Laser cleaning method and system for residual sand of sand casting mold

Abstract

The invention discloses a laser cleaning method and system for residual sand of a sand casting mold, and the method comprises the steps: S1, obtaining point cloud data of the surface of the mold through a three-dimensional scanner, reconstructing a triangular mesh model M (V, E, F) of the surface of the mold, V being a vertex set, E being an edge set, and F being a surface set; s2, based on the triangular mesh model, a laser cleaning path is planned; S3, a laser head is controlled to move along the path, and laser is emitted to remove residual sand; according to the method, the residence time calculation model is introduced, the laser residence time is dynamically set by combining the region area, the material removal difficulty coefficient and the characteristic parameters, the adhesion strength signal of the surface of the mold under the laser action is detected in real time through the acoustic emission sensor, error feedback is formed, the laser power, the pulse frequency and the residence time are dynamically adjusted, and the mold surface quality is improved. And the self-adaptive response to the thickness and the adhesion strength of the residual sand in different areas is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of foundry mold cleaning, in particular to a sand mold foundry mold residual sand laser cleaning method and system. BACKGROUND

[0002] During the use of the foundry mold, a large amount of casting sand is often left in the cavity. If these residual substances are not removed in time, they will affect the forming quality of the subsequent castings. Therefore, foundry mold cleaning is an important part of the casting process. Laser cleaning technology, as a non-contact, high-precision surface treatment technology, has been gradually applied to the field of foundry mold cleaning in recent years. Laser cleaning uses a high-energy-density laser beam to act on the surface of residual sand, causing it to vaporize or peel off instantly. This method has the advantages of high cleaning efficiency, no secondary pollution, and little damage to the mold.

[0003] However, the existing laser cleaning technology still has many problems in actual application: it cannot sense the adhesion strength and cleaning state of the residual sand in real time, which may lead to incomplete cleaning or over-cleaning, affecting the cleaning effect and the safety of the mold. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides the following technical solutions:

[0005] The sand mold foundry mold residual sand laser cleaning method comprises the following steps:

[0006] S1: Obtain the point cloud data of the mold surface by a three-dimensional scanner, and reconstruct a triangular mesh model M(V, E, F) of the mold surface, wherein V is a vertex set, E is an edge set, and F is a face set;

[0007] S2: Plan a laser cleaning path based on the triangular mesh model:

[0008] S3: Control the laser head to move along the path and emit laser to remove residual sand;

[0009] S4: Real-time detection of residual sand adhesion strength, dynamic adjustment of laser power, pulse frequency and cleaning time.

[0010] As an improvement of the above technical solution, the step S2 of planning a laser cleaning path comprises:

[0011] S2.1: Divide the area using a normal vector constrained superpixel segmentation algorithm, and the included angle θ between the vertex normal vector and the average normal vector in the unit is ≤30°;

[0012] S2.2: Extract the unit feature vector f = (h, c, d), where h represents the unit average height, which is used to represent the height position of the region; c represents the unit average curvature, which is used to represent the bending degree of the region; d represents the Hausdorff distance between the point cloud and the CAD model, which is used to predict the residual sand thickness;

[0013] S2.3: Weighted K-means clustering is adopted for the normalized features, and the distance formula is:

[0014]

[0015] where ω1, ω2 and ω3 are weight coefficients, and ω1+ω2+ω3=1, and N cleaning sub-regions are generated.

[0016] As an improvement of the above technical scheme, the reference residence time of each sub-region is calculated in step S3:

[0017]

[0018] where A k is the area of the region, which is obtained by accumulating the area of the triangular facets in the region, η: material cleaning difficulty coefficient, v base is the reference laser scanning speed, h k , c k , d k are the feature vector components of the kth sub-region, respectively.

[0019] As an improvement of the above technical scheme, the area A k of the region is calculated as follows:

[0020] (a) Obtain the triangular facet set F k contained in the sub-region R k ;

[0021] (b) Based on the topological relationship of the vertex set V and the edge set E, calculate the area of each triangular facet f∈F k ,

[0022]

[0023] where V0, V1, V2 are the vertex coordinates associated with the facet f through the edge set E;

[0024] (c) Accumulate the area of all facets:

[0025]

[0026] As an improvement of the above technical scheme, in step S4, the adhesion strength F real is detected in real time by the acoustic emission sensor, and the dynamic adjustment is as follows:

[0027] Laser power P(t) = P0 + alpha u(t)

[0028] Pulse frequency f(t) = f0 + beta u(t)

[0029] Region residence time T k (t) = T 0,k + gamma u(t)

[0030] Wherein

[0031] Error signal e(t) = F real - F traget

[0032] Wherein, F traget is the acoustic emission characteristic calibration value when the mold surface reaches a completely clean state,

[0033] P0 is the laser power, f0 is the pulse frequency, t 0,k is the reference value of the region residence time, alpha is the power adjustment coefficient, beta is the frequency adjustment coefficient, and gamma is the residence time adjustment coefficient.

[0034] The casting mold residual sand laser cleaning system adopts the sand mold casting mold residual sand laser cleaning method described above, comprising:

[0035] A three-dimensional scanner is used to obtain point cloud data of the mold surface.

[0036] A laser execution unit includes a collaborative robot, a laser head support mounted on the execution end thereof, and a pulsed laser head mounted on the laser head support, the collaborative robot drives the laser head to move along the planned path, and the laser head emits pulsed laser to remove residual sand.

[0037] An acoustic emission sensor is mounted on the laser head support for real-time acquisition of mold surface acoustic emission signals and output of adhesion strength quantification value F real .

[0038] A laser execution unit includes a collaborative robot and a laser head fixedly mounted on the execution end of the collaborative robot, the collaborative robot drives the laser head to move along the planned path,

[0039] A dynamic control unit is in communication connection with the acoustic emission sensor and the laser execution unit, for calculating an error signal through the signal output by the acoustic emission sensor, generating a PID control signal, and outputting adjustment instructions of laser power, pulse frequency and region residence time to the laser execution unit.

[0040] The beneficial effects of the present application are:

[0041] ​The mold surface point cloud data is acquired through a three-dimensional scanner and a triangular mesh model is constructed, combined with the normal vector constrained superpixel segmentation and weighted K-means clustering algorithm, the sub-regions are accurately divided and cleaned, the optimal laser cleaning path is planned and the reference residence time is calculated based on the geometric characteristics and residual sand thickness prediction of the sub-regions, and blind cleaning areas and excessive cleaning are avoided;

[0042] The residence time calculation model is introduced, the laser residence time is dynamically set in combination with the area, material removal difficulty coefficient and characteristic parameters, the adhesion strength signal of the mold surface under the action of laser is detected in real time through the acoustic emission sensor, the error feedback is formed, the laser power, pulse frequency and residence time are dynamically adjusted, and adaptive response to the residual sand thickness and adhesion strength of different regions is realized. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 It is a front view of the overall structure of the application.

[0044] The drawings show that: 10 is a collaborative robot; 20 is a laser head; 30 is an acoustic emission sensor. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application and not to limit the application.

[0046] Example 1

[0047] The sand casting mold residual sand laser cleaning method comprises the following steps:

[0048] S1: Acquire the point cloud data of the mold surface through a three-dimensional scanner, and reconstruct the triangular mesh model M(V, E, F) of the mold surface, wherein V is the vertex set, E is the edge set, and F is the face set;

[0049] Using a three-dimensional scanner, based on the principles of optical triangulation, laser ranging, etc., the discrete point cloud data of the mold surface is collected, and the point cloud data completely records the geometric coordinate information of the mold surface;

[0050] Poisson reconstruction, moving least squares method and other grid reconstruction algorithms are used to convert the point cloud data into a triangular mesh model M(V, E, F), wherein the vertex set V stores the three-dimensional coordinates of the discrete vertices of the mold surface; the edge set E defines the connection relationship between the vertices, and constructs the surface topology network; the face set F consists of three vertices to form a triangular facet, which fits the local surface morphology and provides a structured geometric basis for subsequent algorithms.

[0051] S2: Based on the triangular mesh model, the laser cleaning path is planned:

[0052] The laser cleaning path planning includes:

[0053] S2.1: Divide the region using the normal vector constrained superpixel segmentation algorithm, the included angle θ between the vertex normal vector and the average normal vector in the unit is ≤30°;

[0054] According to the normal vector information of the triangular mesh model, the mold surface is divided into superpixel units with similar normal vector directions, which ensures the continuity of the surface orientation in the unit and provides a continuous basic area for path planning. The normal vector is calculated by cross multiplication of the vertex coordinates of the patch (such as patch vertex V0, V1, V2, normal vector n=(V1-V0)×(V2-V0)), and the continuity of the unit is realized by using the adjacency relationship and the angle constraint of the normal vector.

[0055] The normal vector constrained superpixel segmentation algorithm is executed according to the following steps:

[0056] S2.11, initialize the cluster set, each cluster contains a randomly selected triangular patch, and its normal vector is used as the initial cluster average normal vector;

[0057] S2.12, traverse all unassigned triangular patches:

[0058] a. Based on the adjacency relationship defined by the edge set E, find the clusters adjacent to the current patch;

[0059] b. Calculate the included angle between the normal vector of the current patch and the average normal vector of the adjacent cluster;

[0060] c. If there is a cluster with an included angle θ ≤30°, assign the patch to the cluster with the smallest included angle;

[0061] d. If there is no cluster that meets the condition, create a new cluster with the current patch;

[0062] S2.13, update the average normal vector of each cluster;

[0063] S2.14, repeat steps 2-3 until the convergence condition is met: more than 90% of the patch normal vectors have an included angle θ ≤30° with the average normal vector of the cluster to which they belong;

[0064] S2.15, output the final superpixel unit set, each unit contains adjacent triangular patches with similar normal vector directions.

[0065] S2.2: Extract the unit feature vector f=(h, c, d), where h represents the average height of the unit, which is used to represent the height position of the region; c represents the average curvature of the unit, which is used to represent the bending degree of the region; d represents the Hausdorff distance between the point cloud and the CAD model, which is used to predict the residual sand thickness;

[0066] Multi-dimensional features are extracted from the superpixel units to depict the regional geometric position, surface morphology and residual sand thickness information, thereby providing a basis for cluster division.

[0067] S2.3: Weighted K-means clustering is adopted for the normalized features, and the distance formula is:

[0068]

[0069] where ω1, ω2 and ω3 are weight coefficients, and ω1+ω2+ω3=1, and N clean sub-regions are generated.

[0070] Feature normalization: The feature vector f=(h, c, d) of each superpixel unit is normalized to eliminate the dimensional influence.

[0071] Initialization: N points are randomly selected as initial cluster centers,

[0072] Iteration process:

[0073] a. Assignment step: for each superpixel unit, the weighted Euclidean distance between it and each cluster center is calculated, and then the unit is assigned to the nearest cluster center;

[0074] b. Update step: the center of each cluster is recalculated,

[0075] Termination condition: when the maximum number of iterations is reached, the iteration is stopped;

[0076] Output: N clusters are finally obtained, each of which is a clean sub-region.

[0077] When generating N clean sub-regions, the region continuity is mainly guaranteed by the adjacency constraint of superpixel segmentation and the spatial correlation rules of clustering algorithm. The following is a technical logic disassembly:

[0078] Superpixel segmentation stage: the basic continuity is ensured by the "adjacency relationship";

[0079] Superpixel segmentation (S2.1) is the "bottom framework" of sub-region continuity, and the core is to make the units naturally continuous through "adjacency constraint of edge set E + normal vector similarity":

[0080] Adjacency binding: in the triangular mesh model, the edge set E defines the "adjacent topology" of the facets (each facet is connected to the surrounding facets through edges), and when segmenting, step S2.12 "finds adjacent clusters based on edge set E" to force the facet to be assigned to the "spatially adjacent" cluster. For example, if the adjacent facets of facet A all belong to cluster X, A is preferentially assigned to X to avoid "cross-space jumping" assignment, thereby fundamentally ensuring the spatial adjacency of the units;

[0081] Normal vector similarity constraint: the segmentation condition "the angle θ between the vertex normal vector and the average normal vector within a unit is less than or equal to 30°, which ensures that the surface within the unit is approximately oriented and has a gentle curvature. This "geometric continuity" makes the super-pixel unit naturally continuous in space (if the surface is suddenly changed, the angle between the normal vectors will exceed the threshold, and the unit will be automatically divided into a new unit);

[0082] The super-pixel unit has basic continuity, and the weighted K-means clustering (S2.3) further maintains regional continuity through "feature correlation + weight bias":

[0083] Spatial correlation of feature vectors: in the feature vector f=(h, c, d), the average height h and the average curvature c are themselves spatially continuous features (the height and curvature of adjacent regions will not change suddenly). During clustering, the similarity of these features will naturally cause "spatially adjacent and similar features" to be clustered into a class;

[0084] Implicit constraints of weight design: in the distance formula, ω1 (height weight), ω2 (curvature weight) are smaller than ω3 (residual sand thickness weight), but they will still punish units with "sudden spatial feature changes". For example, if the residual sand thickness of two units is similar, but the height difference is extremely large (h i- h j ), the weighted distance will increase, reducing the probability of clustering into a class, and indirectly ensuring spatial continuity;

[0085] The final N cleaning sub-regions can not only ensure the homogeneity of residual sand and geometric features, but also maintain spatial continuity, providing "continuous and controllable, feature uniform" cleaning partitions for laser path planning.

[0086] S3: Control the laser head to move along the path and emit laser to remove residual sand, and calculate the reference residence time of each sub-region:

[0087]

[0088] where A k is the area of the region, obtained by accumulating the area of the triangular facets within the region, η: material cleaning difficulty coefficient, v base is the reference laser scanning speed, h k , c k , d k are the feature vector components of the kth sub-region, respectively.

[0089] The calculation of the area of the region A k includes:

[0090] (a) Obtain the triangular facet set F k contained in the sub-region R k ;

[0091] (b) based on the topological relationship of vertex set V and edge set E, for each triangular facet f e F k Calculate the area,

[0092]

[0093] wherein V0, V1, V2 are the vertex coordinates associated with the facet f through the edge set E;

[0094] (c) accumulate all the facet area:

[0095]

[0096] Surface area A k The greater, the longer the time required.

[0097] The higher the eigenvalue (h k , c k , d k ), the thicker the residual sand, the greater the curvature, and the deeper the location, the longer the residence time required,

[0098] S4: Real-time detection of residual sand adhesion strength, through dynamic adjustment of laser power, pulse frequency and cleaning time;

[0099] Real-time detection of adhesion strength F real by acoustic emission sensor, through dynamic adjustment:

[0100] Laser power P(t) = P0+au(t);

[0101] Pulse frequency f(t) = f0+βu(t);

[0102] Regional residence time T k (t) = T 0,k +γu(t);

[0103] Wherein

[0104] Error signal e(t) = F real -F traget ;

[0105] Wherein F traget is the acoustic emission eigenvalue of the mold surface when it reaches complete cleaning state,

[0106] P0 is the laser power, f0 is the pulse frequency, T 0,k is the baseline value of the regional residence time, a is the power adjustment coefficient, β is the frequency adjustment coefficient, and γ is the residence time adjustment coefficient,

[0107] Real-time response to residual sand removal state: use acoustic emission sensor to collect acoustic emission signal when residual sand is stripped, compare with mold cleaning state calibration value, get error signal, reflect the deviation between actual state and target state of residual sand cleaning, if residual sand is not removed, F real >F traget , then increase power, frequency or residence time; if over-cleaning (F real <F traget ), then reduce parameters, ensure the balance of efficient sand removal + mold protection.

[0108] Through the closed-loop system of "three-dimensional modeling → regional intelligent segmentation → feature clustering → parameter planning → real-time feedback", the precise laser cleaning of mold residual sand is realized. Based on the geometric features of the mold surface, combined with the physical properties (thickness, adhesion strength) of residual sand, the cleaning path and initial parameters are planned through algorithm, and then real-time monitoring is used for dynamic correction, to ensure efficient sand removal and protection of the mold.

[0109] Example two

[0110] The casting mold residual sand laser cleaning system adopts the sand mold casting mold residual sand laser cleaning method recorded in example one, comprising:

[0111] A three-dimensional scanner is used to obtain point cloud data of the mold surface.

[0112] The laser execution unit includes a collaborative robot 10, a laser head bracket installed at the execution end of the collaborative robot 10, and a pulsed laser head 20 installed on the laser head bracket. The collaborative robot 10 drives the laser head 20 to move along the planned path, and the laser head 20 emits pulsed laser to remove residual sand.

[0113] An acoustic emission sensor 30 is installed on the laser head bracket, which is used to collect acoustic emission signals from the mold surface in real time and output adhesion strength quantification value F real ;

[0114] The laser execution unit includes a collaborative robot 10 and a laser head 20 fixedly installed on the execution end of the collaborative robot 10. The collaborative robot 10 drives the laser head 20 to move along the planned path,

[0115] The dynamic control unit is in communication connection with the acoustic emission sensor 30 and the laser execution unit. It is used to calculate the error signal through the signal output by the acoustic emission sensor 30, generate the PID control signal, and output the adjustment instruction of laser power, pulse frequency and regional residence time to the laser execution unit.

[0116] The system is composed of a three-dimensional scanner (perception layer), a dynamic control unit (decision layer), a laser execution unit (execution layer), and an acoustic emission sensor 30 (feedback layer). The core logic is:

[0117] The perception layer: a three-dimensional scanner acquires a digital model of the mold;

[0118] The decision layer: a dynamic control unit runs an embodiment algorithm, plans a path, calculates parameters, and generates adjustment instructions;

[0119] The execution layer: a laser execution unit executes cleaning according to the instructions;

[0120] The feedback layer: acoustic emission sensors monitor in real time and provide correction basis for the decision layer, forming a closed-loop control.

[0121] The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them.

Claims

1. A sand mold casting mold residual sand laser cleaning method characterized by, The method comprises the following steps: S1: obtaining the point cloud data of the mold surface by a three-dimensional scanner, and reconstructing a triangular mesh model M(V, E, F) of the mold surface, wherein V is a vertex set, E is an edge set, and F is a face set; S2: planning a laser cleaning path based on the triangular mesh model; S3: controlling the laser head to move along the path and emit laser to remove residual sand; S4: detecting the residual sand adhesion strength in real time, and adjusting the laser power, pulse frequency and cleaning time dynamically.

2. The sand casting mold sand laser cleaning method and system of claim 1, wherein: The step S2 of planning the laser cleaning path comprises: S2.1: dividing the region by using a normal vector constrained superpixel segmentation algorithm, and the included angle θ between the vertex normal vector and the average normal vector in the unit is ≤30°; S2.2: extracting a unit feature vector f=(h, c, d), wherein h represents the average height of the unit, which is used to represent the height position of the region; c represents the average curvature of the unit, which is used to represent the bending degree of the region; and d represents the Hausdorff distance between the point cloud and the CAD model, which is used to predict the residual sand thickness; S2.3: performing weighted K-means clustering on the normalized features, and the distance formula is: wherein ω1, ω2 and ω3 are weight coefficients, and ω1+ω2+ω3=1, and N cleaning sub-regions are generated.

3. The sand- mold casting-mold residual sand laser cleaning method according to claim 2, characterized by: In step S3, the reference residence time of each sub-region is calculated: Wherein, A k is the area surface, obtained by accumulating the area of triangular facets within the region, η: material cleaning difficulty coefficient, v base is the reference laser scanning speed, h k ,c k ,d k are the feature vector components of the kth sub-region respectively.

4. The sand- mold casting-mold residual sand laser cleaning method according to claim 3, characterized by: The area surface A of the region k The calculation of A includes: (a) obtaining a sub-region R k a set of triangles F comprising k ; (b) based on the topological relations of the vertex set V and the edge set E, for each triangular facet f e F k calculating the area, wherein V0, V1 and V2 are the vertex coordinates associated by the edge set E of the face f; (c) accumulating the areas of all face pieces: 。 5. The sand casting mold sand laser cleaning method of claim 1, wherein: In step S4 the adhesion strength F is detected in real time by the acoustic emission sensor real by dynamic adjustment: Laser power P(t)=P0+αu(t); Pulse frequency f(t)=f0+βu(t); Zone residence time T k (t) = T 0,k + γu(t); wherein Error signal e(t) = F real - F traget ; wherein F traget is the acoustic emission signature value for the mold surface to reach a fully clean condition, P0 is the laser power, f0 is the pulse frequency, T 0,k is the reference value of the zone residence time, a is the power adjustment coefficient, β is the frequency adjustment coefficient, and γ is the residence time adjustment coefficient.

6. A laser cleaning system for foundry mold sand residue, characterized by: The method for laser cleaning of residual sand of a sand mold casting mold according to any one of claims 1-5 comprises: a three-dimensional scanner for obtaining point cloud data of the mold surface; a laser execution unit comprising a collaborative robot, a laser head bracket mounted on the execution end of the collaborative robot, and a pulse laser head mounted on the laser head bracket, wherein the collaborative robot drives the laser head to move along the planned path, and the laser head emits pulse laser to remove residual sand; An acoustic emission sensor is installed on the laser head support to collect acoustic emission signals on the mold surface in real time and output a quantitative value F of the adhesion strength real ; a laser execution unit comprising a collaborative robot and a laser head fixedly mounted on the execution end of the collaborative robot, wherein the collaborative robot drives the laser head to move along the planned path, a dynamic control unit in communication connection with the acoustic emission sensor and the laser execution unit, for calculating an error signal through the signal output by the acoustic emission sensor, generating a PID control signal, and outputting adjustment instructions of the laser power, pulse frequency and region residence time to the laser execution unit.

Images