Industrial part surface defect vision detection system

Abstract

This invention discloses a visual inspection system for surface defects in industrial parts, relating to the field of parts quality control technology. The system adopts a four-layer architecture: a thermoelectric coupling field control layer, a texture-encoded defect development and acquisition layer, a cross-scale linkage modeling layer, and a process closed-loop control layer. The core components include a field-induced texture-encoded defect development unit, a texture-encoded defect cause mapping unit, a cross-scale closed-loop control unit, and a multi-dimensional collaborative verification unit. This invention utilizes the core technologies of active thermoelectric coupling field control + reuse of processing texture process codes, field-induced development of micro-defects, and cross-scale closed-loop control to solve the problems of existing detection methods, such as the inability to trace the cause of defects, high false negative rates for nanoscale micro-defects, disconnect between detection and process optimization, high system costs, and dependence on constant temperature environments. It achieves simultaneous and accurate completion of defect detection and cause tracing, significantly reduces the false negative rate of micro-defects, enables closed-loop control of the entire detection-process optimization process, reduces system costs, and adapts to non-constant temperature industrial scenarios.

Description

Technical Field

[0001] This invention relates to the field of parts quality control technology, specifically to a visual inspection system for surface defects in industrial parts. Background Technology

[0002] In the field of intelligent manufacturing, surface defect detection of industrial parts is a crucial step in ensuring product quality, especially in high-end manufacturing scenarios such as aerospace and precision equipment, where surface defects directly affect the safety and reliability of product service. With the development of detection technology, AI-assisted design drawing defect detection methods based on 3D modeling have gradually become a research hotspot. The core idea is to reconstruct the surface morphology of parts through 3D modeling and then compare the design drawings with the 3D model of the actual parts using AI algorithms to achieve automated defect detection.

[0003] Currently, while existing AI inspection technologies based on 3D modeling offer significant improvements over traditional manual and 2D visual inspections, accurately reconstructing macroscopic defect morphologies on part surfaces using 3D models and enhancing inspection efficiency and accuracy with AI algorithms, several limitations remain: First, surface textures are generally considered interference factors, and their influence is eliminated through filtering during 3D modeling and AI comparison. However, the underlying defect-causing information contained in texture distortion is not recognized, leading to detection only identifying the presence of defects without tracing the dynamic process of defect formation. When batches of defects occur, indirect information such as processing records and equipment parameters must be relied upon for investigation, resulting in low efficiency and a high risk of errors. Second, existing 3D modeling and AI inspection technologies struggle to capture microscopic defects such as nanoscale microcracks and micropores due to insufficient optical contrast on the part surface, leading to a high rate of missed detection for these latent defects. Third, existing inspection systems assume a constant ambient temperature and eliminate temperature interference by constructing temperature-controlled chambers, failing to consider the regulatory effect of temperature fields on the optical properties of part surfaces and thus unable to utilize thermodynamic principles to improve the detectability of microscopic defects.

[0004] Furthermore, existing AI-based defect detection methods for 3D modeling-assisted design drawings are still limited to defect identification and fail to achieve synergistic integration of detection with process optimization and quality traceability, making it difficult to meet the refined quality control requirements of the high-end manufacturing sector. Therefore, this paper proposes a visual inspection system for surface defects in industrial parts to overcome these problems. Summary of the Invention

[0005] The purpose of this invention is to provide a visual inspection system for surface defects of industrial parts to solve the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the present invention provides a visual inspection system for surface defects of industrial parts. The system adopts a four-layer architecture, which consists of a thermoelectric coupling field control layer, a texture coding defect development and acquisition layer, a cross-scale linkage modeling layer, and a process closed-loop control layer. The core includes a field-induced texture coding defect development unit, a texture coding defect cause mapping unit, a cross-scale closed-loop control unit, and a multi-dimensional collaborative verification unit. Each layer forms an organic whole through dynamic interaction of encoding, decoding, feedback, and regulation. Each core unit communicates with the data switch via an industrial bus. The thermoelectric coupling field regulation layer and the texture encoding defect development and acquisition layer work synchronously at 200Hz frequency through a synchronous trigger line. The cross-scale linkage modeling layer communicates with the texture encoding defect development and acquisition layer and the process closed-loop regulation layer through an industrial 5G link. The thermoelectric coupling field actively regulates the physical properties of the part surface, enabling the processing texture to have process parameter encoding function, and micro-defects to generate visually identifiable development signals. At the same time, the cause of defects is traced through encoded texture decoding, achieving synchronous collaboration of defect detection, cause tracing, and process optimization.

[0007] Furthermore, the field-induced texture coding defect development unit includes an array-type thermoelectric synergistic control module, a flexible bonding electrode array, an infrared polarized light synergistic acquisition module, and a coupled field parameter closed-loop controller. The array-type thermoelectric synergistic control module consists of 128×128 independently controllable thermoelectric composite units, each unit integrating a Peltier thermoelectric element and a micro piezoelectric ceramic electrode, connected to the coupled field parameter closed-loop controller via an SPI bus. The flexible bonding electrode array uses a polyimide substrate and is fixed to the periphery of the part-bearing platform via elastic connectors. The infrared polarized light synergistic acquisition module integrates an infrared thermal imager with a pixel resolution of 1 micrometer and a polarized light microscope with a resolution of 0.05 nanometers. The two are connected via a trigger signal line to complete 200Hz synchronous trigger acquisition, and are fixed 30cm directly above the part-bearing platform by a bracket. The coupled field parameter closed-loop controller has a built-in material-coupled field parameter encoding rule mapping library and completes data interaction with other units via an industrial Ethernet interface.

[0008] Furthermore, the texture coding defect cause mapping unit includes a coding texture decoding module, a process coding-defect mapping database, and a chaotic neural network inference engine. The coding texture decoding module is connected to the infrared polarized light collaborative acquisition module via a PCIe interface, and has built-in coding rules consistent with the coupled field parameter closed-loop controller. It completes decoding by extracting the surface potential distortion parameters of the texture region. The process coding-defect mapping database uses a MySQL database, which is installed in the industrial control computer and connected to the coding texture decoding module and the chaotic neural network inference engine via Ethernet. It stores the coding texture features, defect types, and cause data corresponding to different materials and different process parameter deviations, and supports real-time iterative updates using SQL statements. The chaotic neural network inference engine uses an FPGA chip as the core processing unit and has a built-in feature-deviation-cause three-order inference model.

[0009] Furthermore, the cross-scale closed-loop control unit includes a cross-scale information conversion module, an industrial 5G edge computing communication module, an adaptive process parameter adjustment engine, and a batch defect prediction module. The cross-scale information conversion module is connected to the texture-encoded defect cause mapping unit via a serial port, converting defect causes and parameter deviation quantization values ​​into macroscopic process control signals in G-code format. The industrial 5G edge computing communication module uses an industrial-grade 5G module and communicates with the CNC system of the processing equipment via a 5G network. The adaptive process parameter adjustment engine has a built-in cause-control parameter matching model and is connected to the industrial 5G edge computing communication module via a bus, directly interfacing with the CNC system of the processing equipment. The batch defect prediction module uses an embedded microprocessor and is connected to the texture-encoded defect cause mapping unit via Ethernet.

[0010] Furthermore, the multi-dimensional collaborative verification unit includes an atomic force microscope microscopic verification module, a process parameter accuracy detection module, and a part mechanical property testing module. The detection probe of the atomic force microscope microscopic verification module is aligned with the part support platform through a displacement platform and connected to the industrial control computer through a data cable. The process parameter accuracy detection module collects real-time process parameters of the processing equipment through speed sensors and flow sensors and transmits them to the industrial control computer via Ethernet. The part mechanical property testing module includes a tensile testing machine and a fatigue testing machine, and the test results are transmitted to the industrial control computer.

[0011] Furthermore, the preset coding rule of the coupled field parameter closed-loop controller is that the cutting speed corresponds to the thermal field gradient and the feed rate corresponds to the electric field strength, which is used to correspond to the unique surface potential distortion mode for different combinations of process parameters.

[0012] Furthermore, the process parameter deviations stored in the process coding-defect mapping database include cutting speed deviation ±5% and feed rate deviation ±3%, and the surface potential distortion parameters extracted by the coding texture decoding module include distortion amplitude, distribution density, and gradient direction.

[0013] Furthermore, the batch defect prediction module analyzes the defect causes and parameter deviation trends of 10 consecutive parts. When the frequency of defects with the same cause reaches 3% or more, it outputs a prediction warning and triggers preventive adjustment of process parameters.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. Highly efficient and accurate cause tracing: The processing texture is transformed into a process parameter encoding carrier. By decoding the encoded texture, defects and process deviations are directly linked. There is no need to rely on indirect process records, which greatly improves the tracing efficiency and accuracy.

[0015] 2. Upgraded micro-defect detection capability: By actively developing micro-defects through thermoelectric coupling field and combining it with infrared polarized light for collaborative acquisition, the missed detection rate of latent defects such as nanoscale microcracks and micropores is reduced, and the detection accuracy is greatly improved.

[0016] 3. Achieve closed-loop control throughout the entire process: Relying on industrial 5G to build a cross-scale dynamic feedback link, the causes of defects are transformed into process control instructions in real time. Combined with batch defect prediction, the batch defect incidence rate is significantly reduced, and the mechanical properties of parts such as fatigue strength are significantly improved after control.

[0017] 4. Low system cost and strong environmental adaptability: It does not rely on 3D modeling, AI comparison algorithms and constant temperature chambers. High-precision detection is achieved through conventional infrared polarization light equipment and thermoelectric coupling field control, which greatly reduces the system cost and can work stably in non-constant temperature environments. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the visual inspection system for surface defects of industrial parts according to the present invention. Detailed Implementation

[0019] 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.

[0020] Please see Figure 1 The present invention provides a technical solution: See Figure 1 The following is an example of a visual inspection system for surface defects in industrial parts: I. System Overall Architecture: This industrial part surface defect visual inspection system adopts a four-layer architecture: thermoelectric coupling field control layer, texture encoding defect development and acquisition layer, cross-scale linkage modeling layer, and process closed-loop control layer. The core includes a field-induced texture encoding defect development unit, a texture encoding defect cause mapping unit, a cross-scale closed-loop control unit, and a multi-dimensional collaborative verification unit. Each layer forms an organic whole through dynamic interaction of encoding, decoding, feedback, and control. Each core unit achieves data communication with an industrial bus and data switch. Specifically, the thermoelectric coupling field control layer and the texture encoding defect development and acquisition layer operate synchronously at 200Hz via a synchronous trigger line. The cross-scale linkage modeling layer, texture encoding defect development and acquisition layer, and process closed-loop control layer transmit data via an industrial 5G link. It is worth noting that this system does not rely on existing 3D modeling and AI comparison technologies. Instead, it actively controls the physical properties of the part surface through a thermoelectric coupling field, enabling the processed texture to have process parameter encoding capabilities. This allows microscopic defects to generate visually identifiable development signals under the action of the coupling field. Simultaneously, the decoding of the encoded texture directly traces the cause of the defect, achieving synchronous collaboration between defect detection, cause tracing, and process optimization.

[0021] II. Core Unit Structure: 1. Core mechanism carrier of field-induced texture coding defect development unit: This unit is the core of realizing the transformation of interference into resources and the active development of defects. By constructing a controllable thermoelectric coupling field, it completes the encoding of process information of the processed texture and the active development of micro defects, thus simultaneously solving the three major problems of texture interference, insufficient contrast of micro defects and failure to utilize the temperature field from a mechanism perspective.

[0022] The specific structure includes an array-type thermoelectric synergistic control module, a flexible-fit electrode array, an infrared polarization light synergistic acquisition module, and a coupled field parameter closed-loop controller. The connection relationships and parameter configurations of each component are as follows: The array-type thermoelectric synergistic control module consists of 128 x 128 independently controllable thermoelectric composite units. Each unit integrates a Peltier thermoelectric element and a micro piezoelectric ceramic electrode. Each thermoelectric composite unit is connected to the coupled field parameter closed-loop controller via an SPI bus. They are evenly distributed on the component support platform and its periphery, with a spacing of 2 mm between adjacent thermoelectric composite units. This achieves synchronous coupling control of the thermal field control range from -20°C to 100°C with a gradient accuracy of 0.2°C per millimeter and the electric field control range from 0 kV to 5 kV with an electric field intensity gradient accuracy of 0.1 kV per millimeter. The flexible-fit electrode array uses a polyimide substrate with a substrate thickness of 0.1 mm and a copper-plated electrode layer thickness of [missing information]. The sensor has a resolution of 50 micrometers and is fixed to the periphery of the part-bearing platform via elastic connectors. It can adaptively conform to the surface contour of the part, ensuring that the electric field is uniformly applied to the part surface. The infrared polarization light collaborative acquisition module integrates a high-resolution infrared thermal imager and a polarization light microscope. The two are connected by a trigger signal line to achieve synchronous triggering acquisition. The sampling frequency is set to 200Hz. The infrared thermal imager has a pixel resolution of 1 micrometer, and the polarization light microscope has a resolution of 0.05 nanometers. The acquisition module is fixed 30cm directly above the part-bearing platform by a bracket. The coupling field parameter closed-loop controller has a built-in material coupling field parameter encoding rule mapping library. The controller interacts with other units through an industrial Ethernet interface and automatically matches the optimal thermoelectric coupling parameters according to the part material.

[0023] The working principle is based on the surface potential distortion effect caused by thermoelectric coupling and the field polarization difference effect caused by defect region, so as to realize the simultaneous completion of texture encoding and defect development. The specific process is as follows: ① Field-induced texture encoding: The topology of the machined texture is determined by the machining process parameters, including cutting speed, feed rate, and tool pressure. Under the action of a thermoelectric coupling field, texture regions with different topologies will exhibit surface potential distortions of varying magnitudes due to differences in thermal conductivity and dielectric constant. This unit presets encoding rules through a closed-loop controller of the coupling field parameters. The encoding rules are: cutting speed corresponds to thermal field gradient, and feed rate corresponds to electric field strength. This ensures that different combinations of process parameters correspond to a unique surface potential distortion pattern, thus completing the encoding of the machining texture's process information. At this point, the machining texture is no longer an interference but a carrier of encoded process information.

[0024] ② Active Defect Development: Microscopic defects, including nanoscale microcracks and micropores, exhibit significant differences in thermoelectric and physical properties compared to the substrate material. These thermoelectric and physical properties include the coefficient of thermal expansion and dielectric constant. Under the influence of a thermoelectric coupling field, a local field distortion region forms in the defect area. On one hand, the field distortion leads to a difference in thermal radiation intensity between the defect area and the substrate, resulting in a clear grayscale contrast under infrared thermal imaging (development signal one). On the other hand, the polarization charge distribution in the defect area differs from that in the substrate, causing a specific polarization state deflection of polarized light in the defect area. This alternating bright and dark polarization development signal can be captured using a polarized light microscope (development signal two). The superposition of these two development signals completely solves the problem of insufficient natural optical contrast in microscopic defects, enabling active development and precise localization of defects.

[0025] 2. Texture Coding Defect Cause Mapping Unit: Core Unit for Cause Tracing This unit directly establishes a mapping relationship between the causes of defects caused by deviations in the coding feature process parameters by decoding the coded texture, thereby solving the problem of inefficient reliance on indirect data for tracing the causes of defects from the root.

[0026] The specific structure includes an encoding texture decoding module, a process encoding defect mapping database, a chaotic neural network inference engine. The connection relationships and parameter configurations of each component are as follows: The encoded texture decoding module is connected to the infrared polarized light co-acquisition module through the PCIe interface, receives texture image data transmitted by the acquisition module, and completes decoding by extracting the surface potential distortion parameters of the texture region. The surface potential distortion parameters include the distortion amplitude distribution, density, gradient direction, etc. The decoding module has built-in encoding rules consistent with the coupled field parameter closed-loop controller and outputs real-time process parameters. The process encoded defect mapping database adopts a MySQL database, installed in the industrial control computer, and is connected to the encoded texture decoding module and the chaotic neural network inference engine through Ethernet. The database is constructed through a large number of orthogonal experiments and stores the encoded texture feature defect types and cause data corresponding to different materials and different process parameter deviations. The process parameter deviations include cutting speed deviation ±5% and feed rate deviation ±3%. The database supports real-time iterative updates of data through SQL statements. The chaotic neural network inference engine uses an FPGA chip as the core processing unit. The chip model is Xilinx Zynq UltraScale+MPSoC. The engine has built-in a third-order inference model for feature deviation causes. It receives the decoded process parameter defect development signal features through a high-speed interface and outputs the defect causes and process parameter deviation quantification values.

[0027] When the infrared polarized light co-acquisition module detects a defect development signal, the encoded texture decoding module simultaneously receives the encoded texture image of the defect area and its surroundings. After extracting the surface potential distortion parameters, it performs decoding by comparing them with the encoding rules, obtaining the real-time process parameters at the time of defect generation. The encoded texture decoding module compares the real-time process parameters with the preset standard process parameters to obtain the parameter deviation value. The chaotic neural network inference engine calls the data in the process encoded defect mapping database and directly associates the parameter deviation with the cause of the defect through a third-order inference model. For example, when the decoding shows that the cutting speed is 8% higher than the standard value, and the defect development signal is a linear infrared grayscale contrast plus a polarization state deflection angle of 30 degrees, it is directly determined that the defect is caused by thermal stress nanoscale microcracks due to excessive cutting speed, without relying on any indirect data such as process records, thus achieving real-time and accurate tracing of the cause of the defect.

[0028] 3. Core of cross-scale closed-loop control unit detection process collaboration: This unit constructs a cross-scale dynamic feedback link between microscopic defect information and macroscopic process parameters, transforming defect cause information into process control instructions in real time, thereby addressing the root cause of batch defects caused by process optimization lag.

[0029] The specific structure includes a cross-scale information conversion module, an industrial 5G edge computing communication module, an adaptive process parameter adjustment engine, and a batch defect prediction module. The connection relationships and parameter configurations of each component are as follows: The cross-scale information conversion module connects to the texture-encoded defect cause mapping unit via a serial port, receives the quantized value of the defect cause parameter deviation, and converts it into a macroscopic process control signal that can be recognized by the processing equipment. The conversion uses a standardized G-code format. The industrial 5G edge computing communication module uses an industrial-grade 5G module, model Huawei ME909S-821. The module communicates with the CNC system of the processing equipment via the 5G network to achieve millisecond-level transmission of control signals, with transmission latency controlled within 5 milliseconds, avoiding the transmission latency of traditional industrial Ethernet. The adaptive process parameter adjustment engine has built-in cause control... The parameter matching model connects the engine to the industrial 5G edge computing communication module via a bus. It automatically generates the optimal process adjustment scheme based on the cause of defects. The adjustment scheme includes specific values ​​for parameters such as cutting speed and coolant flow rate. The engine directly interfaces with the CNC system of the CNC grinding machine. The batch defect prediction module uses an embedded microprocessor, model STM32H743. The module connects to the texture encoding defect cause mapping unit via Ethernet. By analyzing the defect causes and parameter deviation trends of 10 consecutive parts, when the frequency of defects with the same cause reaches 3% or more, it outputs a prediction warning in advance and triggers preventive adjustment of process parameters.

[0030] After the texture encoding defect cause mapping unit outputs the defect cause and parameter deviation, the cross-scale information conversion module immediately receives it and converts it into a macroscopic process control signal. For example, the cutting speed is reduced from 150 meters per minute to 135 meters per minute, and the coolant flow rate is increased from 10 liters per minute to 15 liters per minute. The industrial 5G edge computing communication module transmits the control signal to the adaptive process parameter adjustment engine, which drives the CNC system of the processing equipment to complete the real-time adjustment of process parameters. At the same time, the batch defect prediction module continuously receives defect data of continuous parts, performs statistical analysis on the defect cause and parameter deviation trend, and when the frequency of scratch defects caused by tool wear reaches 4%, an early warning signal is immediately issued, driving the adaptive process parameter adjustment engine to adjust the tool compensation amount in advance to avoid the generation of batch defects, thus realizing a closed-loop system for detection, cause control and prediction.

[0031] 4. Multi-dimensional collaborative verification unit effect guarantee unit: This unit is used to verify the accuracy of defect detection, the accuracy of cause tracing, and the effect of process control, avoiding errors in a single detection dimension and ensuring the overall reliability of the system. The specific structure includes an atomic force microscope (AFM) microscopic verification module, a process parameter accuracy detection module, and a part mechanical property testing module. The connection relationships and parameter configurations of each component are as follows: The AFM microscopic verification module's detection probe is aligned with the part's support platform via a displacement platform with a positioning accuracy of 0.01 micrometers. The module connects to an industrial control computer via a data cable to accurately measure the morphology of the developed microscopic defects, verifying the detection accuracy. The process parameter accuracy detection module collects real-time process parameters from the processing equipment through sensors, including a speed sensor and a flow sensor, with a measurement accuracy of 0.1%. The module transmits the collected data to the industrial control computer via Ethernet and compares it with the parameter values ​​in the control commands to verify the control accuracy. The part mechanical property testing module includes a tensile testing machine and a fatigue testing machine. The maximum test force of the tensile testing machine is 100 kN, and the frequency range of the fatigue testing machine is 0.1 Hz to 50 Hz. The module performs tensile fatigue tests on the processed parts after control and transmits the test results to the industrial control computer to verify the effect of process optimization on improving part quality.

[0032] III. System Workflow: This system follows a dynamic cyclic process of field-induced modulation encoding, development, decoding, modulation, and verification. Each step works in tandem to simultaneously resolve all existing problems. The specific steps are as follows: 1. Initialization configuration: Based on the material standard process parameters of the part to be tested, the operator sends a configuration command to the coupling field parameter closed-loop controller through the industrial control computer. The coupling field parameter closed-loop controller calls the built-in mapping library to match the optimal thermoelectric coupling field parameters and texture encoding rules to complete the system initialization. 2. Field-induced texture encoding and defect development: The coupled field parameter closed-loop controller sends a start command to the array-type thermoelectric co-control module to build a thermoelectric coupled field on the surface of the part, and at the same time completes the encoding of the process information of the processed texture; the infrared polarized light co-acquisition module is started synchronously to capture the infrared polarized light development signal generated by micro-defects; 3. Encoding, Decoding and Cause Tracing: The infrared polarized light collaborative acquisition module transmits the encoded texture image and the defect development image to the encoding texture decoding module. The encoding texture decoding module completes the decoding and outputs real-time process parameters. The chaotic neural network inference engine receives the real-time process parameters and defect development signal characteristics, calls the process encoding defect mapping database to complete cause tracing, and outputs the defect cause and parameter deviation quantification value. 4. Cross-scale closed-loop control: The cross-scale information conversion module receives the defect causes and parameter deviation quantification values ​​and converts them into process control signals. These signals are then transmitted to the adaptive process parameter adjustment engine via the industrial 5G edge computing communication module. The engine drives the processing equipment to complete parameter adjustment. The batch defect prediction module simultaneously monitors the defect trends of continuous parts and executes preventive early warnings. 5. Multi-dimensional verification and iteration: The atomic force microscope microscopic verification module, the process parameter accuracy detection module, and the part mechanical property testing module each carry out corresponding tests. The verification results are transmitted to the industrial control computer, which feeds back the verification results to the process coding defect mapping database, updates the coding rules and mapping model, and improves system performance.

[0033] Summarize: The process information encoding of the processed texture is reused. The encoded texture is directly used as the direct data carrier for tracing the cause of defects. The efficiency of cause tracing is greatly improved compared with the existing technology, the accuracy of tracing is improved, and there is no need to rely on any indirect process records. Active development of micro-defects improves the accuracy of detecting micro-defects such as nano-cracks and micropores, and reduces the false negative rate. It enables cross-scale real-time collaboration in detecting the causes of defects and predicting process defects, significantly reducing the incidence of batch defects, while the mechanical properties of the adjusted parts, such as fatigue strength, are significantly improved. Without relying on existing 3D modeling AI comparison algorithms and high-end detection instruments, high-precision detection can be achieved through thermoelectric coupling field control and conventional infrared polarized light acquisition equipment. The system cost is significantly reduced compared to existing high-end detection equipment, and it also has stronger environmental adaptability, and can work stably in conventional non-constant temperature environments.

Claims

1. A visual inspection system for surface defects of industrial parts, characterized in that, The system adopts a four-layer architecture: thermoelectric coupling field control layer, texture coding defect development and acquisition layer, cross-scale linkage modeling layer, and process closed-loop control layer. The core includes field-induced texture coding defect development unit, texture coding defect cause mapping unit, cross-scale closed-loop control unit, and multi-dimensional collaborative verification unit. Each layer forms an organic whole through dynamic interaction of encoding, decoding, feedback, and regulation. Each core unit communicates with the data switch via an industrial bus. The thermoelectric coupling field regulation layer and the texture encoding defect development and acquisition layer work synchronously at 200Hz frequency through a synchronous trigger line. The cross-scale linkage modeling layer communicates with the texture encoding defect development and acquisition layer and the process closed-loop regulation layer through an industrial 5G link. The thermoelectric coupling field actively regulates the physical properties of the part surface, enabling the processing texture to have process parameter encoding function, and micro-defects to generate visually identifiable development signals. At the same time, the cause of defects is traced through encoded texture decoding, achieving synchronous collaboration of defect detection, cause tracing, and process optimization.

2. The visual inspection system for surface defects of industrial parts as described in claim 1, characterized in that: The field-induced texture coding defect development unit includes an array-type thermoelectric co-control module, a flexible bonding electrode array, an infrared polarized light co-acquisition module, and a coupled field parameter closed-loop controller. The array-type thermoelectric co-control module consists of 128×128 independently controllable thermoelectric composite units, each integrating a Peltier thermoelectric element and a micro piezoelectric ceramic electrode, and is connected to the coupled field parameter closed-loop controller via an SPI bus. The flexible bonding electrode array uses a polyimide substrate and is fixed to the periphery of the part-bearing platform by elastic connectors. The infrared polarized light co-acquisition module integrates an infrared thermal imager with a pixel resolution of 1 micrometer and a polarized light microscope with a resolution of 0.05 nanometers. The two are connected by a trigger signal line to complete 200Hz synchronous trigger acquisition, and are fixed 30cm directly above the part-bearing platform by a bracket. The coupled field parameter closed-loop controller has a built-in material-coupled field parameter encoding rule mapping library and completes data interaction with other units through an industrial Ethernet interface.

3. The visual inspection system for surface defects of industrial parts as described in claim 1, characterized in that: The texture coding defect cause mapping unit includes a texture coding decoding module, a process coding-defect mapping database, and a chaotic neural network inference engine. The texture coding decoding module is connected to the infrared polarized light collaborative acquisition module via a PCIe interface. It has built-in coding rules consistent with the coupled field parameter closed-loop controller and completes decoding by extracting the surface potential distortion parameters of the texture region. The process coding-defect mapping database uses a MySQL database, which is installed in the industrial control computer and connected to the texture coding decoding module and the chaotic neural network inference engine via Ethernet. It stores the coded texture features, defect types, and cause data corresponding to different materials and different process parameter deviations, and supports real-time iterative updates using SQL statements. The chaotic neural network inference engine uses an FPGA chip as the core processing unit and has a built-in feature-deviation-cause three-order inference model.

4. The visual inspection system for surface defects of industrial parts as described in claim 1, characterized in that: The cross-scale closed-loop control unit includes a cross-scale information conversion module, an industrial 5G edge computing communication module, an adaptive process parameter adjustment engine, and a batch defect prediction module. The cross-scale information conversion module is connected to the texture coding defect cause mapping unit through a serial port, converting the defect cause and parameter deviation quantization value into macroscopic process control signals in G-code format. The industrial 5G edge computing communication module adopts an industrial-grade 5G module and communicates with the CNC system of the processing equipment through a 5G network. The adaptive process parameter adjustment engine has a built-in cause-control parameter matching model, which is connected to the industrial 5G edge computing communication module via bus and directly interfaces with the CNC system of the processing equipment. The batch defect prediction module uses an embedded microprocessor and is connected to the texture-encoded defect cause mapping unit via Ethernet.

5. The visual inspection system for surface defects of industrial parts as described in claim 1, characterized in that: The multi-dimensional collaborative verification unit includes an atomic force microscope microscopic verification module, a process parameter accuracy detection module, and a part mechanical property testing module. The detection probe of the atomic force microscope microscopic verification module is aligned with the part support platform through a displacement platform and connected to the industrial control computer through a data cable. The process parameter accuracy detection module collects real-time process parameters of the processing equipment through speed sensors and flow sensors and transmits them to the industrial control computer via Ethernet. The part mechanical property testing module includes a tensile testing machine and a fatigue testing machine, and the test results are transmitted to the industrial control computer.

6. The visual inspection system for surface defects of industrial parts as described in claim 2, characterized in that: The preset coding rule of the coupled field parameter closed-loop controller is that the cutting speed corresponds to the thermal field gradient and the feed rate corresponds to the electric field strength, which is used to correspond to a unique surface potential distortion mode for different combinations of process parameters.

7. The visual inspection system for surface defects of industrial parts as described in claim 3, characterized in that: The process parameter deviations stored in the process coding-defect mapping database include cutting speed deviation ±5% and feed rate deviation ±3%. The surface potential distortion parameters extracted by the coding texture decoding module include distortion amplitude, distribution density, and gradient direction.

8. The visual inspection system for surface defects of industrial parts as described in claim 4, characterized in that: The batch defect prediction module analyzes the defect causes and parameter deviation trends of 10 consecutive parts. When the frequency of defects with the same cause reaches 3% or more, it outputs a prediction warning and triggers preventive adjustment of process parameters.

Images