Automobile part injection mold detection method and system based on image analysis

Abstract

The invention discloses an automobile part injection mold detection method and system based on image analysis, and the method comprises the following steps: S1, carrying out anti-vibration interference imaging: decomposing the output light of a multispectral laser source through a beam splitter prism, and generating reference light and detection light; s2, dynamic curvature compensation, wherein a high-speed industrial camera with the object distance being 300 mm is adopted for collecting interference fringe view field covering 200 * 200 mm, and mold CAD curvature data are called; s3, orthogonal feature decoupling: inputting the phase height map into a double-branch convolutional neural network; s4, confidence-driven decision making; s5, outputting a detection result; and S6, cooperative control. According to the piezoelectric displacement real-time control technology based on acceleration feedback, vibration interference is restrained to be smaller than or equal to 0.1 g (actually measured 0.07 g), the interference fringe contrast ratio is improved to 78.5% from 12.3% of a traditional scheme, the minimum detectable defect size is reduced to 3 microns from 50 microns, the detection precision is improved by 16 times, and the problem of microcrack missing detection caused by vibration is thoroughly solved.

Description

Technical Field

[0001] This invention relates to the field of injection mold technology, specifically to a method and system for detecting automotive component injection molds based on image analysis. Background Technology

[0002] Injection molds are used to produce various plastic products; they are essential tools in plastic product manufacturing. Injection molds are made from a wide variety of materials, primarily metals, non-metals, and composites. Metals are among the most commonly used materials for manufacturing injection molds, including steel, aluminum alloys, and copper alloys. Steel is the most common material for injection molds due to its high hardness, high strength, high wear resistance, and good machinability, meeting the requirements of most injection molds. Aluminum alloys have lower density, higher thermal conductivity, and better machinability, making them suitable for producing small injection molds. Copper alloys, on the other hand, have high thermal conductivity and good corrosion resistance, making them suitable for producing high-requirement injection molds.

[0003] In the field of surface defect detection for automotive injection molds, existing technologies face three major bottlenecks: 1. Vibration interference causes blurred imaging: Environmental vibrations in injection molding workshops (frequency 5-100Hz, acceleration 0.2-0.5g) reduce the fringe contrast of traditional interferometers to below 15%. Passive vibration isolation solutions are adopted, but the response delay is >200ms, which cannot meet the requirements of high-speed production lines (detection cycle ≤300ms), resulting in a microcrack missed rate as high as 45%.

[0004] 2. Measurement distortion caused by surface distortion: Complex curved molds (radius of curvature R < 5 mm) result in a phase height map fitting residual ≥ 0.9 μm. Using a fixed-order Zernike polynomial, a residual of 0.95 μm is generated in the reflector bowl area of ​​the headlight (R = 2.3 mm), which masks microcracks less than 3 μm wide.

[0005] 3. Texture interference causes misjudgment: Texture features such as EDM spark marks and polishing marks on the mold surface are easily misidentified as defects: When the texture entropy value E>1.5, the false alarm rate of a single CNN model is high. Traditional confidence models do not consider texture interference, which leads to high-value molds being misjudged and scrapped. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for detecting automotive component injection molds based on image analysis, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting injection molds for automotive parts based on image analysis, comprising the following steps: S1, Anti-vibration Interference Imaging: The multi-spectral laser source output light is decomposed by a beam splitter to generate reference light and probe light. Based on the feedback signal of the triaxial accelerometer, the piezoelectric displacement device is driven to maintain the vibration acceleration ≤0.1g. S2. Dynamic Curvature Compensation: A high-speed industrial camera with a 300mm object distance is used to acquire interference fringe field of view covering 200×200mm. The curvature data of the mold CAD is retrieved and calculated according to the formula. Dynamically set the Zernike order (R is the minimum radius of curvature, in mm); S3. Orthogonal Feature Decoupling: Input the phase height map into a two-branch convolutional neural network: Texture branching extracts base features through dilated convolution (dilation=2). ; Defect branches extract defect features through deformable convolution. ; Based on the cross-correlation loss function L corr Perform feature separation; S4. Confidence-driven decision: Calculating the strength of defect features. With texture entropy ; when >0.7 and When the value is less than 1.5, a defect coordinate map is generated; S5. Output of Detection Results: Outputs a defect distribution map with confidence level labels, including: Defect location coordinates and dimension d; Defect type classification: microcracks / scratches / pits; Confidence score ; S6, Cooperative Control: When there is 0.5 < When the confidence level is <0.7: Real-time monitoring of vibration acceleration a and ambient light intensity ; If a≤0.1g, then local light source enhancement is triggered, increasing the power of the marked area by 20%; Return to step S2 to perform a re-inspection and update the output of step S5.

[0008] The dynamic setting of the Zernike order in step S2 includes: When the radius of curvature R ≥ 20 mm, the order k = 15 is fixed. When 5mm≤R<20mm, follow calculate; When R < 5 mm, perform piecewise fitting: Divide the region into 5×5mm sub-regions centered on the curvature extremum point; Calculate the k value independently for each sub-region.

[0009] The generation of the phase height map in step S2 includes: The Levenberg-Marquardt algorithm is used to solve for the Zernike coefficients. :

[0010] The fitting residual is ≤0.12μm.

[0011] The pre-trained model in step S3 must meet the following requirements: The training dataset contains ≥2000 microcrack samples; Characteristic decoupling orthogonality ≥ 85%.

[0012] The confidence score calculation in step S4 includes: right A high confidence defect greater than 0.7 results in a direct output C=. ; For 0.5≤ For regions ≤0.7, an enhanced formula is used: C= ×(2− )×[1+0.1×sign( -1.2).

[0013] The cooperative closed-loop control in step S6 includes: Re-inspection trigger condition: There are 3 consecutive pixels that satisfy 0.5 < <0.7; Light source enhancement constraints: Power increase , =0.05, =100 Lux; The duration of a single re-inspection is ≤50ms; Results update mechanism: Only pixels with a confidence level increase of ≥0.15 in the original report are replaced; It also includes a timeout interrupt policy: Set the total detection time threshold Tmax = 300ms; When the cumulative re-inspection time is greater than or equal to Tmax: Freeze all control actions; Output the current best detection report; Mark the timeout status code.

[0014] A surface defect detection system for automotive component injection molds based on image analysis, comprising: The vibration-resistant interferometric imaging module includes a laser source and a piezoelectric displacement device; The surface deformation analysis module includes an industrial camera and a Zernike processor; The defect feature decoupling module includes a dual-branch CNN computation unit; The detection result generation module includes: a defect coordinate locator, outputting (x, y, d); a type classifier, which matches pre-stored defect templates based on feature vectors; and a confidence synthesizer, which calculates C= ×(2− The collaborative control execution module, which includes a light intensity modulator, is activated only during re-inspection.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention is based on real-time piezoelectric displacement control technology with acceleration feedback, which suppresses vibration interference to ≤0.1g (measured at 0.07g), increases the contrast of interference fringes from 12.3% in the traditional scheme to 78.5%, reduces the minimum detectable defect size from 50μm to 3μm, improves detection accuracy by 16 times, and completely solves the problem of missed detection of microcracks caused by vibration.

[0016] This invention employs dynamic Zernike order adjustment (automatic piecewise fitting when the radius of curvature R < 5 mm) combined with the Levenberg-Marquardt optimization algorithm. In small curvature regions such as the headlight reflector bowl (R = 2.3 mm), the fitting residual is reduced from 0.95 μm to 0.10 μm, improving the accuracy by 89%, and achieving stable detection of 2.1 μm microcracks for the first time.

[0017] This invention employs a dual-branch CNN model combined with a cross-correlation loss function, achieving an orthogonality of 87.6% between texture features and defect features (higher than the required value of 85%). The detection rate of microcracks is increased from 68.2% to 95.3%, and the detection rates of scratches and pits are simultaneously increased to 98.7% and 96.8%, respectively, thus solving the long-standing problem of texture-defect confusion in the industry.

[0018] The innovative confidence calculation formula of this invention reduces the false alarm rate from 23.7% to 9.8% in coarse-grained texture areas, a reduction of 59%, which can avoid mold scrap losses due to false alarms. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1-2 The present invention provides a technical solution: an image analysis-based method for detecting injection molds of automotive parts, comprising the following steps: S1, Anti-vibration Interference Imaging: The multi-spectral laser source output light is decomposed by a beam splitter to generate reference light and probe light. Based on the feedback signal of the triaxial accelerometer, the piezoelectric displacement device is driven to maintain the vibration acceleration ≤0.1g. S2. Dynamic Curvature Compensation: A high-speed industrial camera with a 300mm object distance is used to acquire interference fringe field of view covering 200×200mm. The curvature data of the mold CAD is retrieved and calculated according to the formula. Dynamically set the Zernike order (R is the minimum radius of curvature, in mm); S3. Orthogonal Feature Decoupling: Input the phase height map into a two-branch convolutional neural network: Texture branching extracts base features through dilated convolution (dilation=2). ; Defect branches extract defect features through deformable convolution. ; Based on the cross-correlation loss function L corr Perform feature separation; S4. Confidence-driven decision: Calculating the strength of defect features. With texture entropy ; when >0.7 and When the value is less than 1.5, a defect coordinate map is generated; S5. Output of Detection Results: Outputs a defect distribution map with confidence level labels, including: Defect location coordinates and dimension d; Defect type classification: microcracks / scratches / pits; Confidence score ; S6, Cooperative Control: When there is 0.5 < When the confidence level is <0.7: Real-time monitoring of vibration acceleration a and ambient light intensity ; If a≤0.1g, then local light source enhancement is triggered, increasing the power of the marked area by 20%; Return to step S2 to perform a re-inspection and update the output of step S5.

[0022] The dynamic setting of the Zernike order in step S2 includes: When the radius of curvature R ≥ 20 mm, the order k = 15 is fixed. When 5mm≤R<20mm, follow calculate; When R < 5 mm, perform piecewise fitting: Divide the region into 5×5mm sub-regions centered on the curvature extremum point; Calculate the k value independently for each sub-region.

[0023] The generation of the phase height map in step S2 includes: The Levenberg-Marquardt algorithm is used to solve for the Zernike coefficients. :

[0024] The fitting residual is ≤0.12μm.

[0025] The pre-trained model in step S3 must meet the following requirements: The training dataset contains ≥2000 microcrack samples; Characteristic decoupling orthogonality ≥ 85%.

[0026] The confidence score calculation in step S4 includes: right A high confidence defect greater than 0.7 results in a direct output C=. ; For 0.5≤ For regions ≤0.7, an enhanced formula is used: C= ×(2− )×[1+0.1×sign( -1.2).

[0027] The cooperative closed-loop control in step S6 includes: Re-inspection trigger condition: There are 3 consecutive pixels that satisfy 0.5 < <0.7; Light source enhancement constraints: Power increase , =0.05, =100 Lux; The duration of a single re-inspection is ≤50ms; Results update mechanism: Only pixels with a confidence level increase of ≥0.15 in the original report are replaced; It also includes a timeout interrupt policy: Set the total detection time threshold Tmax = 300ms; When the cumulative re-inspection time is greater than or equal to Tmax: Freeze all control actions; Output the current best detection report; Mark the timeout status code.

[0028] A surface defect detection system for automotive component injection molds based on image analysis, comprising: The vibration-resistant interferometric imaging module includes a laser source and a piezoelectric displacement device; The surface deformation analysis module includes an industrial camera and a Zernike processor; The defect feature decoupling module includes a dual-branch CNN computation unit; The detection result generation module includes: a defect coordinate locator, outputting (x, y, d); a type classifier, which matches pre-stored defect templates based on feature vectors; and a confidence synthesizer, which calculates C= ×(2− The collaborative control execution module, which includes a light intensity modulator, is activated only during re-inspection.

[0029] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

Claims

1. A method for detecting injection molds for automotive parts based on image analysis, characterized in that, Includes the following steps: S1, Anti-vibration Interference Imaging: The multi-spectral laser source output light is decomposed by a beam splitter to generate reference light and probe light. Based on the feedback signal of the triaxial accelerometer, the piezoelectric displacement device is driven to maintain the vibration acceleration ≤0.1g. S2. Dynamic Curvature Compensation: A high-speed industrial camera with a 300mm object distance is used to acquire interference fringe field of view covering 200×200mm. The curvature data of the mold CAD is retrieved and calculated according to the formula. The Zernike order is dynamically set, where R is the minimum radius of curvature in mm. S3. Orthogonal Feature Decoupling: Input the phase height map into a two-branch convolutional neural network: Texture branching extracts the basis features F through dilation=2 dilated convolution. texture ; Defect branches extract defect features F through deformable convolution. defect ; Based on cross-correlation loss function Perform feature separation; S4. Confidence-driven decision: Calculate the defect feature strength S defect (x,y) and texture entropy value E texture (x,y); When S defect >0.7 and E texture When the value is less than 1.5, a defect coordinate map is generated; S5. Output of Detection Results: Outputs a defect distribution map with confidence level labels, including: Defect location coordinates (x, y) and dimension d; Defect type classification: microcracks / scratches / pits; Confidence score C = S defect ×(2-E texture ); S6, Cooperative Control: When there is 0.5 defect When the confidence level is <0.7:​ Real-time monitoring of vibration acceleration a and ambient light intensity I env ; If a≤0.1g, local light source enhancement is triggered, increasing the power of the marked area by 20%; Return to step S2 to perform a re-inspection and update the output of step S5.

2. The image analysis-based method for detecting automotive component injection molds according to claim 1, characterized in that: The dynamic setting of the Zernike order in step S2 includes: When the radius of curvature R ≥ 20 mm, the order k is fixed at 15; When 5mm≤R<20mm, according to calculate; When R < 5 mm, perform piecewise fitting: Divide the region into 5×5mm sub-regions centered on the curvature extremum point; Calculate the k value independently for each sub-region.

3. The image analysis-based method for detecting automotive component injection molds according to claim 1, characterized in that: The generation of the phase height map includes: The Levenberg-Marquardt algorithm is used to solve for the Zernike coefficients a. n : The fitting residual is ≤0.12μm. The pre-trained model in step S3 must meet the following requirements: The training dataset contains ≥2000 microcrack samples; Characteristic decoupling orthogonality ≥ 85%.

4. The image analysis-based method for detecting automotive component injection molds according to claim 1, characterized in that: The confidence score calculation in step S4 includes: For S defect For high confidence defects >0.7, directly output C=S. defect ; For 0.5≤S defect For regions ≤0.7, an enhanced formula is used: C=S defect ×(2-E texture )×[1+0.1×sign(E texture -1.2)]。 5. The image analysis-based method for detecting automotive component injection molds according to claim 1, characterized in that: The cooperative closed-loop control in step S6 includes: Re-inspection trigger condition: There are 3 consecutive pixels that meet the 0.5 standard. defect <0.7;​ Light source enhancement constraints: Power increase γ = 0.05, I0 = 100 Lux; The duration of a single re-inspection is ≤50ms; Results update mechanism: Only replace pixels in the original report whose confidence level has increased by ≥0.

15.

6. The image analysis-based method for detecting automotive component injection molds according to claim 1, characterized in that: It also includes a timeout interrupt policy: Set the total detection time threshold Tmax = 300ms; When the cumulative re-inspection time is greater than or equal to Tmax: freeze all control actions; output the current best inspection report; mark the timeout status code.

7. A surface defect detection system for automotive component injection molds based on image analysis, used to implement the detection method according to any one of claims 1-6, characterized in that, include: The vibration-resistant interferometric imaging module includes a laser source and a piezoelectric displacement device; The surface deformation analysis module includes an industrial camera and a Zernike processor; The defect feature decoupling module includes a dual-branch CNN computation unit; The detection result generation module includes: a defect coordinate locator, outputting (x, y, d); a type classifier, which matches pre-stored defect templates based on feature vectors; and a confidence synthesizer, which calculates C = S. defect ×(2-E texture ); The collaborative control execution module, which includes a light intensity modulator, is activated only during re-inspection.

Images