CN-122023271-A - Defect detection method and system for casting machining surface

CN122023271ACN 122023271 ACN122023271 ACN 122023271ACN-122023271-A

Abstract

The application discloses a defect detection method and system for a casting machining surface, wherein the method comprises the steps of acquiring a casting machining surface of a machined casting through an industrial camera preset in machining equipment to obtain a casting image to be detected in the machining process of the machining equipment based on current tool technological parameters, identifying each suspected defect area in the casting image to be detected through a preset basic visual defect detection model, extracting image characteristics of each suspected defect area, fusing the image characteristics with the current tool technological parameters to obtain comprehensive characteristics of each suspected defect area, inputting the comprehensive characteristics into a pre-trained machining parameter-tool image characteristic correlation model, outputting probability that each suspected defect area is of a tool type, and screening the suspected defect areas with probability smaller than a preset tool judgment threshold as real defect areas. Therefore, the application can effectively avoid erroneous judgment of the cutter grain, reduce the production cost and improve the economic benefit of enterprises.

Inventors

GUO ZHIMING
WANG SHANWEI
KONG GANG
ZHUANG SHUNXU
LI SHUANG
WU QIHUA
DOU YUNXIA
FANG YUNTAO
ZHAI QIANG
ZHU YAOWEN
QI WEI
WANG HAINING

Assignees

潍柴动力股份有限公司

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (10)

1.A method for detecting defects in a machined surface of a casting, the method comprising: In the casting processing process of processing equipment based on the current cutter technological parameters, acquiring a casting processing surface of a processed casting through an industrial camera preset in the processing equipment to obtain a casting image to be detected; Identifying each suspected defect area in the casting image to be detected through a preset basic visual defect detection model; extracting the image characteristics of each suspected defect area, and fusing the image characteristics with the current cutter technological parameters to obtain the comprehensive characteristics of each suspected defect area; Inputting the comprehensive characteristics into a pre-trained processing parameter-tool mark image characteristic association model, outputting the probability that each suspected defect area is of a tool mark type, and screening the suspected defect areas with the probability smaller than a preset tool mark judgment threshold value as real defect areas.
2. The method of claim 1, wherein generating the pre-trained process parameter-moire image feature correlation model comprises: acquiring a history marking data set which is generated in the casting processing process of the processing equipment based on the key tool process parameters of the current tool, wherein the marking data set comprises each history casting processing surface image corresponding to the key tool process parameters of the current tool, the key tool process parameters of the current tool and marking labels of each history casting processing surface image; Preprocessing the processing surface images of the historical castings to obtain preprocessed images; respectively extracting the cutter grain image characteristics of cutter grains from each preprocessed image to obtain the cutter grain image characteristics of each preprocessed image; Performing associated feature fusion on the cutter grain image features of each preprocessing image and the key cutter technological parameters of the current cutter to obtain fusion feature vectors of each preprocessing image; And generating a pre-trained processing parameter-tool grain image characteristic association model based on the fusion characteristic vector.
3. The method of claim 2, wherein generating the historical annotation dataset comprises: collecting the cutter state and cutting condition of a current cutter in the processing equipment as key cutter technological parameters of the current cutter; In the casting machining process of the machining equipment based on the key tool technological parameters of the current tool, acquiring images of the machined casting machining surface of the machined casting to obtain each historical casting machining surface image corresponding to the key tool technological parameters of the current tool; Correlating each historical casting machined surface image corresponding to the key tool technological parameters of the current tool with the key tool technological parameters of the current tool to obtain an image-parameter correlation data set; and marking the knife grain area, the actual defect area and the normal area on each historical casting machining surface image in the image-parameter association data set to obtain a marked data set.
4. The method of claim 2, wherein said preprocessing the historical cast finish images comprises: denoising the processing surface images of the historical castings to clarify the contour of the tool grain; Carrying out gray scale normalization on each history casting processing surface image after denoising treatment so as to convert a color image into a gray scale image and normalize pixel values to a [0,1] interval; and (5) enhancing the image contrast of each history casting processing surface image after the gray level normalization so as to highlight the texture details of the tool grains.
5. The method according to claim 2, wherein extracting the moire image features of the moire for each of the preprocessed images, respectively, to obtain the moire image features of each of the preprocessed images, comprises: Extracting energy, entropy, contrast and correlation characteristics of each preprocessed image to be used as texture characteristics of each preprocessed image; Extracting the width, length, spacing and trend angle of the cutter lines of each preprocessing image to obtain morphological characteristics for describing the geometric shape and distribution rule of the cutter lines; calculating the average gray value and gray variance of each preprocessed image to obtain gray features for representing the depth and uniformity of the moire; And taking the texture feature, the morphological feature and the gray level feature as knife grain image features of the preprocessed images.
6. The method according to claim 2, wherein the performing the associated feature fusion of the moire image feature of each preprocessed image and the key tool process parameter of the current tool to obtain the fused feature vector of each preprocessed image comprises: carrying out vector standardization processing on the cutter grain image characteristics of each preprocessing image and the key cutter technological parameters of the current cutter to obtain each image characteristic standardization vector and processing parameter standardization vector; and carrying out weighted fusion on the image feature standardization vectors and the processing parameter standardization vectors to obtain fusion feature vectors of the preprocessed images.
7. The method of claim 2, wherein the labeling label of each historical cast machined surface image comprises a knife grain area label, a non-knife grain area label; the generating a pre-trained processing parameter-tool grain image feature association model based on the fusion feature vector comprises the following steps: Constructing a correlation model by adopting a gradient lifting tree algorithm; Taking the logarithmic loss function as a target loss function of the association model to obtain a classification model; Training a classification model by taking the fusion feature vector of each preprocessed image as input and the tool grain area label and the non-tool grain area label of each history casting processing surface image as output to obtain a first loss value; And generating a pre-trained processing parameter-tool grain image characteristic association model under the condition that the first model loss value reaches the minimum.
8. The method of claim 2, wherein the labeling labels of each historical cast machined surface image include normal area labels, actual defect area labels; Generating a preset basic visual defect detection model according to the following steps: Constructing a basic detection model by adopting a target detection algorithm; Taking the processing surface images of the historical castings as input, taking the normal area labels and the actual defect area labels as output, training the basic detection model, and outputting a second loss value; And generating a preset basic visual defect detection model under the condition that the second loss value reaches the minimum.
9. The method according to claim 1, wherein the method further comprises: Performing secondary labeling on the casting image to be detected by adopting the real defect area to obtain current labeling information; combining the current tool technological parameters, the casting image to be detected and the current labeling information into an incremental data set; And performing model fine adjustment on the preset basic visual defect detection model and the pre-trained processing parameter-tool grain image characteristic association model by adopting the incremental data set.
10. A system for detecting defects in a casting work surface, the system comprising: The image acquisition module is used for acquiring a casting machining surface of a machined casting through an industrial camera preset in the machining equipment in the casting machining process of the machining equipment based on the current tool technological parameters, so as to obtain a casting image to be detected; the defect area identification module is used for identifying each suspected defect area in the casting image to be detected through a preset basic visual defect detection model; The feature fusion module is used for extracting the image features of each suspected defect area, and fusing the image features with the current cutter technological parameters to obtain the comprehensive features of each suspected defect area; The real defect region screening module is used for inputting the comprehensive characteristics into a pre-trained processing parameter-tool mark image characteristic association model, outputting the probability that each suspected defect region is of a tool mark type, and screening the suspected defect region with the probability smaller than a preset tool mark judgment threshold value as a real defect region.

Description

Defect detection method and system for casting machining surface Technical Field The application relates to the technical field of computer vision, in particular to a defect detection method and system for a casting machining surface. Background In the field of modern mechanical manufacturing, casting processing is a key link for producing various mechanical equipment parts. After the casting is processed, the surface quality of the casting directly influences the reliability and the service life of the product. Therefore, the defect detection of the casting machining surface is an important process for guaranteeing the product quality. In the related art, a visual image processing method is mainly relied on. Among them, surface defect detection systems based on deep learning are more common. The technology directly learns defect characteristics from images through a deep convolutional neural network, and achieves end-to-end defect detection. The core principle is that a large amount of marked image data is utilized to train the model, so that the model can automatically distinguish the defects from the normal surface characteristics. However, in the casting machining process, cutting actions of machining equipment form cutter marks on a machining surface, the cutter marks belong to normal machining marks, but are regular or irregular texture features in visual images, the prior art only depends on the image features for judgment, key information such as cutter parameters and cutting conditions in the machining process is not considered, and the cutter marks cannot be effectively distinguished from actual defects by a traditional visual detection system, so that the detection efficiency is reduced, qualified castings are discarded due to misjudgment, and the production cost is increased. Disclosure of Invention The embodiment of the application provides a defect detection method and system for a casting machining surface. The following presents a simplified summary in order to provide a basic understanding of some aspects of the disclosed embodiments. This summary is not an extensive overview and is intended to neither identify key/critical elements nor delineate the scope of such embodiments. Its sole purpose is to present some concepts in a simplified form as a prelude to the more detailed description that is presented later. In a first aspect, an embodiment of the present application provides a method for detecting a defect of a machined surface of a casting, including: In the casting processing process of processing equipment based on the current cutter technological parameters, acquiring a casting processing surface of a processed casting through an industrial camera preset in the processing equipment to obtain a casting image to be detected; Identifying each suspected defect area in the casting image to be detected by presetting a basic visual defect detection model; extracting the image characteristics of each suspected defect area, and fusing the image characteristics with the current cutter technological parameters to obtain the comprehensive characteristics of each suspected defect area; Inputting the comprehensive characteristics into a pre-trained processing parameter-tool mark image characteristic association model, outputting the probability that each suspected defect area is of a tool mark type, and screening the suspected defect areas with the probability smaller than a preset tool mark judgment threshold value as real defect areas. Optionally, generating a pre-trained processing parameter-tool mark image feature correlation model according to the following steps comprises: The method comprises the steps of obtaining a historical labeling data set, wherein the historical labeling data set is generated in the casting machining process of machining equipment based on key tool technological parameters of a current tool, and the labeling data set comprises all historical casting machining face images corresponding to the key tool technological parameters of the current tool, key tool technological parameters of the current tool and labeling labels of all historical casting machining face images; Preprocessing each history casting machining surface image to obtain each preprocessed image; respectively extracting the cutter grain image characteristics of the cutter grains from each preprocessed image to obtain the cutter grain image characteristics of each preprocessed image; carrying out associated feature fusion on the cutter grain image features of each preprocessed image and key cutter technological parameters of the current cutter to obtain fusion feature vectors of each preprocessed image; And generating a pre-trained processing parameter-tool grain image feature correlation model based on the fusion feature vector. Optionally, generating the history annotation dataset comprises: collecting the cutter state and cutting condition of the current cutter in the processing equipment, and taking t