High-precision rapid measurement method for small-size precision rotary body part

Abstract

The invention relates to the field of precise measurement of complex parts, in particular to a high-precision rapid measurement method for small-size precise rotary parts. Comprising the following steps: acquiring point cloud data of a precise rotary body part, and performing Gaussian blur processing on the point cloud data; constructing a point cloud segmentation model based on a course learning method; training a point cloud segmentation model by using the point cloud data after Gaussian blur processing; extracting and fitting the part information after the point cloud segmentation; and based on the fitted point cloud data, obtaining the three-dimensional geometric dimension and tolerance information of the precise rotary body part as a measurement result. The generalization ability and convergence speed of the model are improved, a multi-scale processing flow is used in a coding stage to realize feature extraction and fusion, and the feature extraction ability and overall performance of the model are improved.

Description

Technical Field

[0001] The invention relates to the field of complex parts precision measurement, in particular to a high-precision and rapid measurement method for small-size precision rotating body parts. Background Art

[0002] In actual engineering applications, small-sized precision rotating parts are core components with small size, precise structure, high technical requirements, and complex manufacturing process. The key geometric features such as spatial special-shaped grooves and micro-holes have a vital role in the overall system. Traditional geometric measurement methods for mechanical parts cannot measure micro-complex structures. Digital measurement technology can convert the surface geometry of the parts to be measured into point cloud data, and analyze the size error and shape error of the workpiece through point cloud processing algorithms.

[0003] The existing high-precision three-dimensional shape and size detection methods are mainly achieved through non-contact measurement methods such as optical multi-focal plane superposition, white light interferometry, confocal, CT, and contact measurement methods of three-coordinate measuring instruments. Among them, optical multi-focal plane superposition, white light interferometry, confocal, and CT methods have a long detection time and are only used for measurement, and cannot achieve rapid measurement; three-coordinate measuring instruments have low measurement efficiency and are easy to damage the workpiece surface, and are not suitable for the construction of complex structural morphology of small-sized precision rotating parts. This method aims at the problems and challenges of the lack of systematic measurement methods for complex structural parts and the difficulty of high-precision measurement. It proposes a digital measurement design method for precision workpieces based on deep learning point cloud processing algorithms, combines structured light three-dimensional scanning devices with multi-angle positioning rotary mechanisms and quick-change tooling, and realizes the rapid detection of small-sized precision rotating parts, breaking through the limitations of traditional measurement principles and measurement methods under complex on-site application conditions, solving the bottleneck problems in the processing and manufacturing of precision structural parts, and providing theoretical basis and technical support for the precision manufacturing and on-site measurement of small-sized complex structural parts. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a high-precision and rapid measurement method for small-sized precision rotating parts, which can realize automatic detection of key dimensional parameters of small-sized precision rotating parts, and the measurement efficiency is greatly improved compared with the current measurement method.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0006] A high-precision and rapid measurement method for small-sized precision rotating parts comprises the following steps:

[0007] 1) Obtain the point cloud data of the precision rotating parts and perform Gaussian blur processing on it;

[0008] 2) Construct a point cloud segmentation model based on curriculum learning method;

[0009] 3) Use the point cloud data after Gaussian blur processing to train the point cloud segmentation model;

[0010] 4) Extract and fit the part information after point cloud segmentation;

[0011] 5) Based on the fitted point cloud data, the three-dimensional geometric dimensions and tolerance information of the precision rotating parts are obtained as measurement results.

[0012] The following steps are also included:

[0013] The measurement results of the precision rotating parts are mapped with the coding information of the identified parts to form a measurement data-part coding information pair, which is then stored in the database.

[0014] The step 1) is specifically as follows: use a Gaussian kernel to scan each data in the point cloud, use a convolution operation to determine the weighted average grayscale value of the pixels in the neighborhood, and replace the value of the central pixel of the template. The relationship is expressed as:

[0015] N=2×σ+1

[0016] Where N*N*N represents the Gaussian kernel size and σ represents the Gaussian kernel standard deviation.

[0017] The step 3) is specifically as follows:

[0018] Assume that the total number of model training rounds is t, the number of Gaussian blur parameters, i.e., the standard deviation σ of the Gaussian kernel, is γ, and they are (σ1, σ2, σ3, ..., σ γ}, the training round is divided into floor(t / γ) segments, floor represents the floor function, then in the first training segment, i.e., (0, floor(t / γ)] round, the input is the point cloud image Gaussian blurred by σ1, in the second training segment, i.e., (floor(t / γ), 2floor(t / γ)] round, the input is the point cloud image Gaussian blurred by σ2, in the third training segment, i.e., (2floor(t / γ), 3floor(t / γ)] round, the input is the point cloud image Gaussian blurred by σ3, and so on, until the input is the point cloud image Gaussian blurred by σ γ Gaussian blurred point cloud image to complete the training.

[0019] In the step 4), fitting is performed based on the least squares principle.

[0020] A high-precision and rapid measurement system for small-size precision rotating parts, comprising:

[0021] Parts accompanying tooling assembly, used to fix precision rotating parts;

[0022] Multi-angle positioning and rotating mechanism, used to fix the parts accompanying tooling components and make them rotate at opposite angles;

[0023] Structured light scanning sensor, used for multi-angle scanning of different posture point cloud data of precision rotating parts;

[0024] Automatic control programming module, used to control the structured light scanning sensor and the multi-angle positioning rotation mechanism to work synchronously;

[0025] The main processor is used to perform Gaussian blur processing on the point cloud data, build and train a point cloud segmentation model based on the curriculum learning method, extract and fit the part information after the point cloud segmentation, and obtain the three-dimensional geometric dimensions and tolerance information of the precision rotating parts as the measurement results.

[0026] It also includes: a part identification module, which is used to identify the coding information of the part, so that the measurement result of the precision rotating part is mapped with the coding information of the identified part to form a measurement data-part coding information pair.

[0027] The present invention has the following beneficial effects and advantages:

[0028] 1. It can realize the automatic detection of key dimensional parameters of small-sized precision rotating parts, storage and processing of detection data, and the measurement efficiency is greatly improved compared with the current measurement method.

[0029] 2. The present invention applies artificial intelligence technology to the automatic segmentation of industrial point cloud three-dimensional data, which reduces the process of human participation, improves data processing efficiency, and avoids errors in measurement results caused by human factors.

[0030] 3. The present invention applies point cloud automatic segmentation technology and proposes a curriculum learning method that adjusts visual acuity at different stages of training, thereby improving the generalization ability and convergence speed of the model. In the encoding stage, a multi-scale processing flow is used to realize feature extraction and fusion, thereby improving the model's feature extraction capability and overall performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic diagram of the structure of a high-precision and rapid measurement method for small-sized precision rotating parts designed by the present invention;

[0032] Figure 2 The present invention is a flowchart of the three-dimensional point cloud data processing process. DETAILED DESCRIPTION

[0033] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0034] like Figure 1The structure of the high-precision and rapid measurement device for small-size precision rotating parts designed by the present invention is shown, including a main processor 01, an automatic control programming module 02, a part identification module 03, a structured light scanning sensor 04, a multi-angle positioning and rotating mechanism 05, and a part accompanying tooling assembly 06, wherein: the main processor 01 is used for overall control of the workflow and processing of the sensor's three-dimensional acquisition data; the automatic control programming module 02 is used to control the structured light sensor 04 and the multi-angle positioning and rotating mechanism 05 to work synchronously. Specifically, the design program allows the angle positioning and rotating mechanism 05 to move to a certain position and posture for acquisition through the structured light sensor 04, and all set position and posture acquisitions are performed. After the completion, the sensor obtains the three-dimensional point cloud data information of the whole part through the splicing algorithm; the part identification module 03 is a barcode or two-dimensional code reading device. In the example, a handheld barcode reader is used to bind the barcode or two-dimensional code on the part accompanying tooling component 06, so that the product code is unique and convenient for quality traceability; the structured light sensor 04 is a high-precision point cloud data acquisition sensor, which is fixedly installed and cooperates with the multi-angle positioning rotary mechanism 05 to image the part at multiple angles; the multi-angle positioning rotary mechanism 05 is a cradle-type dual-axis turntable, which can drive the accompanying tooling component and the workpiece to rotate and swing, and cooperate with the measuring sensor to measure the workpiece from different angles. The dual-axis turntable can adapt to different accompanying tooling with the same interface through the rotary positioning flange to adapt to the measurement of different workpieces. The part accompanying tooling assembly 06 is a part mounting table containing a tooling base, an O-ring, and a QR code label. The positioning shaft section of the accompanying tooling base and the inner stop of the slewing mechanism 05 flange are clearance-matched to achieve rapid replacement of the accompanying tooling assembly. The O-ring compensates for the clearance between the accompanying tooling base and the slewing mechanism 05 flange, and uses the elastic deformation of the sealing ring to increase the friction between the two, preventing the accompanying tooling from escaping from the slewing positioning flange when the slewing mechanism 05 swings.

[0035] like Figure 2 The flowchart of the process of processing three-dimensional point cloud data according to the present invention is shown.

[0036] Step 1: Gaussian blur processing of point cloud data. In order to use the course learning training strategy, let the model start learning from easy samples and gradually advance to complex samples and knowledge, it is necessary to perform Gaussian blur processing on the point cloud image to varying degrees. Gaussian blur uses a Gaussian kernel to scan each data in the point cloud, uses convolution operation to determine the weighted average grayscale value of the pixels in the neighborhood, and replaces the value of the central pixel of the template. The relationship is expressed as:

[0037] N=2×σ+1

[0038] Where (N xN xN) represents the Gaussian kernel size and σ represents the Gaussian kernel standard deviation.

[0039] Step 2, point cloud-based PointNet++ point cloud segmentation algorithm network framework application, PointNet++ is an extension of PointNet, point-based methods directly use point cloud as input, directly use convolution operations to process the original point cloud to obtain global feature representation for segmentation tasks, this method retains the geometric structure characteristics of the three-dimensional shape, and the network performance has been greatly improved. PointNet++ adds a multi-level structure (hierarchical structure) on the basis of PointNet, so that the network can provide higher-level features over larger and larger areas. The present invention improves on the basis of the PointNet++ model, and improves the receptive field by gradually improving the visual acuity of the point cloud image during training through a curriculum learning method.

[0040] Step 3, visual acuity course learning training method, training on point cloud images with gradually increasing resolution can reduce the texture bias of the segmentation model. Therefore, a visual acuity-based model training process is proposed, starting to train the network with blurred and low-resolution point cloud images, and gradually increasing the clarity of the training samples. The initial low-resolution point cloud images allow the model to focus on learning global features while ignoring any noisy or irrelevant point cloud data. In the later stages, clear point cloud images allow the model to pay more attention to relevant local features to make more accurate predictions.

[0041] Suppose the total number of model training rounds is t, the number of Gaussian blur parameters σ is γ, and they are (σ1, σ2, ..., σγ} from large to small, and the training rounds are divided into floor(t / γ) segments, where floor represents the floor rounding function. In the first training segment, i.e., the first floor(t / γ) rounds, the point cloud image Gaussian blurred by σ1 is input, in the second training segment, i.e., (floor(t / γ), 2floor(t / γ)] rounds, the point cloud image Gaussian blurred by σ2 is input, and in the third training segment, i.e., (2floor(t / γ), 3floor(t / γ)] rounds, the point cloud image Gaussian blurred by σ3 is input, and this process is executed in sequence until the entire training process is completed.

[0042] Step 4, extracting part information after point cloud segmentation. The purpose of the measurement method in the present invention is to obtain key three-dimensional geometric dimensions and tolerance information in the rotating part. Through steps 1-3, the segmented point cloud is subjected to information extraction to obtain the segmentation results of each part of the rotating part, such as the cylindrical part, the conical part, and each plane of the spatial special-shaped groove of the rotating part, and each part is fitted accordingly. The fitting method adopts the least squares principle-based method for fitting.

[0043] Step 5, output of measurement results. According to the requirements of the drawing, combined with the planes of the cylindrical part, conical part, and spatial special-shaped groove of the part fitted in step 4, the three-dimensional geometric dimensions and tolerance information of the part are obtained, such as the diameter of the cylinder, the cone angle, the angular relationship between the planes of the special-shaped groove, the groove distribution of the special-shaped groove on the cylinder, etc. At this point, the single part measurement process is completed.

Claims

1. A high-precision and rapid measurement method for small-size precision rotating parts, characterized in that: The following steps are involved: 1) Obtain the point cloud data of the precision rotating parts and perform Gaussian blur processing on it; 2) Construct a point cloud segmentation model based on curriculum learning method; 3) Use the point cloud data after Gaussian blur processing to train the point cloud segmentation model; 4) Extract and fit the part information after point cloud segmentation; 5) Based on the fitted point cloud data, the three-dimensional geometric dimensions and tolerance information of the precision rotating parts are obtained as measurement results.

2. A high-precision and rapid measurement method for small-sized precision rotating parts according to claim 1, characterized in that: The following steps are also included: The measurement results of the precision rotating parts are mapped with the coding information of the identified parts to form a measurement data-part coding information pair, which is then stored in the database.

3. A high-precision and rapid measurement method for small-sized precision rotating parts according to claim 1, characterized in that: The step 1) is specifically as follows: use a Gaussian kernel to scan each data in the point cloud, use a convolution operation to determine the weighted average grayscale value of the pixels in the neighborhood, and replace the value of the central pixel of the template. The relationship is expressed as: N=2×σ+1 Where N*N*N represents the Gaussian kernel size and σ represents the Gaussian kernel standard deviation.

4. A high-precision and rapid measurement method for small-sized precision rotating parts according to claim 1, characterized in that: The step 3) is specifically as follows: Assume that the total number of model training rounds is t, the number of Gaussian blur parameters, i.e., the standard deviation σ of the Gaussian kernel, is γ, and they are {σ1, σ2, σ3, ..., σ γ }, the training round is divided into floor(t / γ) segments, floor represents the floor function, then in the first training segment, i.e., (0, floor(t / γ)] round, the input is the point cloud image Gaussian blurred by σ1, in the second training segment, i.e., (floor(t / γ), 2floor(t / γ)] round, the input is the point cloud image Gaussian blurred by σ2, in the third training segment, i.e., (2floor(t / γ), 3floor(t / γ)] round, the input is the point cloud image Gaussian blurred by σ3, and so on, until the input is the point cloud image Gaussian blurred by σ γ Gaussian blurred point cloud image to complete the training.

5. The high-precision and rapid measurement method for small-sized precision rotating parts according to claim 1 is characterized in that: In the step 4), fitting is performed based on the least squares principle.

6. A high-precision and rapid measurement system for small-size precision rotating parts, characterized in that: include: Parts accompanying tooling assembly, used to fix precision rotating parts; Multi-angle positioning and rotating mechanism, used to fix the parts accompanying tooling components and make them rotate at opposite angles; Structured light scanning sensor, used for multi-angle scanning of different posture point cloud data of precision rotating parts; Automatic control programming module, used to control the structured light scanning sensor and the multi-angle positioning rotation mechanism to work synchronously; The main processor is used to perform Gaussian blur processing on the point cloud data, build and train a point cloud segmentation model based on the curriculum learning method, extract and fit the part information after the point cloud segmentation, and obtain the three-dimensional geometric dimensions and tolerance information of the precision rotating parts as the measurement results.

7. A high-precision and rapid measurement system for small-size precision rotating parts according to claim 6, characterized in that: Also includes: The part identification module is used to identify the coding information of the part, so that the measurement result of the precision rotating part is mapped with the coding information of the identified part to form a measurement data-part coding information pair.

Images