CN-121998936-A - Method and system for carrying out three-dimensional flaw detection on industrial product based on multi-source data

CN121998936ACN 121998936 ACN121998936 ACN 121998936ACN-121998936-A

Abstract

The application discloses a three-dimensional flaw detection method and system for industrial products based on multi-source data, which belong to the technical field of industrial product flaw detection and comprise the steps of determining spatial resolutions of training samples and ultrasonic flaw detection probes, dividing a world coordinate system based on the spatial resolutions to obtain a three-dimensional space containing a plurality of discrete units, configuring the internal discretized point cloud data of the training samples in the three-dimensional space, assigning the discrete units with the point cloud data as a first type unit, assigning the discrete units without the point cloud data as a second type unit, acquiring an internal flaw detection model based on the assigned discrete units, acquiring an external flaw detection model through external image data of the training samples, determining a workpiece to be detected, inputting the internal point cloud data of the workpiece to be detected into the internal flaw detection model to obtain an internal flaw detection result, and inputting the external image data of the workpiece to be detected into the external flaw detection model to obtain the external flaw detection result.

Inventors

SHEN YULONG
YUAN BO
CHEN SENLIN
HU KAI

Assignees

南京栢拓视觉科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (9)

1. The method for three-dimensional flaw detection of industrial products based on multi-source data is characterized by being applied to flaw detection equipment, wherein the flaw detection equipment comprises a mechanical arm, an ultrasonic flaw detection probe and a camera, wherein the ultrasonic flaw detection probe is arranged on the mechanical arm and used for acquiring target internal point cloud data, and the camera is used for acquiring target external image data, and the method comprises the following steps: Determining the spatial resolution of a training sample and the ultrasonic flaw detection probe; dividing a world coordinate system based on the spatial resolution to obtain a three-dimensional space containing a plurality of discrete units; the discretized point cloud data in the training sample are configured in the three-dimensional space; Assigning discrete units with the point cloud data as first class units, and assigning discrete units without the point cloud data as second class units; acquiring an internal flaw detection model based on the assigned discrete units; acquiring an external flaw detection model through external image data of the training sample; the method comprises the steps of determining a workpiece to be detected, inputting internal point cloud data of the workpiece to be detected into the internal flaw detection model to obtain an internal flaw detection result, inputting external image data of the workpiece to be detected into the external flaw detection model to obtain an external flaw detection result.
2. The method of three-dimensional inspection of industrial products based on multi-source data of claim 1, wherein the training samples comprise qualified samples and the step of obtaining an internal inspection model based on the assigned discrete units comprises: Acquiring the average value of the discrete units at the same position in a plurality of qualified samples, wherein the average value is a standard value; comparing the values of the same discrete units in any two qualified samples in a plurality of qualified samples, and determining the average value of the number of the discrete units with differences as a standard range value; and acquiring the internal flaw detection model based on the standard value and the standard range value.
3. The method for three-dimensional inspection of industrial products based on multi-source data according to claim 2, wherein the step of inputting the internal point cloud data of the workpiece to be inspected into the internal inspection model to obtain the internal inspection result comprises: the discretized point cloud data in the workpiece to be detected are configured in the three-dimensional space; In the workpiece to be detected, assigning discrete units with the point cloud data as first class units and assigning discrete units without the point cloud data as second class units; and judging the workpiece to be detected as a qualified product in response to the fact that the number of discrete units which are different from the standard value in the workpiece to be detected is smaller than or equal to the standard range value.
4. The method of three-dimensional inspection of industrial products based on multi-source data of claim 1, wherein the training samples comprise qualified samples, and wherein the step of obtaining an external inspection model comprises: acquiring the average value of R channels, the average value of G channels and the average value of B channels of the same pixel point in a plurality of images in the RGB format of the qualified sample, and constructing a standard channel value; Acquiring total differences of the same pixel points in an R channel, a G channel and a B channel and variance values of a plurality of total differences of the same pixel points in any two qualified samples in a plurality of images in the RGB format of the qualified samples; Obtaining an average value of the pixel point variance values, and constructing a standard channel range value; And acquiring the external flaw detection model based on the standard channel range value.
5. The method of three-dimensional inspection of industrial products based on multi-source data of claim 1, wherein the step of obtaining an internal inspection model based on the assigned discrete units comprises: taking the first type unit as a center to outwards cover a plurality of discrete units to obtain a flaw detection convolution kernel; layering the flaw detection convolution kernels along a preset direction to obtain a plurality of layered convolution surfaces; Sequentially splicing the plurality of layered convolution surfaces along the layering sequence of the preset direction to obtain spliced convolution surfaces; and determining the internal flaw detection model to be trained, training the internal flaw detection model based on the spliced convolution surface of the training sample, and obtaining the trained internal flaw detection model.
6. The method of three-dimensional inspection of an industrial article based on multi-source data of claim 5, wherein the internal inspection model comprises an MLP neural network comprising: the number of neurons of the input layer is the same as the number of discrete units of the spliced convolution face; The number of neurons of the middle layer is twice the number of discrete units of the spliced convolution face; And the output layer comprises 3 neurons which are respectively used for representing qualification, disqualification and undetermining.
7. The method for three-dimensional flaw detection of industrial products based on multi-source data according to claim 1, wherein the external flaw detection model constrains the characteristics in the dual-time domain characteristic extraction and the characteristic fusion, and a non-local characteristic pyramid network NL-FPN is arranged in the center of a main network of the external flaw detection model for extracting and fusing multi-scale characteristics so as to construct a characteristic fusion module DFM based on dense connection to perform robust fusion on dual-temporal characteristics; And comparing the qualified sample under the same pose with external image data of the workpiece to be detected through an external flaw detection model to obtain an external flaw detection result.
8. The method for three-dimensional inspection of industrial products based on multi-source data according to claim 1, characterized in that, after assigning the discrete units, the method further comprises: taking the first type unit as a center to outwards cover a plurality of discrete units to obtain a filtering enhancement convolution kernel; in the filter enhancement convolution kernel, assigning discrete units on the two first type unit lines as first type units in response to the existence of two first type units with an L2 norm distance of less than or equal to 2; in the filter enhanced convolution kernel, responsive to only one first type of element being present, the first type of element is assigned as a second type of element.
9. A system for three-dimensional inspection of an industrial article based on multi-source data, characterized in that it is applied to inspection equipment, the inspection equipment includes a mechanical arm, and an ultrasonic inspection probe and a camera for obtaining target external image data, wherein the ultrasonic inspection probe is arranged on the mechanical arm and used for obtaining target internal point cloud data, and the system comprises: A training sample configuration module (10) for determining the spatial resolution of the training sample and the ultrasonic inspection probe; A space construction module (20) for dividing a world coordinate system based on the spatial resolution, obtaining a three-dimensional space comprising a plurality of discrete units; A point cloud configuration module (30) configured to configure the discretized point cloud data inside the training sample in the three-dimensional space; A assigning module (40) for assigning discrete units having the point cloud data as a first type of units and assigning discrete units having no point cloud data as a second type of units; an external flaw detection model acquisition module (50) for acquiring an internal flaw detection model based on the assigned discrete units; an internal flaw detection model acquisition module (60) for acquiring an external flaw detection model from the external image data of the training sample; The flaw detection result module (70) is used for determining a workpiece to be detected, inputting the internal point cloud data of the workpiece to be detected into the internal flaw detection model to obtain an internal flaw detection result, and inputting the external image data of the workpiece to be detected into the external flaw detection model to obtain an external flaw detection result.

Description

Method and system for carrying out three-dimensional flaw detection on industrial product based on multi-source data Technical Field The application belongs to the technical field of industrial flaw detection, and particularly relates to a method and a system for carrying out three-dimensional flaw detection on industrial products based on multi-source data. Background Industrial products like automobile mechanical parts are produced in large quantities, and the internal quality of the industrial products is difficult to detect due to the solid sealing and isomerism of the industrial products, and in many cases, only a quality detection scheme of one-time destructive spot check can be adopted. Not only is the full inspection impossible, but it is also destructive, and it is apparent that such a detection is defective. In addition, for the problem of external detection of objects, three-dimensional damage detection is still needed by human eyes at present, and in the production process, the requirement on the human eyes is high, and the full detection cannot be realized. Disclosure of Invention The application aims to develop a method and a system for carrying out three-dimensional flaw detection on industrial products based on multi-source data so as to solve the technical problems. According to a first aspect, the application provides a three-dimensional flaw detection method for industrial products based on multi-source data, which is applied to flaw detection equipment, wherein the flaw detection equipment comprises a mechanical arm, an ultrasonic flaw detection probe and a camera, wherein the ultrasonic flaw detection probe is arranged on the mechanical arm and used for acquiring target internal point cloud data, and the camera is used for acquiring target external image data, and the method comprises the following steps: Determining the spatial resolution of a training sample and the ultrasonic flaw detection probe; dividing a world coordinate system based on the spatial resolution to obtain a three-dimensional space containing a plurality of discrete units; the discretized point cloud data in the training sample are configured in the three-dimensional space; Assigning discrete units with the point cloud data as first class units, and assigning discrete units without the point cloud data as second class units; acquiring an internal flaw detection model based on the assigned discrete units; acquiring an external flaw detection model through external image data of the training sample; the method comprises the steps of determining a workpiece to be detected, inputting internal point cloud data of the workpiece to be detected into the internal flaw detection model to obtain an internal flaw detection result, inputting external image data of the workpiece to be detected into the external flaw detection model to obtain an external flaw detection result. In some embodiments, the training samples comprise qualified samples, and the step of obtaining an internal inspection model based on the assigned discrete units comprises: Acquiring the average value of the discrete units at the same position in a plurality of qualified samples, wherein the average value is a standard value; comparing the values of the same discrete units in any two qualified samples in a plurality of qualified samples, and determining the average value of the number of the discrete units with differences as a standard range value; and acquiring the internal flaw detection model based on the standard value and the standard range value. In some embodiments, the step of inputting the internal point cloud data of the workpiece to be inspected into the internal inspection model, and obtaining an internal inspection result includes: the discretized point cloud data in the workpiece to be detected are configured in the three-dimensional space; In the workpiece to be detected, assigning discrete units with the point cloud data as first class units and assigning discrete units without the point cloud data as second class units; and judging the workpiece to be detected as a qualified product in response to the fact that the number of discrete units which are different from the standard value in the workpiece to be detected is smaller than or equal to the standard range value. In some embodiments, the training samples comprise qualified samples, and the step of acquiring the external inspection model comprises: acquiring the average value of R channels, the average value of G channels and the average value of B channels of the same pixel point in a plurality of images in the RGB format of the qualified sample, and constructing a standard channel value; Acquiring total differences of the same pixel points in an R channel, a G channel and a B channel and variance values of a plurality of total differences of the same pixel points in any two qualified samples in a plurality of images in the RGB format of the qualified samples; Obtaining an average value