An automatic detection method for casting quality in water glass sand casting process

Abstract

The present invention relates to the technical field of casting quality detection, and in particular to an automatic detection method for casting quality in a water glass sand casting process. The method comprises: obtaining point cloud data of the casting surface in the water glass sand casting process; obtaining data distribution characteristic values ​​of each similar point set according to the relative position distribution of the point cloud data in each similar point set and the distribution of the point cloud data in the neighborhood of each point cloud data; determining the noise evaluation value of each similar point set in combination with the clustered distribution characteristics between the point cloud data in the similar point set and the discrete distribution characteristics between different similar point sets; obtaining the filtering radius corresponding to each similar point set according to the data distribution characteristic value and the noise evaluation value; filtering the corresponding similar point set using the filtering radius, and evaluating the quality of the casting based on the point cloud data after filtering. The present invention ensures the filtering effect of the point cloud data on the casting surface and improves the accuracy of the detection result of the casting quality.

Description

Technical Field

[0001] The invention relates to the technical field of casting quality detection, and in particular to an automatic casting quality detection method for a water glass sand mold casting process. Background Art

[0002] The water glass sand casting process is widely used in precision casting and the manufacture of complex castings. It uses water glass as a binder and sand as aggregate to produce high-precision castings. It is particularly suitable for casting high-strength, high-temperature resistant and complex-shaped metal parts. However, in the water glass sand casting process, due to the complexity and variability of the casting mold and the large size of the casting, it is easy to cause subtle defects, dimensional deviations and shape errors on the surface of the casting. These errors often affect the final quality of the casting, and then affect the reliability and economy of the entire production process. With the development of laser scanning technology, laser measurement technology has been widely used in the automated detection of castings. The laser scanner can quickly and non-destructively collect three-dimensional data of the casting surface. These data represent the geometric shape of the casting in the form of point clouds. The use of these three-dimensional data can realize the detection of casting quality.

[0003] In the actual inspection process, due to the complex surface, changeable shape and environmental factors of the casting, the laser scanning data is often affected by noise, and the collected point cloud data cannot fully reflect the true geometric characteristics of the casting, which in turn affects the subsequent quality inspection results. Therefore, before the casting quality inspection, the laser measurement data of the casting needs to be filtered; the laser measurement data of the casting is a dense point cloud distribution, and the existing radius filtering algorithm can better adapt to this local dense point cloud distribution. By adjusting the radius size and the number of point thresholds, noise and outliers can be effectively removed. However, the effect of radius filtering mainly depends on the setting of the filter radius. The casting surface often contains a variety of different structures and features. These different structures present different geometric forms and distribution characteristics in the point cloud data. The noise level and distribution characteristics of different point cloud data may also be different. Therefore, the use of the global filter radius cannot adapt to this difference, resulting in poor filtering effect in some areas, and excessive filtering in other areas, which in turn affects the accuracy of subsequent casting quality assessment. Summary of the invention

[0004] In order to solve the problem that the existing filtering method using the global filtering radius has poor filtering effect when filtering the point cloud data of the casting, thereby affecting the accuracy of the casting quality assessment result, the purpose of the present invention is to provide a casting quality automatic detection method for the water glass sand casting process, and the technical scheme adopted is as follows:

[0005] The present invention provides a method for automatically detecting casting quality in a water glass sand casting process, the method comprising the following steps:

[0006] Obtain point cloud data of the casting surface in the water glass sand casting process;

[0007] According to the relative position distribution of the point cloud data in each similar point set and the distribution of the point cloud data in the neighborhood of each point cloud data, the data distribution characteristic value of each similar point set is obtained; the similar point set is obtained by dividing all the point cloud data based on the density distribution of the point cloud data;

[0008] Determine the noise evaluation value of each similar point set by combining the clustered distribution characteristics between point cloud data in the similar point set and the discrete distribution characteristics between different similar point sets; obtain the filter radius corresponding to each similar point set according to the data distribution characteristic value and the noise evaluation value;

[0009] The corresponding similar point set is filtered using the filtering radius, and the quality of the casting is evaluated based on the point cloud data after filtering.

[0010] Preferably, the acquisition of the similar point set includes:

[0011] The DBSCAN clustering algorithm is used to cluster all point cloud data into several clusters, and all point cloud data in each cluster are regarded as a similar point set.

[0012] Preferably, the data distribution characteristic value of each similar point set is obtained according to the relative position distribution of the point cloud data in each similar point set and the distribution of the point cloud data in the neighborhood of each point cloud data, including:

[0013] Calculate the average value of the horizontal coordinate, the average value of the vertical coordinate and the average value of the vertical coordinate of all the point cloud data in the candidate set respectively;

[0014] The position formed by the average value of the horizontal coordinate, the average value of the vertical coordinate and the average value of the vertical coordinate is recorded as the center position; according to the relative position between all the point cloud data in the candidate set and the center position, the sparse factor of the candidate set is obtained;

[0015] For any point cloud data in the candidate set, principal component analysis is performed on the point cloud data and all point cloud data in its neighborhood to obtain a feature vector corresponding to the maximum eigenvalue as the feature vector corresponding to the point cloud data;

[0016] According to the degree of fluctuation of the modulus of the feature vector corresponding to the point cloud data in the candidate set and the sparsity factor, the data distribution characteristic value of the candidate set is obtained, and both the degree of fluctuation and the sparsity factor are negatively correlated with the data distribution characteristic value;

[0017] The candidate set is any similar point set.

[0018] Preferably, obtaining the sparse factor of the candidate set according to the relative positions between all point cloud data in the candidate set and the central position includes:

[0019] The average value of the Euclidean distances between all point cloud data in the candidate set and the central position is determined as the sparse factor of the candidate set.

[0020] Preferably, obtaining the degree of fluctuation of the modulus length of the feature vector corresponding to the point cloud data in the candidate set includes:

[0021] The variance of the modulus of the feature vectors corresponding to all point cloud data in the candidate set is used as the fluctuation degree.

[0022] Preferably, the step of combining the clustered distribution characteristics between point cloud data in the similar point set and the discrete distribution characteristics between different similar point sets to determine the noise evaluation value of each similar point set includes:

[0023] Based on the clustering distribution characteristics between the point cloud data in the candidate set, the cohesion of the candidate set is determined;

[0024] Based on the discrete distribution characteristics between the candidate set and its adjacent similar point set, the separation degree corresponding to the candidate set is determined;

[0025] The noise evaluation value of the candidate set is obtained by comprehensively considering the cohesion and the separation, and both the cohesion and the separation are negatively correlated with the noise evaluation value.

[0026] Preferably, determining the cohesion of the candidate set based on the clustered distribution characteristics between the point cloud data in the candidate set includes:

[0027] The negatively correlated normalized value of the sparse factor of the candidate set is determined as the cohesion of the candidate set.

[0028] Preferably, determining the separation degree corresponding to the candidate set based on the discrete distribution characteristics between the candidate set and its adjacent similar point set includes:

[0029] The cumulative sum of the Euclidean distances between the center position of the candidate set and the center positions of all its neighboring similar point sets is determined as the separation degree corresponding to the candidate set.

[0030] Preferably, obtaining the filter radius corresponding to each similar point set according to the data distribution characteristic value and the noise evaluation value includes:

[0031] Calculate the result of rounding up the ratio between the noise evaluation value of the candidate set and the data distribution characteristic value of the candidate set;

[0032] The product of the rounded-up result and the preset filtering radius is determined as the filtering radius corresponding to the candidate set.

[0033] Preferably, the evaluating the quality of the casting based on the filtered point cloud data includes:

[0034] All filtered point cloud data are input into the trained neural network to obtain the quality assessment results of the casting.

[0035] The present invention has at least the following beneficial effects:

[0036] The present invention first divides all point cloud data into multiple similar point sets based on the density distribution of point cloud data on the surface of a casting in a water glass sand casting process. Similar data in the point cloud data on the surface of a casting usually have similar geometric structures and distribution characteristics. By classifying similar data into a subset and performing unified filtering processing on the subset, the processing time can be significantly reduced, and it can also be ensured that they are subjected to the same processing, thereby maintaining data consistency. All point cloud data in the same similar point set use the same filtering radius, and then according to the relative position distribution of point cloud data in each similar point set and the distribution of point cloud data in the neighborhood of each point cloud data, the data distribution characteristic value of each similar point set is obtained. The data distribution characteristic value is used to reflect the sparsity and regularity of the distribution of point cloud data in the similar point set. For the same structure of a casting, the corresponding point cloud data distribution should be more clustered and more regular. Furthermore, combined with the clustered distribution characteristics between point cloud data in a similar point set and the discrete distribution characteristics between different similar point sets, the noise of each similar point set is filtered. The noise level is evaluated and the noise evaluation value is obtained. In order to ensure the filtering effect of the point cloud data, the larger the noise evaluation value, the larger the filter radius should be selected to remove the noise. The more obvious the distribution characteristics of the point cloud data in the similar point set, the greater the possibility that the point cloud data in the similar point set belongs to the same design structure, and a smaller filter radius needs to be selected for fine processing. Therefore, the filter radius corresponding to the similar point set is determined in combination with the data distribution characteristic value and the noise evaluation value. Each similar point set corresponds to a filter radius. According to the filter radius corresponding to each similar point set, all point cloud data in the same similar point set are uniformly filtered. This avoids the problem of poor filtering effect in some areas and possible over-filtering in other areas caused by the global use of the same filter radius for filtering, and avoids the problem of large consumption of computing resources caused by the use of one filter radius for one point cloud data, ensuring the filtering effect of the point cloud data on the surface of the casting. Finally, the point cloud data after filtering is used to evaluate the quality of the casting, thereby improving the accuracy of the evaluation results of the casting quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A flow chart of a method for automatically detecting casting quality in a water glass sand casting process provided by an embodiment of the present invention;

[0039] Figure 2 The present invention provides a structural block diagram of an automatic casting quality detection system for a water glass sand casting process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of an automatic detection method for casting quality for water glass sand casting process proposed by the present invention in combination with the accompanying drawings and preferred embodiments.

[0041] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0042] The specific scheme of the automatic detection method for casting quality of water glass sand casting process provided by the present invention is described in detail below with reference to the accompanying drawings.

[0043] An automatic detection method for casting quality in water glass sand casting process:

[0044] The specific scenario targeted by this embodiment is: when the quality of castings produced by a water glass sand casting process is inspected based on a machine vision method, a laser is used to scan the surface of the casting to determine the geometric shape of the casting. However, due to the complexity of the casting surface, the changeable shape and the influence of environmental factors, the laser scanning data will be affected by noise. Therefore, it is necessary to filter the collected point cloud data. When using a radius filtering algorithm for filtering, different point cloud data are filtered using different filtering radii in combination with the characteristics of different point cloud data to ensure the filtering effect and improve the casting quality inspection results.

[0045] This embodiment proposes a casting quality automatic detection method for water glass sand casting process, such as Figure 1 As shown, a method for automatically detecting casting quality in a water glass sand casting process according to this embodiment includes the following steps:

[0046] Step S1, obtaining point cloud data of the casting surface of the water glass sand casting process.

[0047] Aim the laser scanner at the casting surface. The laser scanner measures the position of each point on the casting surface by emitting a laser beam and receiving the reflected signal. That is, the laser scanner calculates the distance between each point on the casting surface and the laser scanner by emitting a laser beam and measuring the time it takes for it to be reflected back. Then, point cloud data is generated based on the measured distance data and the corresponding angle. Each point represents a reflection point of each point on the casting surface, including the spatial coordinates of the point.

[0048] So far, this embodiment has acquired the point cloud data of the casting surface of the water glass sand casting process.

[0049] Step S2, obtaining the data distribution characteristic value of each similar point set according to the relative position distribution of the point cloud data in each similar point set and the distribution of the point cloud data in the neighborhood of each point cloud data; the similar point set is obtained by dividing all the point cloud data based on the density distribution of the point cloud data.

[0050] Similar data in the point cloud data of the casting surface often come from the same design structure. For example, the design of the casting may form geometric structures such as serrated and trapezoidal shapes on the surface. This design is usually to improve strength or reduce material usage. Each serrated or trapezoidal structure on the casting surface has the same shape and size. These structures have high similarity in terms of geometric shape, spatial distribution, and noise characteristics presented in the point cloud data. If the same filter radius is used to filter similar data in the point cloud data of the casting surface, it can avoid the problem of poor filtering effect in some areas caused by the global use of the same filter radius for filtering, and the problem of excessive filtering in other areas can be avoided. It also avoids the problem of large consumption of computing resources caused by one filter radius for one point cloud data. Therefore, the point cloud data will be classified first, and the data with higher similarity will be divided together to facilitate the subsequent determination of the filter radius.

[0051] Specifically, the DBSCAN clustering algorithm is used to cluster all point cloud data to obtain multiple clusters, and all point cloud data in each cluster is regarded as a similar point set, that is, multiple clusters are obtained. The DBSCAN clustering algorithm is a prior art and will not be described in detail here.

[0052] In the point cloud data of the casting surface, similar point sets usually correspond to specific design structure areas in the casting surface. These structures have high similarity in terms of geometric shape, spatial distribution, and noise characteristics in the point cloud data. By analyzing the distribution characteristics of the point cloud data, it can provide an important basis for determining the filter radius. The level of noise can directly affect the selection of the filter radius. In areas with high noise levels, a larger filter radius needs to be selected to remove noise; while in areas with low noise levels, a smaller filter radius can be selected to retain more point cloud information. Therefore, when obtaining the filter radius corresponding to the similar point set in the point cloud data of the casting surface, it can be analyzed from the distribution characteristics of the point cloud data in the similar point set and the noise level.

[0053] When obtaining the distribution characteristics of point cloud data in the similar point set, the position information of the point cloud data can be used for analysis. The more regular the position distribution of the point cloud data in the point cloud similar point set and the lower the sparsity, the greater the possibility that this part of the point cloud data belongs to the same design structure. This provides strong support for the acquisition of the filter radius in the similar point set. If the point cloud data in the similar point set belong to the same design structure, the filter radius needs to be smaller to provide fine filtering.

[0054] The lower the distribution sparsity of the point cloud data in the similar point set, the more likely it is that the point cloud data in the similar point set belongs to the same design structure in the water glass sand casting; because water glass sand castings usually contain multiple design structures or functional areas, the geometric features of each area will be highly consistent during the manufacturing process. If the point cloud data shows a low sparsity feature in some areas, it means that the area may correspond to a continuous and regular design part in the casting. On the contrary, areas with high sparsity mean that the data points are more scattered and lack similarity. Therefore, in water glass sand castings, point cloud data with higher sparsity often means that they are more likely to come from different design structures rather than the same design structure.

[0055] Based on the above characteristics, the sparsity of the similar point set is analyzed according to the relative position distribution of the point cloud data in the similar point set to determine the sparse factor of the similar point set.

[0056] This embodiment is described by taking a similar point set as an example, and other similar point sets can be processed by the method provided in this embodiment.

[0057] Specifically, any similar point set is recorded as a candidate set, and the average values ​​of the horizontal coordinates, the average values ​​of the vertical coordinates, and the average values ​​of the vertical coordinates of all the point cloud data in the candidate set are calculated respectively; the position formed by the average values ​​of the horizontal coordinates, the average values ​​of the vertical coordinates, and the average values ​​of the vertical coordinates is recorded as the center position of the candidate set; and the average value of the Euclidean distances between all the point cloud data in the candidate set and the center position of the candidate set is determined as the sparse factor of the candidate set.

[0058] In water glass sand castings, parts with more regular and symmetrical structures usually produce more regular point cloud data. Such point cloud data are more likely to come from the same design structure, and these point cloud data have higher regularity.

[0059] For any point cloud data in the candidate set, principal component analysis (PCA) is performed on the point cloud data and all point cloud data in its neighborhood to obtain the eigenvector corresponding to the largest eigenvalue as the eigenvector corresponding to the point cloud data. By using this method, the eigenvector corresponding to each point cloud data in the candidate set can be obtained. Principal component analysis is a prior art and will not be described in detail here.

[0060] The modulus of the eigenvector reflects the surface characteristics and local geometric shape of the point cloud data. When the point cloud data has good regularity, the surface should be smooth and regular, and the modulus length of the eigenvector should be relatively consistent. The regularity of the point cloud data can be quantified by calculating the fluctuation degree of the modulus length of the eigenvector. A smaller fluctuation degree indicates that the casting surface is more regular; a larger fluctuation degree indicates that the surface is more irregular.

[0061] When the concentration of similar points in the point cloud data is high, it means that the distribution of the point cloud data is more orderly and regular, which indicates that the collected data comes from parts with similar geometric shapes. The production process of water glass sand castings usually involves the precise design and manufacture of specific shapes and structures. Therefore, if the points in the point cloud data are clustered together, it means that these points represent parts of the same design structure. Specifically, as a common casting process, the design structure of water glass sand castings generally has a certain degree of symmetry or repeatability. For example, some designs may have multiple similar holes, edges or curved parts. When these similar structures form concentrated and regular point sets in the point cloud data, the concentrated existence of these point sets indicates that they belong to a part of the same whole; on the contrary, if the point cloud data is sparse and irregular, it means that these data are from parts of different structures.

[0062] Based on the above characteristics, combined with the fluctuation degree and sparse factor of the modulus length of the feature vector corresponding to the point cloud data in the candidate set, the data distribution characteristic value of the candidate set is obtained, and the fluctuation degree and the sparse factor are negatively correlated with the data distribution characteristic value.

[0063] Among them, the negative correlation relationship means that the dependent variable will decrease as the independent variable increases, and the dependent variable will increase as the independent variable decreases. It can be a subtraction relationship, a division relationship, etc., which is determined by practical applications.

[0064] In this embodiment, the variance of the modulus of the feature vectors corresponding to all the point cloud data in the candidate set is used as the fluctuation degree of the modulus of the feature vectors corresponding to the point cloud data in the candidate set.

[0065] In this embodiment, a specific calculation formula for the data distribution characteristic value is given, and the data distribution characteristic value of the j-th similar point set can be expressed as:

[0066]

[0067] in, represents the data distribution characteristic value of the jth similar point set, Represents the variance of the modulus length of the eigenvector corresponding to all point cloud data in the jth similar point set, represents the number of point cloud data in the jth similar point set, represents the position of the t-th point cloud data in the j-th similar point set, represents the center position of the jth similar point set, represents the Euclidean distance between the t-th point cloud data in the j-th similar point set and the center position of the j-th similar point set, represents the preset first adjustment parameter, and norm() represents the normalization function.

[0068] In this embodiment, a preset first adjustment parameter is introduced into the calculation formula of the data distribution characteristic value to prevent the denominator from being 0. In this embodiment, the preset adjustment parameter is 0.01. In specific applications, the implementer can set it according to the specific situation.

[0069] It represents the average value of the Euclidean distance between all point cloud data in the jth similar point set and the center position of the jth similar point set, that is, the sparse factor of the jth similar point set, which is used to reflect the sparsity of the distribution of point cloud data in the jth similar point set. Therefore, the larger the sparse factor of the jth similar point set, the smaller the possibility that the point cloud data in the jth similar point set belongs to the same design structure in the water glass sand mold casting. The smaller the variance of the modulus length of the feature vector of all point cloud data in the jth similar point set, the higher the consistency of the modulus length of the feature vector of all point cloud data in the jth similar point set, the more regular the distribution of the point cloud data in the jth similar point set, and the greater the possibility that the point cloud data in the jth similar point set belongs to the same design structure in the water glass sand mold casting.

[0070] The higher the regularity and the smaller the sparsity of the point cloud data in the jth similar point set, the greater the possibility that the point cloud data in the jth similar point set belongs to the same design structure, and the more obvious the distribution characteristics are, that is, the larger the data distribution characteristic value of the jth similar point set.

[0071] By adopting the above method, the data distribution characteristic value of each similar point set can be obtained.

[0072] Step S3, combining the clustered distribution characteristics between the point cloud data in the similar point set and the discrete distribution characteristics between different similar point sets, determining the noise evaluation value of each similar point set; and obtaining the filter radius corresponding to each similar point set according to the data distribution characteristic value and the noise evaluation value.

[0073] Since the similar point set is the clustering result of the point cloud data of the casting surface collected by the laser scanner, the noise points in the point cloud data of the casting surface may come from the errors of the laser scanner, environmental interference (such as light, reflection), irregularities of the casting surface, etc. These noise points will have a certain impact on the clustering effect; therefore, when evaluating the noise level in the similar point set, it can be indirectly reflected through the clustering effect of the corresponding cluster of the similar point set. The better the clustering effect, the lower the noise level of the similar point set.

[0074] Water glass sand castings are usually composed of different geometric structures, and these structures often follow certain rules or patterns when they are designed. If there are multiple points in the point cloud data that are clustered together to form a high cohesion cluster, then the area represented by these points is likely to be in a part of the object with similar geometric features, such as similar surfaces, holes or edges. These features are usually part of the same design structure, because the manufacture of water glass sand castings is often made according to design drawings, so there is regularity and consistency.

[0075] Secondly, water glass sand castings usually have relatively fixed shapes and geometric features. If the point cloud data come from the same design structure, they should show a high degree of similarity in geometric properties, shape, size, etc. The clustering algorithm will cluster these point cloud data into the same cluster. If the differences between different clusters are large, this indicates that different structures or design features are clearly distinguished in the point cloud data, which means that each cluster corresponds to a different design structure or mold part, and the design structure is very different from that of other clusters, that is, the greater the separation between the corresponding cluster and its surrounding clusters.

[0076] Therefore, this embodiment evaluates the noise level by combining the clustered distribution characteristics between point cloud data in a similar point set and the discrete distribution characteristics between different similar point sets to obtain the noise evaluation value of the similar point set.

[0077] Next, the candidate set is still taken as an example for explanation. Specifically, based on the clustered distribution characteristics between the point cloud data in the candidate set, the cohesion of the candidate set is determined; based on the discrete distribution characteristics between the candidate set and its adjacent similar point set, the separation degree corresponding to the candidate set is determined; and the noise evaluation value of the candidate set is obtained by combining the cohesion degree and the separation degree, and both the cohesion degree and the separation degree are negatively correlated with the noise evaluation value.

[0078] As a specific implementation method, a method for obtaining cohesion and separation is given, specifically: the negative correlation normalized value of the sparse factor of the candidate set is determined as the cohesion of the candidate set; the cumulative sum of the Euclidean distances between the center position of the candidate set and the center positions of all its adjacent similar point sets is determined as the separation corresponding to the candidate set.

[0079] In this embodiment, a specific calculation formula for the noise evaluation value is given, and the noise evaluation value of the j-th similar point set can be expressed as:

[0080]

[0081] in, represents the noise evaluation value of the j-th similar point set, represents the number of point cloud data in the jth similar point set, represents the position of the t-th point cloud data in the j-th similar point set, represents the center position of the jth similar point set, represents the Euclidean distance between the t-th point cloud data in the j-th similar point set and the center position of the j-th similar point set, K represents the number of neighboring similar point sets of the j-th similar point set, represents the center position of the kth adjacent similar point set of the jth similar point set, represents the Euclidean distance between the center position of the j-th similar point set and the center position of the k-th adjacent similar point set of the j-th similar point set, represents an exponential function with a natural constant as base, represents the normalization function, Indicates the preset second adjustment parameter.

[0082] In this embodiment, the method for obtaining the adjacent similar point sets of any similar point set is: respectively obtaining the Euclidean distance between the center position of the similar point set and the center position of each other similar point set, and sequentially sorting the other similar point sets except the similar point set in the order of the Euclidean distance from small to large, to obtain a similar point set sequence, and using the first preset number of similar point sets in the similar point set sequence as the adjacent similar point sets of the similar point set. In this embodiment, the preset number is 5, and in specific applications, the implementer can set the preset number according to specific circumstances.

[0083] The purpose of introducing the preset second adjustment parameter into the calculation formula of the noise evaluation value is to prevent the denominator from being 0. In this embodiment, the preset second adjustment parameter is 0.01. In specific applications, the implementer can set it according to the specific situation. It represents the negative correlation normalized value of the sparse factor of the j-th similar point set, that is, the cohesion of the j-th similar point set. The larger the value, the more clustered the point cloud data in the j-th similar point set. It represents the separation degree corresponding to the j-th similar point set, which is used to reflect the difference between the j-th similar point set and its neighboring similar point sets. The larger the value, the greater the difference between the j-th similar point set and its neighboring similar point sets, that is, the better the clustering effect of the j-th similar point set. The clustering effect of the jth similar point set is reflected by combining the cohesion of the point cloud data within the jth similar point set and the separation between the jth similar point set and its neighboring similar point sets. When the cohesion of the jth similar point set is higher and the corresponding separation is also higher, it means that the clustering effect of the jth similar point set is better and the noise level of the jth similar point set is lower, that is, the noise evaluation value of the jth similar point set is smaller.

[0084] By using the above method, the noise evaluation value of each similar point set can be obtained.

[0085] The higher the noise level of the similar point set, the larger the filter radius needs to be selected to remove the noise. At the same time, the more obvious the distribution characteristics of the point cloud data in the similar point set, the greater the possibility that the point cloud data in the similar point set belongs to the same design structure, and the smaller the filter radius needs to be selected for fine processing. The filter radius is comprehensively adjusted according to the noise level of the similar point set and the distribution characteristics of the point cloud data in the similar point set. The two check and balance each other to avoid over-filtering or under-filtering caused by too large or too small filter radius, so as to improve the filtering effect.

[0086] Specifically, for the candidate set, the round-up result of the ratio between the noise evaluation value of the candidate set and the data distribution characteristic value of the candidate set is calculated; the product between the round-up result and the preset filter radius is determined as the filter radius corresponding to the candidate set. In this embodiment, the preset filter radius is 0.1 meters. In specific applications, the implementer can set it according to specific circumstances.

[0087] The filter radius corresponding to the j-th similar point set can be expressed as:

[0088]

[0089] In the formula, Represents the filter radius corresponding to the j-th similar point set; represents the noise evaluation value of the j-th similar point set, represents the data distribution characteristic value of the jth similar point set, Indicates the preset filter radius, Represents the ceiling function.

[0090] When the noise evaluation value of the j-th similar point set is larger and the data distribution characteristic value of the j-th similar point set is smaller, a larger filter radius should be used to filter its point cloud data.

[0091] By adopting the above method, the filter radius corresponding to each similar point set can be obtained.

[0092] Step S4, filtering the corresponding similar point set using the filtering radius, and evaluating the quality of the casting based on the filtered point cloud data.

[0093] In this embodiment, the filter radius corresponding to each similar point set is obtained in step S3. Next, each similar point set is filtered using the corresponding filter radius, and the quality of the casting is evaluated using the filtered data.

[0094] For any similar point set, the filter radius corresponding to the similar point set is used to filter the point cloud data in the similar point set. The specific process is: taking each point cloud data in the similar point set as the center, calculating the average value of all point cloud data within the filter radius, and replacing the point cloud data with the calculated average value. Using this method, all point cloud data in all similar point sets are processed to obtain the filtered point cloud data of the casting surface.

[0095] The filtered point cloud data of the casting surface is input into the trained neural network, and the quality assessment result of the casting is output. The neural network in this embodiment is used to assess the quality of the casting. The training of the neural network is a prior art and will not be described in detail here. As another embodiment, the filtered point cloud data can also be compared with the standard point cloud data, and the quality of the casting can be assessed according to the comparison result.

[0096] So far, the quality inspection of the casting has been completed by using the method provided in this embodiment.

[0097] This embodiment first divides all point cloud data into multiple similar point sets based on the density distribution of point cloud data on the surface of the casting in the water glass sand casting process. Similar data in the point cloud data on the surface of the casting usually have similar geometric structures and distribution characteristics. By classifying similar data into a subset and performing unified filtering processing on the subset, the processing time can be significantly reduced, and it can also be ensured that they are subjected to the same processing, thereby maintaining data consistency. All point cloud data in the same similar point set use the same filtering radius, and then according to the relative position distribution of the point cloud data in each similar point set and the distribution of the point cloud data in the neighborhood of each point cloud data, the data distribution characteristic value of each similar point set is obtained. The data distribution characteristic value is used to reflect the sparsity and regularity of the distribution of the point cloud data in the similar point set. For the same structure of the casting, the corresponding point cloud data distribution should be more clustered and more regular. Furthermore, combined with the clustered distribution characteristics between the point cloud data in the similar point set and the discrete distribution characteristics between different similar point sets, the data distribution of each similar point set is analyzed. The noise level is evaluated and the noise evaluation value is obtained. In order to ensure the filtering effect of the point cloud data, the larger the noise evaluation value, the larger the filter radius should be selected to remove the noise. The more obvious the distribution characteristics of the point cloud data in the similar point set, the greater the possibility that the point cloud data in the similar point set belongs to the same design structure, and a smaller filter radius needs to be selected for fine processing. Therefore, the filter radius corresponding to the similar point set is determined in combination with the data distribution characteristic value and the noise evaluation value. Each similar point set corresponds to a filter radius. According to the filter radius corresponding to each similar point set, all point cloud data in the same similar point set are uniformly filtered. This avoids the problem of poor filtering effect in some areas and possible over-filtering in other areas caused by the global use of the same filter radius for filtering, and avoids the problem of large consumption of computing resources caused by the use of one filter radius for one point cloud data, ensuring the filtering effect of the point cloud data on the surface of the casting. Finally, the point cloud data after filtering is used to evaluate the quality of the casting, thereby improving the accuracy of the evaluation results of the casting quality.

[0098] An automatic detection system for casting quality in water glass sand casting process:

[0099] like Figure 2As shown, the figure shows a structural block diagram of an automatic detection system for casting quality in water glass sand casting process, the system includes a data acquisition module, a data distribution characteristic value determination module, a filter radius determination module and a quality assessment module.

[0100] Among them, the data acquisition module is used to obtain point cloud data of the casting surface of the water glass sand casting process;

[0101] A data distribution characteristic value determination module is used to obtain the data distribution characteristic value of each similar point set according to the relative position distribution of the point cloud data in each similar point set and the distribution of the point cloud data in the neighborhood of each point cloud data; the similar point set is obtained by dividing all the point cloud data based on the density distribution of the point cloud data;

[0102] A filter radius determination module is used to determine the noise evaluation value of each similar point set by combining the clustered distribution characteristics between the point cloud data in the similar point set and the discrete distribution characteristics between different similar point sets; and obtain the filter radius corresponding to each similar point set according to the data distribution characteristic value and the noise evaluation value;

[0103] The quality assessment module is used to filter the corresponding similar point set using the filtering radius and evaluate the quality of the casting based on the point cloud data after filtering.

[0104] It should be understood that Figure 2 The structural block diagram of a casting quality automatic detection system for water glass sand casting process shown and its modules can be implemented in various ways. For example, in some embodiments, the system and its modules can be implemented by hardware, software or a combination of software and hardware. Among them, the hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art can understand that the above-mentioned method and system can be implemented using computer executable instructions and / or included in a processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The system and its modules of this specification can not only be implemented by hardware circuits such as ultra-large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., but can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software (for example, firmware).

[0105] For more details about the above modules, please refer to other places in this manual and will not be repeated here.

[0106] In other embodiments, there is also provided a device for automatically detecting the quality of castings for a water glass sand casting process, comprising a memory and a processor. The memory is used to store executable program codes, and the processor is used to call and run the executable program codes from the memory, so that the device executes the above-mentioned method for automatically detecting the quality of castings for a water glass sand casting process. The device can be specifically a chip, a component or a module, and the chip may include a connected processor and a memory; wherein the memory is used to store instructions, and when the processor calls and executes the instructions, the chip can execute the above-mentioned method for automatically detecting the quality of castings for a water glass sand casting process provided in the above-mentioned embodiment.

[0107] In other embodiments, a computer program product is also provided. When the computer program product is run on a computer, the computer executes the above-mentioned related steps to implement an automatic detection method for casting quality for a water glass sand casting process provided in the above embodiment.

[0108] In other embodiments, a computer-readable storage medium is also provided, in which a computer program code is stored. When the computer program code is run on a computer, the computer executes the above-mentioned related method steps to implement an automatic detection method for casting quality for a water glass sand casting process provided in the above embodiment.

[0109] Among them, the provided system, electronic device, computer program product, and computer-readable storage medium are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0110] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for automatically detecting casting quality in a water glass sand casting process, characterized in that: The method comprises the following steps: Obtain point cloud data of the casting surface in the water glass sand casting process; According to the relative position distribution of the point cloud data in each similar point set and the distribution of the point cloud data in the neighborhood of each point cloud data, the data distribution characteristic value of each similar point set is obtained; the similar point set is obtained by dividing all the point cloud data based on the density distribution of the point cloud data; Determine the noise evaluation value of each similar point set by combining the clustered distribution characteristics between point cloud data in the similar point set and the discrete distribution characteristics between different similar point sets; obtain the filter radius corresponding to each similar point set according to the data distribution characteristic value and the noise evaluation value; Filtering the corresponding similar point set using the filtering radius, and evaluating the quality of the casting based on the filtered point cloud data; The data distribution characteristic values ​​of each similar point set are obtained according to the relative position distribution of the point cloud data in each similar point set and the distribution of the point cloud data in the neighborhood of each point cloud data, including: Calculate the average value of the horizontal coordinate, the average value of the vertical coordinate and the average value of the vertical coordinate of all the point cloud data in the candidate set respectively; The position formed by the average value of the horizontal coordinate, the average value of the vertical coordinate and the average value of the vertical coordinate is recorded as the center position; according to the relative position between all the point cloud data in the candidate set and the center position, the sparse factor of the candidate set is obtained; For any point cloud data in the candidate set, principal component analysis is performed on the point cloud data and all point cloud data in its neighborhood to obtain a feature vector corresponding to the maximum eigenvalue as the feature vector corresponding to the point cloud data; According to the degree of fluctuation of the modulus of the feature vector corresponding to the point cloud data in the candidate set and the sparsity factor, the data distribution characteristic value of the candidate set is obtained, and both the degree of fluctuation and the sparsity factor are negatively correlated with the data distribution characteristic value; The candidate set is any similar point set; The step of combining the clustered distribution characteristics between point cloud data in the similar point set and the discrete distribution characteristics between different similar point sets to determine the noise evaluation value of each similar point set includes: Based on the clustering distribution characteristics between the point cloud data in the candidate set, the cohesion of the candidate set is determined; Based on the discrete distribution characteristics between the candidate set and its adjacent similar point set, the separation degree corresponding to the candidate set is determined; The noise evaluation value of the candidate set is obtained by combining the cohesion and the separation, wherein both the cohesion and the separation are negatively correlated with the noise evaluation value; The calculation formula for the data distribution characteristic value of the jth similar point set is: in, represents the data distribution characteristic value of the jth similar point set, Represents the variance of the modulus length of the eigenvector corresponding to all point cloud data in the jth similar point set, represents the number of point cloud data in the jth similar point set, represents the position of the t-th point cloud data in the j-th similar point set, represents the center position of the jth similar point set, represents the Euclidean distance between the t-th point cloud data in the j-th similar point set and the center position of the j-th similar point set, represents the preset first adjustment parameter, norm() represents the normalization function; The calculation formula for the noise evaluation value of the jth similar point set is: in, represents the noise evaluation value of the j-th similar point set, K represents the number of neighboring similar point sets of the j-th similar point set, represents the center position of the kth adjacent similar point set of the jth similar point set, represents the Euclidean distance between the center position of the j-th similar point set and the center position of the k-th adjacent similar point set of the j-th similar point set, represents an exponential function with a natural constant as base, Indicates the preset second adjustment parameter; The step of obtaining the filter radius corresponding to each similar point set according to the data distribution characteristic value and the noise evaluation value comprises: Calculate the result of rounding up the ratio between the noise evaluation value of the candidate set and the data distribution characteristic value of the candidate set; The product of the rounded-up result and the preset filtering radius is determined as the filtering radius corresponding to the candidate set.

2. The automatic detection method for casting quality of water glass sand casting process according to claim 1 is characterized in that: The acquisition of the similar point set includes: The DBSCAN clustering algorithm is used to cluster all point cloud data into several clusters, and all point cloud data in each cluster are regarded as a similar point set.

3. The automatic detection method for casting quality of water glass sand casting process according to claim 1 is characterized in that: The step of obtaining the sparse factor of the candidate set according to the relative positions between all point cloud data in the candidate set and the center position includes: The average value of the Euclidean distances between all point cloud data in the candidate set and the central position is determined as the sparse factor of the candidate set.

4. The automatic detection method for casting quality of water glass sand casting process according to claim 1 is characterized in that: The acquisition of the fluctuation degree of the modulus length of the feature vector corresponding to the point cloud data in the candidate set includes: The variance of the modulus of the feature vectors corresponding to all point cloud data in the candidate set is used as the fluctuation degree.

5. The automatic detection method for casting quality of water glass sand casting process according to claim 1 is characterized in that: The step of determining the cohesion of the candidate set based on the clustered distribution characteristics of the point cloud data in the candidate set includes: The negatively correlated normalized value of the sparse factor of the candidate set is determined as the cohesion of the candidate set.

6. The automatic detection method for casting quality of water glass sand casting process according to claim 1 is characterized in that: Based on the discrete distribution characteristics between the candidate set and its adjacent similar point set, the separation degree corresponding to the candidate set is determined, including: The cumulative sum of the Euclidean distances between the center position of the candidate set and the center positions of all its neighboring similar point sets is determined as the separation degree corresponding to the candidate set.

7. The automatic detection method for casting quality of water glass sand casting process according to claim 1 is characterized in that: The method of evaluating the quality of the casting based on the filtered point cloud data includes: All filtered point cloud data are input into the trained neural network to obtain the quality assessment results of the casting.

Images