an ore data acquisition module is used as an acquisition terminal of ore original data, a multidimensional feature extraction and fusion module is used for extracting key feature values from the original data and fusing the key feature values and attaching an accurate time stamp, a classifier engine module is used for calculating by using a classification model to judge whether ore particles are valuable ores or waste stones, classification results are bound with the original time stamp and packaged into a standard control instruction packet, and a separation control module is used for implementing self-adaptive separation control so as to solve the problems of low precision and poor adaptability of the traditional separation system in barium ore separation, and obtain the technical effect of self-adaptive optimal control. The following scheme is specifically provided: the invention relates to an intelligent control system for barium ore separation screening, which comprises an ore data acquisition module, a multidimensional feature extraction fusion module, a classifier engine module and a separation control module, wherein the modules are connected in sequence; The ore data acquisition module acquires ore raw data through a multi-dimensional sensor; The multidimensional feature extraction fusion module is used for extracting important features from the raw ore data by utilizing a feature extraction technology, and obtaining numerical feature vectors of the ore through feature vector construction and standardization processing; the classifier engine module uses an offline trained random forest classifier model as a classification engine to perform classification screening and result voting on the ore numerical feature vector to form a decision instruction, and the decision instruction is converted into a classification control signal; the sorting control module analyzes the signals based on the sorting control signals, and sends the signals to the hardware control mechanism through the internal control decision rule to finish barium ore separation. Further, the ore data acquisition module comprises a hyperspectral linear array imaging sub-module and a miniature XRF sensing sub-module; the hyperspectral linear array imaging submodule is used for acquiring an original hyperspectral image cube; the miniature XRF sensing sub-module is used for acquiring an original X-ray fluorescence spectrum. Further, the multi-dimensional feature extraction fusion module comprises a hyperspectral feature extraction sub-module and an XRF feature extraction sub-module; The hyperspectral feature extraction submodule selects a minimum rectangular region containing whole ore particles as an interested region based on the space dimension of an original hyperspectral imag