A beneficiation process flow calculation and process design system

Abstract

The application belongs to the technical field of data processing, calculation graph modeling and process design software of mineral processing technological process, and relates to a mineral processing technological process calculation and process design system. The system comprises a storage, a processor and a mineral processing technological process calculation program. The program is configured with instructions of flow chart construction, multi-dimensional mineral flow attribute binding, conservation constraint generation, closed circuit return flow solving and residual path checking. Crushing, grinding, classification, flotation, concentration, scavenging and middling return links are converted into calculation nodes and data edges. Solid content, water content, particle size composition, grade, metal content and circulating load attributes are bound in the data edges. Material conservation, metal content conservation and water content conservation constraints are generated. The return flow is taken as a to-be-solved edge variable to participate in the whole-process constraint solving. The application can keep the closed-circuit mineral processing process calculation consistent with multi-dimensional conservation, reduce the inconsistent return flow values caused by manual trial calculation, and provide error checking and local recalculation basis for process design.

Description

Technical Field

[0001] This invention belongs to the technical field of mineral processing process flow data processing, computational graph modeling and process design software, and relates to a mineral processing process flow calculation and process design system. Background Technology

[0002] Mineral processing flow calculations and process design are typically based on the properties of the raw ore, crushing and grinding processes, classification conditions, separation indices, concentrate and tailings yields, and water allocation. Material balance calculations are performed using spreadsheets, process calculation software, or process design platforms. The conventional approach is to break down crushing, grinding, classification, flotation, cleaning, and scavenging processes into several unit operations according to the process route, and then create calculation fields for each unit operation, including input quantities, output quantities, grade, recovery rate, yield, and concentration. The unit operations are usually passed between each other by using the output results of the previous stage as the input results of the next stage. Designers then manually set or repeatedly calculate the middlings return rate, return sand ratio, and circulating load based on experience.

[0003] In existing calculation methods, the closed-loop grinding and classification process and the flotation return process are among the more complex parts of mineral processing calculations. The grinding and classification process needs to simultaneously handle the fresh feed rate, return sand rate, overflow rate, underflow rate, particle size distribution, and pulp concentration. The flotation process needs to simultaneously handle the material distribution among rougher concentrate, rougher tailings, cleaner middlings, scavenger middlings, and final tailings. Conventional software typically treats the closed-loop return flow as a known input or empirical parameter, and then calculates the node outputs based on local element equations. This method can complete the calculation of a single stage, but the closed-loop return flow itself will simultaneously affect the solids, water, particle size distribution, and metal content of multiple nodes before and after it, creating a mutual constraint relationship between the return flow value and the overall process equilibrium result.

[0004] Existing process design platforms can also express mineral processing routes through flowcharts, using nodes and connections to display mineral flow directions, and filling in parameters such as throughput, yield, grade, recovery rate, and concentration in node attributes. The computational focus of these platforms is primarily on process visualization, unit index calculation, and report generation; the flowchart usually only serves as a process display, and the calculation relationships still rely on manually created tables, formulas, or fixed templates. For closed-loop processes with multiple return branches, local calculation errors at a certain node may continue to propagate along the sequential flow direction or enter the preceding node with the return flow, further altering the input conditions of the preceding node. Due to the lack of a unified identifier for cross-node error paths, designers find it difficult to determine whether the error originates from particle size distribution, grade distribution, water distribution, or improper return flow values.

[0005] The main technical problem with existing technologies is that mineral flow data in complex closed-loop beneficiation processes have multi-dimensional coupling relationships. Conventional calculation methods typically calculate each process unit separately and treat return flows as empirical given values ​​or local iterative values. This makes it difficult to maintain consistency between throughput, metal content, water content, particle size distribution, grade, and circulating load within the same calculation framework. Even when the calculation of local nodes meets the preset formula, the entire process may still experience non-closed loops in material conservation, metal conservation, or water conservation. Furthermore, the source of error cannot be located and corrected along the process topology. Ultimately, the process design results rely on manual trial calculations and post-process verification, making it difficult to form stable and consistent closed-loop process calculation results. Summary of the Invention

[0006] The purpose of this invention is to provide a mineral processing flow calculation and process design system that can effectively solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A mineral processing flow calculation and process design system includes a memory, a processor, and a mineral processing flow calculation program stored in the memory and executed by the processor. The mineral processing flow calculation program is configured with instructions for constructing a flow chart, binding multi-dimensional mineral flow attributes, generating conservation constraints, solving closed-loop return flow, and verifying residual paths. The directed flowchart construction instructions convert the crushing, grinding, classification, flotation, cleaning, scavenging and middlings return process steps into computing nodes and data edges. The multidimensional mineral flow attribute binding instruction binds the solids content, water content, particle size distribution, grade, metal content, and circulating load attributes to the data edge; The conservation constraint generation instruction generates material conservation, metal conservation, and water conservation constraints based on the computing node and the data edge; The closed-loop return flow solving instruction uses the return flow as the edge variable to be solved in the whole process constraint solving; The residual path verification instruction locates the cross-node error propagation path based on the node residual and edge residual.

[0008] Preferably, the directed flowchart construction instruction includes process segment semantic parsing rules and topology adjacency generation rules. The process segment semantic parsing rules divide the input mineral processing process data into sequential segments, parallel segments, merged segments, and return segments. The topology adjacency generation rules generate an adjacency matrix based on the material flow direction between process segments. The row and column elements of the adjacency matrix correspond to the computing nodes, and the matrix elements carry mineral flow category identifiers, so that the raw ore flow, concentrate flow, tailings flow, middlings flow, return sand flow and slurry water flow have different data edge types in the same flow diagram; The directed flowchart construction instruction also assigns loop numbers to closed loops and writes the loop numbers to the data edges corresponding to the return flow.

[0009] Preferably, the conservation constraint generation instruction includes a node equation template and an edge attribute mapping table. The node equation template sets input and output variable relationships according to the grinding and classification node, flotation separation node, cleaning node, scavenging node and concentration and dewatering node respectively. The edge attribute mapping table maps solid quantity, water quantity, particle size distribution, target ore grade, gangue grade, and metal quantity to vector components of the same data edge. When generating each node equation, the conservation constraint generation instruction writes the input edge vector, output edge vector, and node transformation variables into the constraint matrix, and retains the column of unknown variables corresponding to the closed-loop return flow in the constraint matrix.

[0010] Preferably, the closed-loop return flow solution instruction includes loop variable initialization rules, hierarchical solution rules, and residual correction rules; The loop variable initialization rule determines the set of edge variables to be solved based on the loop number. The hierarchical solution rule first solves the basic edge variables for sequential links without return flows, and then solves the loop edge variables for closed links with return flows. The residual correction rule decomposes the residual of the whole process into solid quantity residual, water quantity residual, particle size residual and metal quantity residual, and performs targeted correction on the return flow variables according to the data edge type corresponding to the residual, so that the middlings return, sand return and fine tailings return enter different correction branches respectively.

[0011] Preferably, the semantic parsing rules for the process segment also include node role merging rules and virtual edge completion rules; The node role merging rule merges computing nodes that appear repeatedly within the same process segment and have the same connection relationship into composite nodes with multiple output ports, and retains the mineral flow category identifier corresponding to each output port. The virtual edge completion rule generates virtual data edges based on the material type, process segment sequence, and loop number of the preceding and following nodes when there are merging points in the process data but no explicit intermediate flow direction. The virtual data edges only carry topological connection attributes and attribute identifiers to be completed, and the corresponding vector components are filled in by the edge attribute mapping table when the conservation constraints are generated subsequently.

[0012] Preferably, the node equation template includes a grain size migration sub-template and a grade migration sub-template. The grain size migration sub-template establishes grain size vector constraints according to the input edge grain size mass distribution, node classification variables, and output edge grain size mass distribution. The grade migration sub-template establishes grade vector constraints according to the input edge metal content, target ore grade, yield variables, and output edge metal content. The constraint matrix records both the particle size vector index and the grade vector index for the same output edge, and sets an index association item between the grinding and classification node and the flotation separation node, so that the mineral flow vector after the particle size change can be used as the input edge vector of the subsequent separation node.

[0013] Preferably, the residual correction rule includes a residual source marking rule and a key edge selection rule. The residual source marking rule writes the solid quantity residual, water quantity residual, particle size residual and metal quantity residual generated by each node into the residual label table. The residual label table records the node to which the residual belongs, the vector component to which the residual belongs, the associated input edge and the associated output edge. The key edge selection rule determines data edges that are simultaneously associated with two or more residual components and are located on closed loops based on the residual label table, and arranges the determined data edges into a return stream correction sequence. The return stream correction sequence is generated according to the loop number, the number of residual components, and the node topological distance.

[0014] Preferably, the residual path verification instruction includes local closure verification rules, global closure verification rules, and cross-node path tracing rules; The local closure verification rule verifies the solid quantity, water quantity, and metal quantity of the input and output edge vectors for each computing node. The global closure verification rule establishes a full-process verification equation with the raw ore input edge, concentrate output edge, tailings output edge, and all return stream data edges as boundaries; When the local closure check passes but the global closure check fails, the cross-node path tracing rule generates an error propagation path based on the adjacency matrix, loop number, and residual label table, and writes the return stream data edge in the error propagation path into the edge set to be reviewed.

[0015] Preferably, the return stream correction sequence is configured with an association constraint locking rule. Before selecting a return stream data edge for variable correction, the association constraint locking rule reads other data edges that share the same computing node or the same loop number with the selected return stream data edge, and sets the conservation constraint corresponding to the read data edge as an adjoint constraint. When the closed-loop return flow solution instruction updates the solid quantity vector component, water quantity vector component, particle size vector component, or metal quantity vector component of the selected return flow data edge, it simultaneously updates the same vector index item in the accompanying constraint and writes the vector components before and after the update, residual labels, and loop numbers into the correction record table.

[0016] Preferably, the set of edges to be reviewed is configured with a closed-loop recalculation boundary generation rule, which generates a local recalculation subgraph with the first residual node, the last residual node and the return stream data edge between them in the error propagation path as the boundary. The local recalculation subgraph inherits the node equation template, edge attribute mapping table, and loop number from the original directed flowchart, and masks data edge variables that are not in the local recalculation subgraph. The closed-loop return flow solution instruction only performs a second solution on the edge variables and associated constraints within the local recalculation subgraph, and writes the second solution result back to the variable column corresponding to the full-process constraint matrix.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By employing a technical solution combining directed flow graph construction, multi-dimensional mineral flow attribute binding, conservation constraint generation, closed-loop return flow solution, and residual path verification, the process steps such as crushing, grinding, classification, flotation, cleaning, scavenging, and middlings return are transformed into nodes and data edges with computational relationships. Solids quantity, water quantity, particle size distribution, grade, metal quantity, and circulating load are bound to the same mineral flow data edge. When generating material conservation, metal conservation, and water conservation constraints, the system incorporates the return flow as an edge variable to be solved in the entire process solution, rather than as an external empirical value input. Therefore, the value of the closed-loop return flow is jointly determined with the input-output relationship of the preceding and following nodes in the same constraint matrix. Local node balance and overall process boundary balance can form a consistent computational relationship, reducing the possibility of non-closed-loop results caused by separate unit calculations.

[0018] 2. By employing techniques such as process segment semantic parsing, topological adjacency generation, node equation templates, edge attribute mapping tables, residual source marking, key edge selection, and local recalculation subgraphs, the process calculation results possess a traceable correction path. Sequential segments, parallel segments, merged segments, and return segments are written into the adjacency matrix. Raw ore flow, concentrate flow, tailings flow, middlings flow, return sand flow, and slurry water flow participate in the calculation with different data edge types. Residuals are classified into solids residuals, water residuals, particle size residuals, and metal residuals, and are recorded corresponding to their respective nodes, associated input edges, and associated output edges. When a local closure check passes but a global closure check fails, the error propagation path can be generated based on the adjacency matrix, loop number, and residual label table. Subsequently, only the edge variables and accompanying constraints within the relevant closed loop subgraphs are recalculated, avoiding repeated rewriting of irrelevant process segments and ensuring that the calculation correction process corresponds to the error source. Attached Figure Description

[0019] Figure 1 This is a flowchart of the overall processing flow of the mineral processing flow calculation and process design system of the present invention. Figure 2 This is the flowchart for the directed flowchart construction and closed loop identification of the present invention; Figure 3 This is a flowchart of the conservation constraint generation and closed-loop return flow solution of the present invention; Figure 4 This is a flowchart of the residual path verification, critical edge correction, and local recalculation process of the present invention. Detailed Implementation

[0020] 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, not all, of the embodiments of the present invention. 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.

[0021] Please refer to Figure 1This embodiment provides a mineral processing flow calculation and process design system, including a memory, a processor, and a mineral processing flow calculation program stored in the memory and executed by the processor. The mineral processing flow calculation program is configured with directed flow graph construction instructions, multi-dimensional mineral flow attribute binding instructions, conservation constraint generation instructions, closed-loop return flow solution instructions, and residual path verification instructions. The directed flow graph construction instructions convert crushing, grinding, classification, flotation, cleaning, scavenging, and middlings return process steps into computational nodes and data edges. The multi-dimensional mineral flow attribute binding instructions... The command binds solids quantity, water quantity, particle size distribution, grade, metal content, and cyclic load attributes to the data edge. The conservation constraint generation command generates material conservation, metal conservation, and water conservation constraints based on the computing node and the data edge. The closed-loop return flow solution command uses the return flow as the edge variable to participate in the full-process constraint solution. The residual path verification command locates the cross-node error propagation path based on the node residual and edge residual. During execution, the processor reads the user-input process route text, tabular data, or process configuration data, and converts each process step into a specific... The computational nodes, with input and output ends, convert the mineral flows between any adjacent process stages into data edges. Then, the solids content, water content, particle size distribution, grade, metal content, and circulating load of each data edge are written into the same mineral flow attribute record. This allows the grinding and classification closed loop, flotation ore return, and tailings return to participate in the calculation within a unified data structure. The conservation constraint generation instructions do not use the local formulas of a single node as the final calculation boundary; instead, they write all computational nodes, all data edges, and the unknown variables of the closed loop return flow into the overall process constraint matrix. The closed loop return flow solution instructions... Within the constraint matrix, the return flow variables are solved and the input and output vectors of the relevant nodes are updated synchronously. When an unclosed residual appears in the solution result, the residual path verification instruction reads the node, edge, and attribute component to which the residual belongs, and traces the propagation chain of the error between the sequential flow direction and the return flow direction along the directed flow graph. In this embodiment, the return flow values ​​that originally relied on manual trial calculations are transformed into variables to be solved under the constraint of full-process conservation, so that the solids quantity, water quantity, particle size composition, grade, metal quantity, and circulating load of the complex closed-circuit mineral processing process can form consistent results within the same calculation framework.

[0022] In this embodiment, the mineral flow attributes of each data edge are stored using a vectorized structure, and the mineral flow vector can be represented as: ; in, This represents the data edge vector with the number e. Indicates the amount of solids. Indicates water volume. Represents the granular composition vector. Indicates the grade of the target ore. Indicates the amount of metal. This indicates the cyclic load; if the return flow of a certain intermediate mine... The value is 120. The value is 80. The value can be [0.15, 0.35, 0.50]. The value is 0.32. The value is 38.4. If the value is 1.6, the processor can simultaneously read the mass attribute, water attribute, grain size attribute, and metal attribute of the returned stream when executing subsequent conservation constraints, instead of breaking them down into unrelated table fields. This value selection method enables the multidimensional attributes of the same data edge to change synchronously when the returned stream is corrected. This embodiment provides a unified data object for multidimensional conservation calculations, avoiding inconsistent attribute references for the same mineral stream in different calculation tables.

[0023] Preferably, refer to Figure 2 The directed flowchart construction instruction includes process segment semantic parsing rules and topological adjacency generation rules. The process segment semantic parsing rules divide the input mineral processing process data into sequential segments, parallel segments, merged segments, and return segments. The topological adjacency generation rules generate an adjacency matrix based on the material flow direction between process segments. The row and column elements of the adjacency matrix correspond to the computing nodes, and the matrix elements carry mineral flow category identifiers, so that the raw ore flow, concentrate flow, tailings flow, middlings flow, return sand flow, and slurry water flow have different data edge types in the same flow diagram. The directed flowchart construction instruction also assigns loop numbers to closed loops and writes the loop numbers into the data edges corresponding to the return flows. Specifically, after reading the process flow data, the system first identifies the sequential connection relationships such as crushing followed by grinding, grinding followed by classification, and classification overflow followed by flotation, and then identifies the roughing concentrate entering the flotation process. The system establishes parallel or branch connections between the fine and rough tailings and the scavenging processes. It then identifies return connections such as middlings from the fined process returning to the rougher, middlings from the scavenging process returning to the rougher, and graded sand returning to the grinding process. The process segment type obtained from semantic parsing is not directly used as a display label, but rather as the basis for determining edge direction, edge type, and loop boundaries in constraint solving. The topological adjacency generation rules encode each computation node into a node set, writing the relationship between any two nodes with mineral flow transmission into matrix elements, and attaching category identifiers for raw ore flow, concentrate flow, tailings flow, middlings flow, return sand flow, or slurry water flow to the matrix elements. When it is found that the endpoint of a data edge is earlier than its starting point in the topological order, the system identifies this data edge as a return edge and configures a loop number. This embodiment makes the flowchart no longer just for graphical display, but a fundamental data structure for generating conservation constraints, solving return flows, and tracking residuals.

[0024] In a preferred embodiment, the adjacency matrix is ​​stored in a class-encoded form, and the matrix elements can be represented as: ; in, This represents the adjacent elements between the i-th and j-th computing nodes. Indicates the corresponding data edge, This indicates the data edge category encoding. Raw ore flow, middlings flow, return sand flow, concentrate flow, tailings flow, and slurry water flow each correspond to different non-zero codes. Indicates the connection direction symbol. It takes a value of 1 when data edges are passed sequentially along the process flow, a value of -1 when the data edge is a return flow, and a value of -1 when there is no mineral flow connection between the two nodes. The value is 0. For example, when graded sand returns to grinding, if the return sand flow category code is 3 and the connection direction symbol is negative 1, then the corresponding matrix element is negative 3. Based on this, the processor can determine that the data edge belongs to the return sand flow and enter the closed-loop solution branch without reading additional text descriptions. In this embodiment, the same matrix element is used to express the existence of connection, mineral flow type and return direction at the same time, so that the subsequent residual path tracing can directly complete the edge filtering based on matrix operations.

[0025] Further, refer to Figure 3 The conservation constraint generation instruction includes a node equation template and an edge attribute mapping table. The node equation template sets input and output variable relationships for grinding and classification nodes, flotation and separation nodes, cleaning nodes, scavenging nodes, and thickening and dewatering nodes, respectively. The edge attribute mapping table maps solids quantity, water quantity, particle size distribution, target ore grade, gangue grade, and metal quantity to vector components of the same data edge. When generating each node equation, the conservation constraint generation instruction writes the input edge vector, output edge vector, and node transformation variables into a constraint matrix, and retains the column of unknown variables corresponding to the closed-loop return flow in the constraint matrix. Specifically, the node equation template for the grinding and classification node reads the ore edge, return sand edge, overflow edge, and settling sand edge, and follows the principles of solids conservation and water quantity conservation. The conservation and particle size migration relationships form multi-row constraints. The node equation template of the flotation separation node reads the influent edge, concentrate edge, tailings edge, and middlings edge, and forms multi-row constraints according to the yield allocation, target mineral recovery allocation, and metal balance. The clean and scavenging nodes write part of the output edge into the return flow variable column according to the middlings return direction. The concentration and dewatering node sets the solids component as the conservation correlation term and writes the water component into the migration correlation term. The edge attribute mapping table specifies the column position of each data edge vector component in the constraint matrix, so that the solids, water, particle size composition, grade, and metal content of the same mineral flow are consistently called in multiple node equations. This embodiment enables the calculation relationship of different process nodes to be uniformly expressed in the same matrix, reducing the reference break between local table formulas.

[0026] In this embodiment, the conservation relations in the constraint matrix can be expressed as: ; in, This represents the coefficient matrix generated by the node equation template and the edge attribute mapping table. This represents the column vector of variables to be determined, which includes data edge vector components and node transformation variables. This represents a constant vector formed by known raw ore input, design target boundary, and known process parameters. If a certain grinding and classification closed loop includes feed solids, return sand solids, overflow solids, and underflow solids, then the corresponding matrix row can set the feed solids column and return sand solids column as input items, and the overflow solids column and underflow solids column as output items. The return sand solids are retained as a variable to be determined in z. When the feed solids value is 100 and the underflow return relationship enters the closed loop solution, the processor obtains the matching value between the return sand solids and the overflow solids through matrix solution, instead of pre-fixing the return sand amount. In this embodiment, the closed loop return flow is incorporated into a unified linear or piecewise linear constraint framework, so that the return flow variable and the whole process boundary conditions converge together.

[0027] Preferably, the closed-loop return flow solution instruction includes loop variable initialization rules, hierarchical solution rules, and residual correction rules. The loop variable initialization rules determine the set of edge variables to be solved based on the loop number. The hierarchical solution rules first solve the basic edge variables for sequential links without return flows, and then solve the loop edge variables for closed-loop links with return flows. The residual correction rules decompose the overall residual into solids residual, water residual, particle size residual, and metal residual, and perform targeted correction on the return flow variables according to the data edge type corresponding to the residual, so that middlings return, sand return, and tailings return enter different correction branches respectively. Specifically, the loop variable initialization rules search for matrix elements with negative direction signs in the adjacency matrix and read the corresponding data edge category code and loop number. Data belonging to the same loop number... The edges are written into the same set of edge variables to be solved. The hierarchical solution rule first masks the return edges and performs a basic solution on the sequential link from the raw ore input to the staged output to obtain the initial values ​​of the nodes without loop feedback. Then, the return edge masking is removed and the set of edge variables to be solved is written into the constraint matrix. The residual correction rule calculates the solid residual, water residual, particle size residual and metal residual for each node and each data edge according to the solution results. When the main residual appears in the particle size component and the data edge type is sand return flow, the correction branch prioritizes adjusting the sand return edge variables associated with the grading node. When the main residual appears in the metal content component and the data edge type is middlings flow, the correction branch prioritizes adjusting the middlings return variables associated with the coarse or fine separation nodes. This embodiment avoids averaging correction on all return flows, so that different types of closed-loop variables enter the corresponding calculation branches according to their physical properties.

[0028] In a preferred embodiment, the residual correction amount is calculated based on the coupling strength between the residual components and the return edge, and the update relationship of the return edge variables can be expressed as: ; in, This represents the value of edge r returned after the t-th solution. This indicates the variable value after the next correction. This indicates the correction step size coefficient. This represents the matrix coefficients of edge r with respect to residual component k. This represents the set of return edges associated with the residual component k. This indicates a small positive number used to avoid fractions with a denominator of 0. This represents the current value of the residual component k, if a certain metal quantity residual... The value is 6, and the absolute values ​​of the coefficients of the two return edges from the central mine for this residual are 2 and 1, respectively. If the value is 0.5, the return edge with a larger coefficient will receive a larger correction amount. The processor will then concentrate the residual correction on the data edge with a higher contribution. In this embodiment, the residual correction direction is derived from the coupling relationship in the constraint matrix, rather than manually specifying the correction order.

[0029] Furthermore, the semantic parsing rules for the process segment also include node role merging rules and virtual edge completion rules. The node role merging rules merge computational nodes that appear repeatedly within the same process segment and have consistent connection relationships into composite nodes with multiple output ports, while retaining the mineral flow category identifiers corresponding to each output port. The virtual edge completion rules generate virtual data edges based on the material type, process segment order, and loop number of the preceding and following nodes when there are merging points in the process data but no explicit intermediate flow direction. The virtual data edges only carry topological connection attributes and attribute identifiers to be completed, and the corresponding vector components are filled in by the edge attribute mapping table when subsequent conservation constraints are generated. In specific implementation, the system performs hash comparison on the node names, node roles, upstream sources, and downstream destinations in the input process data. If multiple nodes are all... If the input edge type, output edge type, and return relationship are consistent within the same flotation section, they are merged into a composite node, and multiple output ports are established for the composite node. The output ports correspond to concentrate flow, tailings flow, or middlings flow, respectively. If two middlings sources in the input process data converge into the same roughing node but the intermediate flow direction before merging is not recorded, the virtual edge completion rule generates virtual data edges based on middlings type, process section sequence, and the same loop number. In the initial stage of generation, the virtual data edges do not carry specific values ​​of solids, water, particle size composition, and metal content. Only the start point, end point, edge type, and state to be completed are retained. When the conservation constraints are generated, the edge attribute mapping table assigns vector columns to them. This embodiment enables incomplete process inputs to still form a computable topology and avoids virtual connections from generating unsupported data values ​​before entering the conservation constraints.

[0030] In this embodiment, the node equation template includes a grain size migration sub-template and a grade migration sub-template. The grain size migration sub-template establishes grain size vector constraints based on the input edge grain size mass distribution, node classification variables, and output edge grain size mass distribution. The grade migration sub-template establishes grade vector constraints based on the input edge metal content, target ore grade, yield variables, and output edge metal content. The constraint matrix simultaneously records the grain size vector index and grade vector index for the same output edge, and sets an index association term between the grinding classification node and the flotation separation node, so that the mineral flow vector after grain size change can be used as the input edge vector of subsequent separation nodes. Specifically, the grain size migration sub-template maps multiple grain size components of the input edge to the corresponding grain size column of the output edge, and... The particle size distribution relationship between overflow and sediment is expressed by node-level variables. The grade migration sub-template writes the input edge metal content, input edge target ore grade, and output edge yield variables into the grade migration constraint. The overflow output edge of the grinding and classification node has both particle size index and metal content index in the constraint matrix. When the flotation separation node reads the overflow output edge as the input edge, it does not regenerate a set of independent variables, but reads the same index associated item. If the particle size migration changes the fine-grained component in the overflow edge, the yield distribution and metal content distribution of the flotation separation node use the updated overflow vector as input. This embodiment makes the particle size calculation and grade calculation form a common data reference, avoiding component inconsistencies between the grinding and classification results and the subsequent flotation calculation results.

[0031] In a preferred embodiment, the granular migration constraint is represented in matrix mapping form as follows: ; in, This represents the input edge-level mass distribution vector. Represents the output edge-level quality distribution vector, and represents the particle size transfer matrix corresponding to the node classification variable v. The value can be [0.20, 0.30, 0.50]. If the first row has a value of [0.80, 0.10, 0], then the first particle size component of the output edge is composed of 0.80 parts of the first particle size component of the input edge and 0.10 parts of the second particle size component. The processor reads this value directly when calculating the input of the flotation node and writes the corresponding particle size index into the subsequent grade migration association item. In this embodiment, the change of particle size composition caused by grinding and classification is transmitted to the subsequent separation calculation in matrix form, avoiding the use of a single particle size index to replace the complete particle size distribution.

[0032] Further, refer to Figure 4The residual correction rules include residual source marking rules and key edge selection rules. The residual source marking rules write the solid quantity residual, water quantity residual, particle size residual, and metal quantity residual generated by each node into a residual label table. The residual label table records the node to which the residual belongs, the vector component to which the residual belongs, the associated input edge, and the associated output edge. The key edge selection rules determine data edges that simultaneously associate two or more residual components and are located on closed loops based on the residual label table, and arrange the determined data edges into a return flow correction sequence. The return flow correction sequence is generated according to the loop number, the number of residual components, and the node topological distance. Specifically, the residual source marking rules traverse all node equations after each constraint solution. If a certain floating... If there is an unclosed residual between the input and output metal quantities of a selected node, the node number, metal component number, selected edge number, concentrate edge number, tailings edge number, and middlings edge number are recorded in the residual label table. If the same data edge appears in the associated edge fields corresponding to both the solid quantity residual and the metal quantity residual, and the data edge carries a closed loop number, the critical edge selection rule includes the data edge in the candidate critical edge list. The candidate critical edges are then grouped according to their respective loop numbers, and within the same group, they are arranged according to the number of associated residual components. If the number is the same, they are arranged according to the number of topological edges from the residual node, generating a return flow correction sequence. In this embodiment, the closed loop correction object comes from the residual record itself, reducing the repeated rewriting of non-critical edge variables.

[0033] In this embodiment, the ranking of key edges can be represented by a comprehensive scoring relationship as follows: ; in, This represents the correction priority value of data edge e. This indicates the number of residual components associated with data edge e. This represents the topological distance between data edge e and the residual node. This indicates the loop priority identifier for data edge e. , and These represent the weights of the number of residual components, topological distance, and loop priority identifier, respectively. If a return edge of a certain medium-sized mine is associated with 3 residual components, is 1 edge away from the residual node, and has a loop priority identifier of 1, while a tailings edge is associated with 1 residual component, is 2 edges away from the residual node, and has a loop priority identifier of 0, then the former has a higher correction priority value than the latter. Based on this, the processor places the former at the front of the return stream correction sequence. In this embodiment, a computable score replaces manual judgment of key edges, ensuring that the residual correction order is consistent with the error propagation relationship.

[0034] The comprehensive scoring formula is designed based on three core priority principles for residual correction in closed-loop mineral processing: First, the principle of residual correlation priority: the more residual components a data edge is associated with, the broader its contribution to the overall non-closure of the process. Correcting this edge can simultaneously reduce multi-dimensional residuals such as solids, water, particle size, and metal content, thus it is set as a positive weighting term. Second, the principle of topological proximity priority: the closer a data edge is to the residual node, the stronger its causal relationship with the residual, and the higher the directness of the correction response. Therefore, the inverse distance form is used to reflect the proximity priority sorting logic. Third, the principle of closed-loop affiliation priority: the return flow located on a closed loop is the core regulating variable causing global conservation non-closure and should be corrected before ordinary sequential edges. Therefore, a loop priority indicator is set. By using the weighted summation of the three factors, the multi-dimensional priority criteria are quantified into a single score value, realizing the automatic and objective sorting of the return flow correction order and avoiding the subjective bias of manual judgment of key edges.

[0035] This formula is a dimensionless relative scoring formula, with all quantities maintaining consistent dimensions: In the first term, the number of associated residual components is a dimensionless quantity of count type, with a corresponding weight being a dimensionless adjustment coefficient; the product of the two is a dimensionless value. In the second term, the topological distance is a dimensionless quantity of count type of process edge number; its reciprocal is also dimensionless, with a corresponding weight being a dimensionless adjustment coefficient; the product of the two is a dimensionless value. In the third term, the loop priority identifier is a 0-1 type dimensionless identifier, with a corresponding weight being a dimensionless adjustment coefficient; the product of the two is a dimensionless value. The corrected priority value obtained by summing the three terms is a dimensionless relative score, used only for comparing the corrected priority ranking between different data edges, and does not have the problem of inconsistent dimensions.

[0036] The second term of the formula uses the reciprocal of the topological distance, which means establishing a monotonically negative correlation mapping relationship between "topological distance and correction priority": the topological distance represents the number of mineral flow data edges that separate the data edge from the residual node. The smaller the distance value, the more direct the topological association between the data edge and the location where the residual is generated, the shorter the error propagation path, and the more direct the effect of correcting the data edge on eliminating the current residual and the higher the convergence efficiency. By taking the reciprocal operation, the positive correlation count that originally had a larger value and a greater distance is converted into a score contribution that has a larger value and a higher priority. This allows the topological distance factor to be unified with the other two positive weighting terms into the same score increment logic, and can directly participate in the summation operation and be used for sorting.

[0037] Preferably, the residual path verification instruction includes local closure verification rules, global closure verification rules, and cross-node path tracing rules. The local closure verification rules verify the solid quantity, water quantity, and metal quantity of the input edge vector and output edge vector for each computing node. The global closure verification rules establish a full-process verification equation with the raw ore input edge, concentrate output edge, tailings output edge, and all return flow data edges as boundaries. The cross-node path tracing rules generate an error propagation path based on the adjacency matrix, loop number, and residual label table when the local closure verification passes but the global closure verification fails, and write the return flow data edges in the error propagation path into the set of edges to be verified. In specific implementation, the local closure verification rules take a single computing node as the verification scope, verifying all input edge vectors entering the node and all output edge vectors leaving the node. The quantities are compared by components. If the solid quantity, water quantity, and metal quantity within a node all satisfy the closure relationship, the node state is written as locally closed. Subsequently, the global closure verification rule takes the raw ore input edge as the whole process input boundary, the final concentrate output edge and the final tailings output edge as the whole process output boundary, and all return stream data edges as internal loop edges for verification. When the local closure state exists continuously but the whole process boundary still has unclosed residuals, the cross-node path tracing rule reads the residual node and associated edge from the residual label table, and then uses the adjacency matrix to backtrack to the data edge with the same loop number along the sequential direction and return direction. Finally, the return stream data edge belonging to the error propagation path is written into the edge set to be verified. This embodiment distinguishes between local node calculation errors and cross-node loop errors, so that the subsequent recalculation range can be limited to the path related to global non-closure.

[0038] Furthermore, the return flow correction sequence is configured with an associated constraint locking rule. Before selecting a return flow data edge for variable correction, the associated constraint locking rule reads other data edges that share the same computing node or the same loop number as the selected return flow data edge, and sets the conservation constraints corresponding to the read data edges as adjoint constraints. When the closed-loop return flow solution instruction updates the solid quantity vector components, water quantity vector components, particle size vector components, or metal quantity vector components of the selected return flow data edge, it synchronously updates the same vector index item in the adjoint constraints, and writes the vector components, residual labels, and loop numbers before and after the update into the correction record table. Specifically, if a certain medium-sized mine return edge is corrected by the return flow... Once the sequence is selected, the associated constraint locking rules read the coarse selection node, fine selection node, and concentrate edge, tailings edge, and feed edge connected to the middlings return edge, as well as those under the same closed loop number. The solid quantity constraint, metal quantity constraint, and particle size vector constraint in the relevant node equations are set as adjoint constraints. When the closed-loop return flow solution instruction updates the variable components of the middlings return edge, it not only rewrites the data column of the return edge itself, but also synchronously rewrites the matrix items in the adjoint constraints that reference the same vector index. The correction record table saves the variable values ​​before the update, the variable values ​​after the update, the residual label number, the loop number, and the adjoint constraint number. This embodiment avoids destroying the reference relationship of shared nodes or other constraints within the same loop after individually correcting the return edge variables.

[0039] In a preferred embodiment, the set of edges to be reviewed is configured with a closed-loop recalculation boundary generation rule. This rule generates a local recalculation subgraph using the first residual node, the last residual node, and the return flow data edge between them in the error propagation path as boundaries. The local recalculation subgraph inherits the node equation template, edge attribute mapping table, and loop number from the original flowchart, and masks data edge variables not within the local recalculation subgraph. The closed-loop return flow solution instruction only performs a re-solution on the edge variables and accompanying constraints within the local recalculation subgraph, and writes the re-solution result back to the variable column corresponding to the full-process constraint matrix. Specifically, the closed-loop recalculation boundary generation rule locates the first residual node appearing in the error propagation path. The nodes with labels and the last node with a residual label are included in the subgraph scope, along with the sequential edges between the two residual nodes, the return edges, and the data edges that share the accompanying constraints with the return edges. Nodes and data edges in the original flowchart that did not enter the subgraph remain in a variable-locked state. The local recalculation of the subgraph uses the original node equation template and edge attribute mapping table, so no temporary variables incompatible with the whole process are generated. After the closed-loop return flow solution instruction re-solves the edge variables to be solved in the subgraph, the results are written back to the corresponding columns of the whole process constraint matrix according to the data edge number and vector index, and the local closure check and global closure check are re-checked. This embodiment concentrates the recalculation process within the closed-loop scope corresponding to the error propagation path, avoiding new numerical disturbances caused by repeated solutions in irrelevant process segments.

Claims

1. A mineral processing flow calculation and process design system, characterized in that, Includes a memory, a processor, and a mineral processing flow calculation program stored in the memory and executed by the processor; The mineral processing flow calculation program is configured with instructions for constructing a flow chart, binding multi-dimensional mineral flow attributes, generating conservation constraints, solving closed-loop return flow, and verifying residual paths. The directed flowchart construction instructions convert the crushing, grinding, classification, flotation, cleaning, scavenging and middlings return process steps into computing nodes and data edges. The multidimensional mineral flow attribute binding instruction binds the solids content, water content, particle size distribution, grade, metal content, and circulating load attributes to the data edge; The conservation constraint generation instruction generates material conservation, metal conservation, and water conservation constraints based on the computing node and the data edge; The closed-loop return flow solving instruction uses the return flow as the edge variable to be solved in the whole process constraint solving; The residual path verification instruction locates the cross-node error propagation path based on the node residual and edge residual.

2. The mineral processing flow calculation and process design system according to claim 1, characterized in that, The directed flowchart construction instructions include process segment semantic parsing rules and topology adjacency generation rules. The process segment semantic parsing rules divide the input mineral processing process data into sequential segments, parallel segments, merged segments, and return segments. The topology adjacency generation rules generate an adjacency matrix based on the material flow direction between process segments. The row and column elements of the adjacency matrix correspond to the computing nodes, and the matrix elements carry mineral flow category identifiers, so that the raw ore flow, concentrate flow, tailings flow, middlings flow, return sand flow and slurry water flow have different data edge types in the same flow diagram; The directed flowchart construction instruction also assigns loop numbers to closed loops and writes the loop numbers into the data edges corresponding to the return flow.

3. The mineral processing flow calculation and process design system according to claim 1, characterized in that, The conservation constraint generation instruction includes a node equation template and an edge attribute mapping table. The node equation template sets the input and output variable relationships according to the grinding and classification node, flotation separation node, cleaning node, scavenging node and concentration and dewatering node respectively. The edge attribute mapping table maps solid quantity, water quantity, particle size distribution, target ore grade, gangue grade, and metal quantity to vector components of the same data edge. When generating each node equation, the conservation constraint generation instruction writes the input edge vector, output edge vector, and node transformation variables into the constraint matrix, and retains the column of unknown variables corresponding to the closed-loop return flow in the constraint matrix.

4. The mineral processing flow calculation and process design system according to claim 1, characterized in that, The closed-loop return flow solution instruction includes loop variable initialization rules, hierarchical solution rules, and residual correction rules; The loop variable initialization rule determines the set of edge variables to be solved based on the loop number. The hierarchical solution rule first solves the basic edge variables for sequential links without return flows, and then solves the loop edge variables for closed links with return flows. The residual correction rule decomposes the residual of the whole process into solid quantity residual, water quantity residual, particle size residual and metal quantity residual, and performs targeted correction on the return flow variables according to the data edge type corresponding to the residual, so that the middlings return, sand return and fine tailings return enter different correction branches respectively.

5. The mineral processing flow calculation and process design system according to claim 2, characterized in that, The semantic parsing rules for the process segment also include node role merging rules and virtual edge completion rules; The node role merging rule merges computing nodes that appear repeatedly within the same process segment and have the same connection relationship into composite nodes with multiple output ports, and retains the mineral flow category identifier corresponding to each output port. The virtual edge completion rule generates virtual data edges based on the material type, process segment sequence, and loop number of the preceding and following nodes when there are merging points in the process data but no explicit intermediate flow direction. The virtual data edges only carry topological connection attributes and attribute identifiers to be completed, and the corresponding vector components are filled in by the edge attribute mapping table when the conservation constraints are generated subsequently.

6. The mineral processing flow calculation and process design system according to claim 3, characterized in that, The node equation template includes a grain size migration sub-template and a grade migration sub-template. The grain size migration sub-template establishes a grain size vector constraint according to the input edge grain size mass distribution, node classification variables and output edge grain size mass distribution. The grade migration sub-template establishes a grade vector constraint according to the input edge metal content, target ore grade, yield variables and output edge metal content. The constraint matrix records both the particle size vector index and the grade vector index for the same output edge, and sets an index association item between the grinding and classification node and the flotation separation node, so that the mineral flow vector after the particle size change can be used as the input edge vector of the subsequent separation node.

7. The mineral processing flow calculation and process design system according to claim 4, characterized in that, The residual correction rules include residual source marking rules and key edge selection rules. The residual source marking rules write the solid quantity residual, water quantity residual, particle size residual and metal quantity residual generated by each node into the residual label table. The residual label table records the node to which the residual belongs, the vector component to which the residual belongs, the associated input edge and the associated output edge. The key edge selection rule determines data edges that are simultaneously associated with two or more residual components and are located on closed loops based on the residual label table, and arranges the determined data edges into a return stream correction sequence. The return stream correction sequence is generated according to the loop number, the number of residual components, and the node topological distance.

8. The mineral processing flow calculation and process design system according to claim 4, characterized in that, The residual path verification instructions include local closure verification rules, global closure verification rules, and cross-node path tracing rules; The local closure verification rule verifies the solid quantity, water quantity, and metal quantity of the input and output edge vectors for each computing node. The global closure verification rule establishes a full-process verification equation with the raw ore input edge, concentrate output edge, tailings output edge, and all return stream data edges as boundaries; When the local closure check passes but the global closure check fails, the cross-node path tracing rule generates an error propagation path based on the adjacency matrix, loop number, and residual label table, and writes the return stream data edge in the error propagation path into the edge set to be reviewed.

9. The mineral processing flow calculation and process design system according to claim 7, characterized in that, The return stream correction sequence is configured with an association constraint locking rule. Before selecting a return stream data edge for variable correction, the association constraint locking rule reads other data edges that share the same computing node or the same loop number with the selected return stream data edge, and sets the conservation constraint corresponding to the read data edge as an adjoint constraint. When the closed-loop return flow solution instruction updates the solid quantity vector component, water quantity vector component, particle size vector component, or metal quantity vector component of the selected return flow data edge, it simultaneously updates the same vector index item in the accompanying constraint and writes the vector components before and after the update, residual labels, and loop numbers into the correction record table.

10. The mineral processing flow calculation and process design system according to claim 8, characterized in that, The set of edges to be reviewed is configured with a closed-loop recalculation boundary generation rule. The closed-loop recalculation boundary generation rule uses the first residual node, the last residual node, and the return stream data edge between them in the error propagation path as boundaries to generate a local recalculation subgraph. The local recalculation subgraph inherits the node equation template, edge attribute mapping table, and loop number from the original directed flowchart, and masks data edge variables that are not in the local recalculation subgraph. The closed-loop return flow solution instruction only performs a second solution on the edge variables and associated constraints within the local recalculation subgraph, and writes the second solution result back to the variable column corresponding to the full-process constraint matrix.

Images