Jade internal crack intelligent detection and carving path optimization system and method thereof

Abstract

The invention relates to the technical field of jade detection, in particular to a jade internal crack intelligent detection and engraving path optimization system and a method thereof, and the method comprises multiple innovative technologies such as fusion multi-angle high-resolution image acquisition, multispectral confocal scanning, crack characteristic data analysis, engraving planning information generation, real-time engraving data acquisition and finite element simulation analysis. Meanwhile, a jade material dynamic adjustment model trained based on historical processing data further optimizes carving process parameters, the intelligence level of jade processing is comprehensively improved, and in the aspect of crack detection accuracy, the system keeps high accuracy of 95% or above on all types of samples, the material utilization rate is high, and the system is suitable for popularization and application. According to the system disclosed by the invention, all types of samples are remarkably improved by about 15% on average. The processing time is averagely shortened by about 40%; and on the most complex C-type sample, the high quality of 85% or above can still be maintained, while the mass fraction of the traditional method is remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of jade detection, in particular to an intelligent crack detection and carving path optimization system and method for jade interior cracks. Background Art

[0002] As a precious natural material, the processing of jade has always been an important topic in the field of arts and crafts. Traditional jade processing mainly relies on the experience and skills of craftsmen. Although this method can create exquisite works, it also faces problems such as low efficiency, large material waste, and unstable quality. In recent years, with the development of computer technology and automation equipment, the jade processing industry has begun to develop towards intelligence and refinement.

[0003] Currently, relatively advanced jade processing methods in the industry usually use machine vision technology for crack detection and combine simple path planning algorithms for carving. This method has indeed made certain progress compared to traditional manual processing, improving processing efficiency and accuracy. However, this method still has many deficiencies when dealing with complex jade samples.

[0004] Firstly, the existing crack detection technologies mainly rely on a single optical imaging method and are difficult to accurately identify the tiny cracks deep inside the jade. Especially for those high-value jades with a smooth surface but complex internal structures, the limitations of this detection method may lead to serious economic losses. Secondly, the current carving path planning algorithms are often preset fixed patterns and lack the dynamic adaptation ability to the material properties and crack distribution of jade. This "one-size-fits-all" method often causes unnecessary material waste in actual processing and may even cause secondary damage to the jade due to stress concentration.

[0005] In addition, the existing technologies also have deficiencies in real-time monitoring and adjustment during the carving process. Most systems lack an effective vibration suppression mechanism and are difficult to cope with various unstable factors in high-speed precision carving. At the same time, for abnormal situations that may occur during the carving process, such as crack propagation and material spalling, existing systems often have difficulty in detecting and handling them in a timely manner, increasing the risk of processing failure.

[0006] Finally, the existing jade processing systems generally lack the ability of self-learning and continuous optimization. They usually operate based on fixed algorithms and parameters and are difficult to adaptively adjust according to the characteristics of different types of jade and processing experience. This rigid processing method limits the adaptability of the system when facing diverse and personalized processing requirements.

[0007] In the field of internal crack detection, CN107655970A (Hetian jade crack detection system based on ultrasonic waves) adopts a uniaxial ultrasonic scanning scheme. When the included angle between the crack direction and the ultrasonic wave propagation direction is <15°, the detection accuracy drops to 62%, and limited by the fixed frequency design of the transducer, the signal-to-noise ratio drops sharply by 12 dB when facing high-density materials such as jadeite. More seriously, this technology relies on multi-point manual detection, and the complete detection of a single piece of 200mm×200mm jade material takes up to 15 minutes, and the efficiency is difficult to meet the requirements of industrial production.

[0008] In terms of surface defect recognition, CN116993958A (gold-inlaid jade recognition method based on image processing) realizes the surface quality inspection of gold-inlaid jade through machine vision, but its technical solution has obvious shortcomings: the detection object is limited to "product image information", resulting in a zero internal crack recognition rate; and its rigid threshold judgment mechanism has a misjudgment rate as high as 28.6% when the light change > 200 lux, and it cannot adapt to the non-destructive processing requirements of high-value jade pieces. Summary of the Invention

[0009] In view of the above problems, the present invention proposes a jade internal crack intelligent detection and carving path optimization system and its method. The system aims to improve the accuracy of crack detection through multi-spectral confocal scanning technology, use improved deep learning algorithms and intelligent optimization methods to realize the dynamic planning of carving paths, and ensure the stability and safety of the processing process through real-time monitoring and adaptive control. At the same time, the present invention also introduces digital twin and reinforcement learning technologies, enabling the system to have the ability of self-learning and continuous optimization.

[0010] The present invention proposes a jade internal crack intelligent detection and carving path optimization system and its method, including:

[0011] Jade sample detection module, for:

[0012] Collect multi-angle high-resolution images of the jade sample to be detected;

[0013] Use a multi-spectral confocal scanning device to obtain the spectral data of the internal structure of the jade;

[0014] Generate internal crack feature data of the jade based on the high-resolution images and the spectral data of the internal structure;

[0015] Parameter transfer module, coupled with the jade sample detection module, for:

[0016] Receive the internal crack feature data of the jade;

[0017] Generate carving planning information based on the internal crack feature data of the jade;

[0018] Convert the carving planning information into carving processing parameters;

[0019] An engraving processing module, coupled to the parameter transfer module, is used for:

[0020] Receiving the engraving processing parameters;

[0021] Based on the engraving processing parameters, controlling an engraving device to perform a jade carving operation;

[0022] During the engraving process, collecting real-time engraving data;

[0023] A physical simulation module, coupled to the engraving processing module, is used for:

[0024] Receiving the real-time engraving data;

[0025] Based on the real-time engraving data, performing a finite element simulation analysis;

[0026] Generating an engraving process path and historical processing data;

[0027] A data center module, coupled to the physical simulation module and the parameter transfer module, is used for:

[0028] Receiving the engraving process path and historical processing data;

[0029] Based on the engraving process path and historical processing data, training a dynamic adjustment model for jade materials;

[0030] Generating an engraving process parameter model under different material hardnesses;

[0031] Sending the engraving process parameter model to the parameter transfer module for optimizing engraving planning information.

[0032] Preferably, the jade sample detection module includes:

[0033] A multi-spectral imaging unit for collecting multi-spectral image data of a jade sample to be detected;

[0034] A confocal scanning unit for obtaining three-dimensional internal structure information of the jade;

[0035] A data fusion unit, coupled to the multi-spectral imaging unit and the confocal scanning unit, for fusing the multi-spectral image data with the three-dimensional structure information to generate comprehensive feature data;

[0036] A deep learning unit, coupled to the data fusion unit, is used for:

[0037] Receiving the comprehensive feature data;

[0038] Using an improved U-Net++ network structure to analyze the comprehensive feature data;

[0039] Output the detection results of internal cracks in jade.

[0040] Preferably, the improved U-Net++ network structure includes:[[]]

[0041] A shallow feature extraction module for extracting low-level features of jade images through multi-layer convolution operations;

[0042] A semantic feature fusion module, coupled with the shallow feature extraction module, for:[[]]

[0043] Receiving the low-level features;

[0044] Fusing feature information of different resolutions through five feature addition modules;

[0045] Outputting the fused multi-scale features;

[0046] A global feature extraction module, coupled with the semantic feature fusion module, for:[[]]

[0047] Receiving the fused multi-scale features;

[0048] Extracting global context information through a serial 1×1 convolutional layer and global pooling layer;

[0049] Outputting the final crack detection feature map.

[0050] Preferably, the parameter transfer module includes:[[]]

[0051] A carving planning sub-module for:[[]]

[0052] Receiving the internal crack feature data of the jade;

[0053] Generating an initial carving path based on an improved Monte Carlo tree search algorithm;

[0054] Optimizing the initial carving path in combination with the material hardness distribution information;

[0055] A parameter conversion sub-module, coupled with the carving planning sub-module, for:[[]]

[0056] Receiving the optimized carving path;

[0057] Converting the carving path into specific carving processing parameters, including tool path, feed speed, and cutting depth;

[0058] A YOLOv8 object detection module, coupled with the carving planning sub-module, for:[[]]

[0059] Using a pre-trained improved U-Net++ model as a feature extractor;

[0060] Based on the jade image data, perform real-time crack positioning;

[0061] Feed back the crack positioning result to the carving planning sub-module for dynamically adjusting the carving path.

[0062] Preferably, the carving processing module includes:

[0063] A robotic arm control unit for controlling the movement of the carving robotic arm according to the carving processing parameters;

[0064] A tool management unit coupled to the robotic arm control unit for:

[0065] Selecting a suitable carving tool;

[0066] Controlling the replacement and adjustment of the tool;

[0067] An adaptive vibration suppression unit coupled to the robotic arm control unit and the tool management unit for:

[0068] Real-time detecting the tool vibration signal;

[0069] Calculating the vibration frequency and intensity based on the detected vibration signal;

[0070] When the vibration intensity exceeds a preset threshold, adaptively adjusting the tool rotation speed and movement direction;

[0071] A data acquisition unit for acquiring real-time processing data during carving, including tool position, cutting force, and material removal rate.

[0072] Preferably, the adaptive vibration suppression unit is specifically used for:

[0073] Smoothing the vibration signal by using a digital filter;

[0074] Performing a Fourier transform on the smoothed vibration signal to obtain the vibration fundamental frequency;

[0075] Calculating the vibration frequency and vibration intensity based on the vibration fundamental frequency;

[0076] When the vibration intensity exceeds a preset threshold V0, perform the following operations:

[0077] Changing the tool movement direction parameter;

[0078] Adjusting the tool rotation speed parameter;

[0079] Smoothing the threshold V0 by using a sine fitting method;

[0080] Based on the smoothed threshold, realizing the adaptive compensation of vibration.

[0081] Preferably, the physical simulation module includes:

[0082] A finite element analysis unit for:

[0083] Constructing a physical model of the jade carving process based on real-time carving data;

[0084] Performing stress analysis and deformation simulation;

[0085] A process path generation unit, coupled to the finite element analysis unit, for:

[0086] Generating an optimized carving process path based on the simulation results;

[0087] Recording the specific parameters of each carving step;

[0088] A historical data management unit, coupled to the process path generation unit, for:

[0089] Storing and managing historical processing data;

[0090] Providing data query and analysis interfaces.

[0091] 8. The system according to claim 1, characterized in that it further includes:

[0092] An acoustic emission monitoring module, coupled to the carving processing module, for:

[0093] Collecting acoustic emission signals during the carving process through an acoustic emission sensor;

[0094] Extracting the feature vectors of the acoustic emission signals using the Hilbert-Huang transform method;

[0095] Inputting the feature vectors into an SVM classifier to perform crack anomaly identification;

[0096] When an anomaly is detected, sending an alarm signal to the carving processing module.

[0097] Preferably, the data center module further includes:

[0098] A digital twin sub-module for:

[0099] Constructing a digital twin model of the jade carving process based on the carving process path and historical processing data;

[0100] Updating the digital twin model in real time to reflect the state of the physical carving process;

[0101] Providing a visual interface to display the real-time dynamics of the carving process;

[0102] A reinforcement learning sub-module, coupled to the digital twin sub-module, for:

[0103] Train a reinforcement learning agent using the simulation environment provided by the digital twin model;

[0104] Optimize the utilization rate of rough materials and generate the optimal carving path and tool parameters;

[0105] Feed back the optimization result to the parameter transfer module.

[0106] An intelligent detection method for internal cracks of jade and an optimization method for carving paths, based on the above system, includes the following steps:

[0107] S1. Collect multi-angle high-resolution images of the jade sample to be inspected, and use a multi-spectral confocal scanning device to obtain the spectral data of the internal structure of the jade;

[0108] S2. Based on the high-resolution images and the spectral data of the internal structure, use an improved U-Net++ network structure to generate the characteristic data of internal cracks of the jade;

[0109] S3. Use an improved Monte Carlo tree search algorithm to generate an initial carving path based on the characteristic data of internal cracks of the jade, and optimize the initial carving path in combination with the material hardness distribution information;

[0110] S4. Convert the optimized carving path into specific carving processing parameters, including tool paths, feed speeds, and cutting depths;

[0111] S5. Control the carving equipment to perform the jade carving operation, and collect real-time carving data during the carving process;

[0112] S6. Use an adaptive vibration suppression algorithm to detect and suppress the tool vibration during the carving process in real time;

[0113] S7. Perform finite element simulation analysis based on the real-time carving data to generate an optimized carving process path and historical processing data;

[0114] S8. Use acoustic emission monitoring technology to detect abnormal conditions during the carving process in real time;

[0115] S9. Build a digital twin model of the jade carving process to update and visualize the carving state in real time;

[0116] S10. Use the reinforcement learning method to optimize the utilization rate of rough materials based on the digital twin model and generate the optimal carving path and tool parameters;

[0117] S11. Dynamically adjust the carving processing parameters based on the optimization result to achieve continuous optimization of the carving process.

[0118] The beneficial effects of the present invention are reflected at multiple levels. From a macroscopic perspective, the system realizes the intelligentization and automation of the entire jade processing process, greatly improving the processing efficiency and the quality of finished products, reducing material waste, and providing strong technical support for the transformation and upgrading of the jade processing industry. This not only helps to improve the economic benefits of enterprises but also promotes the sustainable development of the entire industry.

[0119] From the perspective of the system architecture, the present invention forms a closed-loop intelligent processing system through the organic combination of multiple functional modules. The collaborative work among the jade sample detection module, parameter transfer module, carving processing module, physical simulation module, and data center module realizes the overall optimization of the process from crack detection to carving execution. This modular design not only improves the reliability and maintainability of the system but also provides convenience for future function expansion and upgrading.

[0120] At the level of technical implementation, the present invention has innovations and breakthroughs in multiple key links. The combination of multi-spectral confocal scanning technology and the improved U-Net++ network significantly improves the accuracy and reliability of crack detection, especially for jade samples with complex internal structures. The improved Monte Carlo tree search algorithm takes into account the material hardness distribution and realizes more intelligent and efficient carving path planning. The introduction of the adaptive vibration suppression algorithm and acoustic emission monitoring technology greatly improves the stability and safety of the carving process.

[0121] In addition, the present invention also cleverly solves some technical contradictions existing in traditional methods. For example, the contradiction between high-precision detection and high-efficiency processing is well balanced through multi-spectral rapid scanning and parallel computing. The contradiction between carving precision and processing efficiency is solved through intelligent path planning and real-time adaptive control. These innovations not only improve the overall performance of the system but also enhance its adaptability and reliability in practical applications.

[0122] Finally, the digital twin and reinforcement learning technologies introduced in the present invention endow the system with the ability of self-learning and continuous optimization. This enables the system to continuously accumulate experience and optimize processing strategies, thus showing better and better performance during long-term use. This "the more it is used, the smarter it becomes" characteristic not only improves the practical value of the system but also provides a powerful example for future intelligent manufacturing.

[0123] In summary, the intelligent internal crack detection and carving path optimization system and method for jade proposed by the present invention comprehensively improve the intelligent level of jade processing through the organic combination of multiple innovative technologies. It not only solves many problems existing in the prior art but also achieves significant improvements in terms of efficiency, quality, safety, and adaptability, opening up a new path for the development of the jade processing industry. Brief Description of the Drawings

[0124] Figure 1 This is the top-level logic diagram of the entire system of the present invention.

[0125] Figure 2 This is the internal logic diagram of the jade sample detection module of the present invention.

[0126] Figure 3 This is the internal logic diagram of the parameter transfer module of the present invention.

[0127] Figure 4 This is the internal logic diagram of the carving and processing module of the present invention.

[0128] Figure 5 This is the internal logic diagram of the physical simulation module of the present invention

[0129] Figure 6 This is the internal logic diagram of the data center module of the present invention Detailed implementation manners

[0130] Please refer to the appendix Figure 1 - 6 , the present invention provides a jade internal crack intelligent detection and carving path optimization system and its method. The system aims to solve the problems of low accuracy in internal crack detection and low efficiency in carving path planning existing in traditional jade processing. Through advanced multi-spectral imaging technology, deep learning algorithms and intelligent optimization methods, accurate detection of internal cracks in jade and automatic optimization of carving paths are achieved.

[0131] The system of the present invention includes a jade sample detection module 1, a parameter transfer module 2, a carving and processing module 3, a physical simulation module 4 and a data center module 5. These modules form a closed-loop intelligent detection and processing system through close cooperation.

[0132] The jade sample detection module 1 is the entrance of the entire system. Its main function is to collect multi-angle high-resolution images of the jade sample to be detected and obtain the spectral data of the internal structure of the jade using a multi-spectral confocal scanning device. Preferably, this module uses a high-precision CCD camera and a multi-band light source, which can capture the minute details on the surface and inside of the jade. For example, in an embodiment of the present invention, an industrial camera with a resolution of 4096×3072 pixels is used, in cooperation with 10 narrow-band filters in the range of 400 - 1000 nm, to achieve all-round and high-precision imaging of the jade sample.

[0133] The multi-spectral confocal scanning device is an innovation of the present invention. By simultaneously obtaining information from multiple spectral channels, it can more comprehensively reflect the structural characteristics inside the jade. The scanning depth of this device can reach 5 mm, and the resolution is better than 10 μm, which enables the system to detect minute cracks that are difficult to be discovered by the naked eye.

[0134] Based on the collected high-resolution images and internal structure spectral data, the jade sample detection module 1 generates jade internal crack feature data. This process involves complex image processing and data fusion algorithms. The present invention adopts an improved U-Net++ network structure to achieve this goal.

[0135] The parameter transfer module 2 receives the jade internal crack feature data and generates carving planning information based on these data. The present invention adopts an improved Monte Carlo tree search (MCTS) algorithm to optimize the carving path. Preferably, the search depth of the MCTS algorithm is set to 10 layers, and 100 nodes are expanded in each layer, so that better path planning results can be obtained while ensuring computational efficiency.

[0136] The carving planning information is then converted into specific carving processing parameters, such as tool paths, feed speeds, and cutting depths. For example, for jade with higher hardness, the feed speed can be set to 0.1 - 0.5 mm / s, and the cutting depth is controlled between 0.05 - 0.2 mm to ensure the balance between carving quality and efficiency.

[0137] The carving processing module 3 receives these parameters and controls the carving equipment to perform the jade carving operation. During the carving process, this module is also responsible for collecting real-time carving data, such as tool positions, cutting forces, and material removal rates. These data are crucial for subsequent processing optimization.

[0138] The physical simulation module 4 receives the real-time carving data and performs finite element simulation analysis based on these data. The present invention adopts a high-precision finite element model with a mesh division accuracy of up to 0.01 mm, which can accurately simulate the stress distribution and deformation during the carving process. Through the simulation analysis, the system generates optimized carving process paths and detailed historical processing data.

[0139] The data center module 5 is the core of the entire system. It receives the carving process paths and historical processing data and trains a jade material dynamic adjustment model based on this information. This model adopts a deep reinforcement learning algorithm and can adaptively adjust carving parameters according to the hardness characteristics of different jades. Preferably, the reward function design of the reinforcement learning algorithm takes into account multiple factors such as carving quality, efficiency, and material utilization rate, enabling the system to find the best balance among multiple goals.

[0140] The carving process parameter models under different material hardnesses generated by the data center module 5 are sent back to the parameter transfer module 2 for further optimization of the carving planning information. This forms a closed-loop feedback mechanism, enabling the system to continuously learn and improve to adapt to the processing requirements of different types of jades.

[0141] The jade sample detection module 1 further includes a multi - spectral imaging unit 11, a confocal scanning unit 12, a data fusion unit 13, and a deep learning unit 14. The collaborative work of these units enables the system to comprehensively and accurately capture the internal structural features of the jade.

[0142] The multi - spectral imaging unit 11 adopts advanced spectral imaging technology and can simultaneously obtain image information in multiple bands. Preferably, the present invention uses 10 narrow - band filters with central wavelengths of 450nm, 500nm, 550nm, 600nm, 650nm, 700nm, 750nm, 800nm, 850nm, and 900nm respectively, and the bandwidth is 20nm. This configuration can fully reflect the characteristics of the jade under different spectra and helps to identify tiny internal cracks.

[0143] The confocal scanning unit 12 adopts a high - precision confocal microscopy system and can obtain three - dimensional structure information inside the jade. In one embodiment of the present invention, the axial resolution of the confocal scanning system reaches 0.5μm, the lateral resolution is 0.2μm, and the scanning depth can reach 5mm. This high - resolution three - dimensional scanning ability enables the system to accurately locate tiny cracks inside the jade.

[0144] The data fusion unit 13 is responsible for fusing the multi - spectral image data with the three - dimensional structure information to generate comprehensive feature data. The present invention adopts a multi - modal data fusion algorithm based on tensor decomposition, which can effectively integrate information from different sources and improve the richness and accuracy of feature expression.

[0145] The deep learning unit 14 is the core of the jade sample detection module 1. It uses an improved U - Net++ network structure to analyze the comprehensive feature data and outputs the detection results of internal cracks in the jade. The U - Net++ network is an advanced image segmentation network, and the present invention has improved it to meet the special requirements of internal crack detection in jade.

[0146] The improved U - Net++ network structure includes a shallow feature extraction module 141, a semantic feature fusion module 142, and a global feature extraction module 143. This network structure design fully considers the characteristics of multi - scale and complexity of internal cracks in jade and can effectively extract and fuse feature information at different levels.

[0147] The shallow feature extraction module 141 extracts low - level features of the jade image through multiple convolutional operations. Preferably, the present invention uses two 7×7 convolutional layers and four 3×3 convolutional layers, and batch normalization operations are followed after each convolutional layer to improve the stability and efficiency of feature extraction.

[0148] The semantic feature fusion module 142 is an innovation of the present invention. It realizes the effective fusion of features with different resolutions through five feature addition modules. This design can make full use of multi-scale information and improve the network's recognition ability for complex crack structures. For example, the first feature addition module is directly connected to the features from the input image, the output of the first 3×3 convolutional layer, and the first max-pooling layer, while the remaining four feature addition modules gradually fuse deeper features.

[0149] The global feature extraction module 143 extracts global context information through three serial 1×1 convolutional layers and a global pooling layer. This design helps the network understand the distribution of cracks in the entire jade sample, thereby improving the detection accuracy. Preferably, the number of channels of the 1×1 convolutional layers is set to 512, 256, and 128, and average pooling operation is used for global pooling.

[0150] The improved U-Net++ network of the present invention performs excellently in the task of detecting internal cracks in jade, achieving a detection accuracy of more than 95% on the test data set, far exceeding the performance of traditional image processing methods.

[0151] Through the collaborative work of the above-mentioned modules and units, the intelligent detection and carving path optimization system for internal cracks in jade of the present invention can efficiently and accurately complete the detection of internal cracks in jade and the optimization of carving paths. The modular design of the system makes it have good scalability and adaptability, and can be flexibly adjusted according to different jade types and processing requirements.

[0152] The system of the present invention significantly improves the precision and efficiency of jade processing by integrating advanced technologies such as multi-spectral imaging, deep learning, and intelligent optimization. Especially when dealing with high-value and complex-structured jade, the advantages of this system are more obvious. It can not only accurately identify tiny cracks that are difficult to be found by the naked eye, but also automatically plan the optimal carving path according to the crack distribution, maximizing the protection of the integrity and value of the jade.

[0153] In addition, the system of the present invention also has the ability of self-learning and continuous optimization. Through the reinforcement learning algorithm of the data center module 5, the system can continuously accumulate experience and optimize carving parameters and strategies to adapt to the processing requirements of different types of jade. This intelligent and adaptive characteristic makes this system have high application value in actual production, can significantly improve the quality and efficiency of jade processing, reduce material waste, and bring a revolutionary change to the jade processing industry.

[0154] The system of the present invention further includes more detailed functional modules to achieve more accurate and efficient detection of internal cracks in jade and optimization of carving paths.

[0155] The parameter transfer module 2 includes a carving planning sub-module 21, a parameter conversion sub-module 22, and a YOLOv8 object detection module 23. The collaborative work of these sub-modules ensures the precise calculation and dynamic adjustment of carving parameters.

[0156] The carving planning sub-module 21 is the core part of the parameter transfer module 2. It receives the internal crack feature data of the jade and generates an initial carving path based on the improved Monte Carlo tree search (MCTS) algorithm. The present invention improves the traditional MCTS algorithm to adapt to the special requirements of jade carving. Preferably, the improved MCTS algorithm adopts an adaptive expansion strategy and dynamically adjusts the search depth according to the importance of the nodes. For example, in the area with dense cracks, the search depth can be increased to 15 layers, while in the relatively intact area, it can be maintained at 8 - 10 layers, which can improve the accuracy of path planning while ensuring the calculation efficiency.

[0157] In addition, the carving planning sub-module 21 also optimizes the initial carving path by combining the material hardness distribution information. The system of the present invention adopts a fast hardness distribution estimation method based on Fourier transform, which can obtain the hardness distribution map of the jade sample in a short time. Preferably, the resolution of the hardness distribution map is set to 0.1mm×0.1mm, which can fully reflect the hardness changes inside the jade and provide an accurate reference for path optimization.

[0158] The parameter conversion sub-module 22 is responsible for converting the optimized carving path into specific carving processing parameters. These parameters include tool paths, feed rates, and cutting depths, etc. The system of the present invention adopts an adaptive parameter adjustment strategy and dynamically adjusts the processing parameters according to the hardness and crack distribution of the jade. For example, in the area with higher hardness, the feed rate can be reduced to 0.05mm / s, and the cutting depth is controlled between 0.02 - 0.05mm; while in the area with lower hardness, the feed rate can be increased to 0.5 - 1mm / s, and the cutting depth can be increased to 0.1 - 0.3mm. This adaptive adjustment strategy can significantly improve the carving efficiency while ensuring the processing quality.

[0159] The YOLOv8 object detection module 23 is an innovation point of the system of the present invention. It uses a pre-trained improved U-Net++ model as a feature extractor to perform real-time crack localization based on the jade image data. YOL Ov8 is one of the most advanced object detection algorithms at present. The present invention makes an adaptive improvement to it to improve the detection ability of tiny cracks inside the jade. Preferably, the backbone of the YOLOv8 network adopts the C SPDarknet53 structure, and the neck part uses the PANet structure. This configuration can improve the detection accuracy of small targets (such as tiny cracks) while ensuring the detection speed.

[0160] The output result of the YOLOv8 object detection module 23 is fed back to the engraving planning sub-module 21 in real time for dynamically adjusting the engraving path. This real-time feedback mechanism enables the system of the present invention to quickly respond to new situations that may occur during the engraving process, such as newly discovered cracks or changes in material properties, thereby ensuring the safety and efficiency of the engraving process.

[0161] The engraving processing module 3 includes a robotic arm control unit 31, a tool management unit 32, an adaptive vibration suppression unit 33, and a data acquisition unit 34. The collaborative work of these units ensures the precise control and real-time monitoring of the engraving process.

[0162] The robotic arm control unit 31 is responsible for controlling the movement of the engraving robotic arm according to the engraving processing parameters. The system of the present invention uses a high-precision six-axis industrial robot with a repeat positioning accuracy of up to ±0.02 mm, which ensures a high degree of accuracy of the engraving trajectory. Preferably, the control algorithm of the robotic arm adopts an adaptive PID control strategy, which can dynamically adjust the control parameters according to different engraving stages and material properties to obtain the best motion performance.

[0163] The tool management unit 32 is responsible for selecting appropriate engraving tools and controlling the replacement and adjustment of the tools. The system of the present invention is equipped with various types of engraving tools, including ball nose cutters, flat end mills, and pointed cutters, etc., which can be selected according to different engraving requirements. Preferably, the tool replacement process adopts an automated mechanism, and the replacement time is controlled within 5 seconds, greatly improving the engraving efficiency.

[0164] The adaptive vibration suppression unit 33 is another innovation point of the system of the present invention. It detects the tool vibration signal in real time, calculates the vibration frequency and intensity, and adaptively adjusts the tool rotation speed and movement direction when the vibration intensity exceeds a preset threshold. The present invention adopts a vibration signal analysis method based on wavelet transform, which can quickly and accurately identify the vibration components of different frequencies. Preferably, the sampling frequency of vibration detection is set to 10 kHz, so that vibration signals up to 5 kHz can be captured, covering most of the vibration frequencies that may occur during the engraving process.

[0165] The data acquisition unit 34 is responsible for collecting real-time processing data during the engraving process, including tool position, cutting force, and material removal rate, etc. The system of the present invention uses high-precision force sensors and position encoders, which can accurately record various parameters during the engraving process. Preferably, the data acquisition frequency is set to 100 Hz, so that sufficient detailed processing process information can be obtained without imposing too much burden on the system.

[0166] The specific working process of the adaptive vibration suppression unit 33 includes multiple steps, and each step is carefully designed to achieve the best vibration suppression effect.

[0167] First, the adaptive vibration suppression unit 33 smooths the vibration signal using a digital filter. In the present invention, a Butterworth low-pass filter is employed, and its cut-off frequency is set to 1 / 2 of the highest frequency of the original signal, which can effectively remove high-frequency noise and retain the useful vibration signal.

[0168] Next, a Fourier transform is performed on the smoothed vibration signal to obtain the vibration fundamental frequency. In the present invention, the fast Fourier transform (FFT) algorithm is adopted, which has high computational efficiency and can process a large amount of vibration data in real time. Preferably, the number of points of the FFT is set to 1024, which can control the calculation time at the millisecond level while ensuring the frequency resolution.

[0169] Based on the obtained vibration fundamental frequency, the adaptive vibration suppression unit 33 calculates the vibration frequency and vibration intensity. The calculation of the vibration intensity adopts the root mean square (RMS) method, which can more accurately reflect the energy level of the vibration.

[0170] When the vibration intensity exceeds the preset threshold V0, the system will perform a series of operations to suppress the vibration. First, the tool motion direction parameter is changed. In the present invention, a direction optimization strategy based on the genetic algorithm is adopted, which can find the optimal motion direction in a short time. Second, the tool rotation speed parameter is adjusted. The rotation speed adjustment adopts the fuzzy PID control algorithm, which can smoothly adjust the tool rotation speed according to the change trend of the vibration intensity.

[0171] In addition, the system of the present invention also smooths the threshold V0 using the sine fitting method. This processing can avoid system instability caused by threshold mutation and make the vibration suppression process smoother and more continuous. Preferably, the sine fitting adopts the least squares method, and the fitting period is set to the reciprocal of the vibration fundamental frequency, which can accurately capture the changes in the vibration characteristics.

[0172] Finally, based on the smoothed threshold, the system realizes the adaptive compensation of the vibration. The compensation strategy adopts a feedforward-feedback combined control method, which can ensure the accuracy of the engraving trajectory while suppressing the vibration.

[0173] The physical simulation module 4 includes a finite element analysis unit 41, a process path generation unit 42, and a historical data management unit 43. The collaborative work of these units realizes the precise simulation and optimization of the engraving process.

[0174] The finite element analysis unit 41 constructs a physical model of the jade carving process based on real-time carving data and performs stress analysis and deformation simulation. In the present invention, the adaptive mesh refinement technology is adopted, which can automatically refine the mesh in key areas (such as the crack tip) to improve the simulation accuracy. Preferably, the minimum mesh size can reach 0.01 mm, which can accurately capture the stress distribution around the microcracks.

[0175] The process path generation unit 42 generates an optimized engraving process path based on the simulation results and records the specific parameters of each engraving step. The present invention adopts a path optimization algorithm based on the stress field, which can minimize the risk of crack propagation while ensuring the engraving quality. Preferably, the path optimization takes into account multiple factors, including stress distribution, material removal rate, and surface finish, etc., and finds the optimal engraving path through a multi-objective optimization algorithm.

[0176] The historical data management unit 43 is responsible for storing and managing historical processing data and providing data query and analysis interfaces. The present invention adopts a distributed database system, which can efficiently process a large amount of historical data. Preferably, the data storage uses the time series database InfluxDB. This database is particularly suitable for processing time-related sensor data, has high query efficiency, and can support complex time series analysis.

[0177] Through the collaborative work of the above-mentioned modules and units, the intelligent crack detection and engraving path optimization system for jade in the present invention realizes the full-process intelligence from crack detection to engraving execution. Each link of the system is carefully designed and optimized, fully considering the special requirements of jade engraving, which can significantly improve the engraving quality and efficiency, reduce material waste, and bring a revolutionary change to the jade processing industry.

[0178] The system of the present invention also includes some advanced functional modules, which further improve the safety, intelligence, and traceability of the jade engraving process.

[0179] The system of the present invention also includes an acoustic emission monitoring module 6. This module is coupled to the engraving processing module 3 and plays an important monitoring role during the engraving process. Acoustic emission monitoring technology is a non-destructive testing method that can capture the acoustic wave signals generated by the internal structure changes of materials in real time, and has unique advantages for detecting the propagation of micro-cracks inside jade.

[0180] The acoustic emission monitoring module 6 collects the acoustic emission signals during the engraving process through highly sensitive acoustic emission sensors. Preferably, the present invention adopts piezoelectric ceramic sensors, whose sensitivity can reach -65dB (referring to 1V / μbar), and the frequency response range is 100kHz - 1MHz. This configuration can effectively capture various acoustic emission signals generated by the jade material during the engraving process, including key events such as crack propagation and material fracture.

[0181] In order to extract useful features from complex acoustic emission signals, the present invention adopts the Hilbert-Huang transform method. This is an advanced time-frequency analysis technique, which is particularly suitable for processing non-linear and non-stationary signals. The Hilbert-Huang transform first performs empirical mode decomposition (EMD) on the signal, decomposing the complex signal into a series of intrinsic mode functions (IMFs). Then, the Hilbert transform is performed on each IMF to obtain the instantaneous frequency and instantaneous amplitude. This method can effectively extract the time-frequency features of acoustic emission signals, providing a reliable basis for subsequent anomaly recognition.

[0182] The extracted feature vectors are then input into the SVM classifier to perform crack anomaly recognition. The present invention adopts a multi-class SVM classifier, which can simultaneously identify multiple anomaly types, such as crack propagation, material spalling, etc. Preferably, the SVM classifier adopts a radial basis function (RBF) kernel, and the kernel parameter γ and the penalty parameter C are optimized by the grid search method to obtain the best classification performance. In an embodiment of the present invention, the optimized SVM classifier achieves an anomaly recognition accuracy rate of over 95% on the test data set.

[0183] When an anomaly is detected, the acoustic emission monitoring module 6 will immediately send an alarm signal to the engraving processing module 3. This real-time monitoring and feedback mechanism greatly improves the safety of the engraving process, and can stop the processing in time before the crack expands to an uncontrollable extent, avoiding unnecessary losses.

[0184] The data center module 5 further includes a digital twin sub-module 51 and a reinforcement learning sub-module 52. The introduction of these two sub-modules enables the system of the present invention to have stronger intelligent and adaptive capabilities.

[0185] The digital twin sub-module 51 constructs a digital twin model of the jade carving process based on the carving process path and historical processing data. Digital twin technology is one of the core technologies of Industry 4.0. It realizes the real-time mapping and interaction between the physical world and the virtual world by creating a digital copy of the physical entity in the virtual space. In the present invention, the digital twin model not only includes the geometric information of the jade, but also includes multi-dimensional information such as material properties, processing parameters, and stress distribution.

[0186] Preferably, the present invention adopts a visualization interface based on the Unity3D engine to display the dynamics of the engraving process in real time. This intuitive visualization method enables the operator to better understand and control the engraving process. The update frequency of the digital twin model is set to 10Hz, which can ensure real-time performance without imposing too much computational burden on the system.

[0187] The reinforcement learning sub-module 52 closely cooperates with the digital twin sub-module 51 and uses the simulation environment provided by the digital twin model to train the reinforcement learning agent. The present invention adopts the Deep Deterministic Policy Gradient (DDPG) algorithm, which is a reinforcement learning algorithm suitable for continuous action spaces. The DDPG algorithm includes an Actor network and a Critic network. The Actor network is responsible for generating actions (such as adjusting carving parameters), and the Critic network is responsible for evaluating the value of the actions.

[0188] In the system of the present invention, the goal of reinforcement learning is to optimize the utilization rate of rough materials while considering carving quality and efficiency. The design of the reward function comprehensively considers multiple factors, including material removal rate, surface finish, energy consumption, etc. Preferably, the reward function adopts the form of a weighted sum, and the weights of each factor can be adjusted according to specific requirements. For example:

[0189] R = w1 * MRR + w2 * (1 - Ra) + w3 * (1 - E) + w4 * (1 - C),

[0190] where R is the total reward, MRR is the material removal rate, Ra is the surface roughness, E is the energy consumption, C is the risk of crack propagation, and w1, w2 are the weight coefficients.

[0191] Through continuous training and optimization, the reinforcement learning sub-module 52 can generate the optimal carving path and tool parameters. These optimization results will be fed back to the parameter transfer module 2 to guide the actual carving process. This optimization method based on digital twin and reinforcement learning enables the system of the present invention to continuously learn and improve, adapt to the processing requirements of different types of jade, and achieve true intelligence and self-adaptation.

[0192] The present invention also provides a method for intelligent detection of internal cracks in jade and optimization of carving paths corresponding to the above system. This method includes 11 main steps, covering the entire process from data collection to final optimization.

[0193] Step S1 is the data collection stage. The present invention adopts a method combining multi-angle high-resolution imaging and multi-spectral confocal scanning to comprehensively capture the surface and internal information of the jade sample. Preferably, the high-resolution imaging uses a 4K industrial camera with an imaging resolution of up to 3840×2160 pixels. The multi-spectral confocal scanning uses 10 narrow-band light sources with different wavelengths, covering the spectral range of 400 - 1000 nm. This multi-dimensional data collection method provides a rich information basis for subsequent crack detection.

[0194] Step S2 uses an improved U-Net++ network structure to generate jade internal crack feature data. As mentioned before, this network structure includes three key modules: shallow feature extraction, semantic feature fusion, and global feature extraction, which can effectively capture crack features at different scales. In an embodiment of the present invention, transfer learning strategy is adopted for the training of the U-Net++ network. First, it is pre-trained on a large-scale natural image dataset, and then fine-tuned on the jade crack dataset. This method greatly reduces the amount of labeled data required and improves the generalization ability of the model.

[0195] Steps S3 and S4 involve the generation and optimization of the carving path. The present invention uses an improved Monte Carlo tree search algorithm to generate the initial carving path, and then optimizes it in combination with the material hardness distribution information. Preferably, the material hardness distribution is obtained by nanoindentation testing method, and the spatial resolution can reach 1μm. The optimized carving path will be converted into specific carving processing parameters, including tool path, feed rate, and cutting depth, etc.

[0196] Steps S5 and S6 are the carving execution and vibration control stages. The present invention uses a high-precision six-axis industrial robot to execute the carving task and uses an adaptive vibration suppression algorithm to adjust the processing parameters in real time. The response time of the vibration suppression algorithm is controlled within 10ms, which can effectively suppress high-frequency vibration and improve the carving accuracy.

[0197] Step S7 involves finite element simulation analysis. The present invention uses adaptive mesh refinement technology to automatically refine the mesh in key areas (such as crack tips). Preferably, the minimum mesh size can reach 0.01mm, and the J-integral method is used for stress analysis, which can accurately predict the crack propagation trend.

[0198] Steps S8 and S9 introduce acoustic emission monitoring and digital twin technology. Acoustic emission monitoring can capture the minute changes inside the material in real time, providing a new dimension for anomaly detection. The digital twin model realizes the real-time mapping between the physical world and the virtual world, providing a reliable simulation environment for subsequent optimization.

[0199] Steps S10 and S11 are the optimization stages based on reinforcement learning. By continuously trial and error and learning in the digital twin environment, the system can find the optimal carving strategy to maximize the utilization rate of rough materials. Preferably, the reinforcement learning adopts an offline learning method based on experience replay, which can make full use of historical data to improve the learning efficiency.

[0200] Through these 11 carefully designed steps, the method of the present invention realizes the full process intelligence of jade internal crack detection and carving path optimization. The method makes full use of advanced artificial intelligence technologies, such as deep learning, reinforcement learning and digital twins, etc., which greatly improves the accuracy, efficiency and safety of jade processing. At the same time, the modular design of the method also makes it have good scalability and adaptability, and can be flexibly adjusted according to different jade types and processing requirements.

[0201] In general, the jade internal crack intelligent detection and carving path optimization system and method provided by the present invention realize the comprehensive intelligence and optimization of the jade processing process through the organic combination of multiple advanced technologies. The system can not only accurately detect the micro cracks inside the jade, but also automatically plan the optimal carving path according to the crack distribution and material properties, and perform real-time monitoring and adjustment during the carving process. This intelligent and adaptive processing method greatly improves the quality and efficiency of jade processing, reduces material waste, and provides new technical support for the upgrading and development of the jade processing industry.

[0202] In order to verify the superiority of the intelligent detection and carving path optimization system for jade internal cracks and the method thereof proposed in the present invention, a set of simulation experiments was designed. Three different jade samples were selected in the experiment, representing different internal structure complexity and crack distribution characteristics. The three samples are: Type A (simple structure, a small number of cracks), Type B (medium complexity, multiple small cracks), and Type C (high complexity, staggered distribution of large and small cracks).

[0203] The test data of this embodiment are strictly set based on a standardized experimental environment: In the experiment, three types of typical jade samples, Hetian jade (Mohs hardness 6.0 - 6.5), jadeite (6.5 - 7.0), and Dushan jade (6.0 - 7.5), are selected, and their hardness ranges cover the material spectrum of common processing materials in the industry. The engraving equipment uses a KUKA KR 500 six-axis industrial robot, equipped with an NSK spindle (maximum rotational speed 40,000 rpm) and a Φ0.1 - 2 mm diamond-coated tool system. The tool management system has a built-in parameter library of 20 ISO standard tool types. The design of the reward function of the reinforcement learning sub-module constructs theoretical rationality through a three-stage verification process - first, based on the production data analysis in the jade processing field, a quantitative correlation model (Pearson correlation coefficient > 0.82) between engraving quality, efficiency, material utilization rate, and tool life is established; second, the sensitivity analysis of the weight coefficients is carried out using the orthogonal experiment method to determine the optimization intervals of the quality factor of 0.4 and the efficiency factor of 0.3; finally, through the Pareto front optimization algorithm, the weight combination with the optimal comprehensive benefit (Q = 0.4, E = 0.3, U = 0.2, S = 0.1) is selected from 300 groups of simulation experiments. In the three-month production verification in a jade carving factory in Yunnan, this parameter set reduced the standard deviation of the yield rate fluctuation from 12.6% to 3.2%, proving its engineering practicability. The material utilization rate is tested by using a Sartorius CPA225D precision balance (resolution 0.1 mg) for three repeated weighings and taking the average value, and the material removal volume is verified by combining high-precision three-dimensional scanning and reverse modeling to ensure the integrity of the data traceability chain. The stress field simulation of the engraving path optimization effect is based on the ANSYS Workbench platform, using the adaptive mesh refinement technology (minimum element size 0.01 mm) and the J-C plastic constitutive model. The matching degree between its simulation results and the dynamic stress fluctuation characteristics of the acoustic emission monitoring data reaches 89%, forming a dual verification system of physical experiments and digital twins.

[0204] Example 1 uses the complete system of the present invention for jade processing. Comparative Example 1 uses the traditional manual inspection and fixed-path engraving method. Comparative Example 2 uses a simple machine vision inspection and preset path planning method. We comprehensively tested these three methods on three different types of jade samples, mainly focusing on the following key indicators: crack detection accuracy, engraving path optimization effect, material utilization rate, processing efficiency, and finished product quality.

[0205] The crack detection accuracy is obtained by comparing the manually annotated ground truth with the system detection results. The carving path optimization effect is evaluated by simulating the deviation between the final carving path and the ideal path. The material utilization rate is calculated as: final product weight / raw material weight*100%. Processing efficiency is measured by the time required to complete carving of the same complexity. The quality of the finished product is comprehensively scored by a professional evaluation team based on factors such as surface finish and detail restoration, with a full score of 100.

[0206] The following are the specific detection methods for each indicator:

[0207] 1. Crack detection accuracy: The cross-validation method is used to test the manually annotated data set and calculate the precision, recall and F1 score of the detection results.

[0208] 2. Engraving path optimization effect: Use finite element analysis software to simulate the stress distribution of different paths and calculate the root mean square error compared with the ideal stress-free state.

[0209] 3. Material utilization rate: Use high-precision electronic scales to measure the weight of raw materials and finished products, accurate to 0.01 grams.

[0210] 4. Processing efficiency: record the entire process time from the start of detection to the completion of engraving, accurate to seconds.

[0211] 5. Finished product quality: Invite 5 experts in the field of jade processing to conduct blind evaluation, and take the average value after removing the highest and lowest scores.

[0212] The test results are shown in the following table:

[0213] Index Sample type Example 1 Comparative Example 1 Comparative Example 2 Crack detection accuracy rate Type A 98.5％ 85％ 90.2％ Type B 96.8％ 72.3％ 83.7％ Type C 95.3％ 60.5％ 75.4％ Engraving path optimization effect (error value) Type A 0.15 0.52 0.38 Type B 0.22 0.78 0.55 Type C 0.31 1.25 0.86 Material utilization rate Type A 78.5％ 65.2％ 70.8％ Type B 72.3％ 58.7％ 63.5％ Type C 68.7％ 50.3％ 57.2％ Processing efficiency (hours) Type A 3.5 5.2 4.8 Type B 5.8 9.5 8.2 Type C 8.2 15.3 12.7 Finished product quality (score) Type A 92 78 83 Type B 88 70 76 Type C 85 62 69

[0214] It can be seen from the test results that the system of the present invention is significantly superior to the traditional method and the simple machine vision method in all test indicators. In particular, when processing the more complex type B and type C samples, the advantages of the present invention are more obvious.

[0215] In terms of crack detection accuracy, the system of the present invention maintains a high accuracy of more than 95% on all types of samples, while the accuracy of the traditional method drops sharply on complex samples. This fully demonstrates the advancement of the multi-spectral confocal scanning and improved U-Net++ network used in the present invention.

[0216] The test results of the engraving path optimization effect show that the system of the present invention can generate an engraving path closer to the ideal state, especially when processing complex samples, the advantage is more prominent. This shows that the improved Monte Carlo tree search algorithm and the optimization strategy considering the material hardness distribution adopted by the present invention are indeed effective.

[0217] In terms of material utilization rate, the system of the present invention has achieved a significant improvement in all types of samples, with an average increase of approximately 15%. This not only means higher economic benefits but also reflects the contribution of this system to resource conservation and environmental protection.

[0218] The improvement in processing efficiency is also a major highlight of the present invention. Compared with traditional methods, this system has shortened the processing time by approximately 40% on average. This improvement in efficiency can not only increase production capacity but also significantly reduce energy consumption.

[0219] The scoring results of the finished product quality further confirm the superiority of the present invention. Even on the most complex Type C samples, this system can still maintain a high quality of over 85 points, while the quality scores of traditional methods have decreased significantly.

[0220] It is worth noting that when the system of the present invention processes samples with different complexities, the decline in various indicators is much smaller than that of the other two methods. This shows that this system has good adaptability and stability and can handle various complex jade processing scenarios.

[0221] In summary, the intelligent crack detection and carving path optimization system and method for jade proposed in the present invention have shown significant advantages in multiple aspects such as crack detection accuracy, carving path optimization, material utilization rate, processing efficiency, and finished product quality. These advantages stem from the innovations of the present invention at multiple technical points, such as multi-spectral confocal scanning technology, improved U-Net++ network, path optimization algorithm based on MCTS, adaptive vibration suppression, acoustic emission monitoring, etc. These innovations not only improve the precision and efficiency of jade processing but also greatly enhance the system's ability to process complex jade samples, providing strong technical support for the intelligent and high-quality development of the jade processing industry.

[0222] It should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A jade internal crack intelligent detection and carving path optimization system, characterized in that: include: Jade sample detection module, used for: Collect multi-angle and high-resolution images of the jade samples to be inspected; The spectral data of the internal structure of jade is obtained by using a multi-spectral confocal scanning device; Generate jade internal crack feature data based on the high-resolution image and internal structure spectrum data; The parameter transfer module is coupled to the jade sample detection module and is used to: Receiving the internal crack characteristic data of the jade; Generate carving planning information based on the internal crack feature data of the jade; Converting the engraving planning information into engraving processing parameters; The engraving processing module is coupled to the parameter transfer module and is used to: Receiving the engraving processing parameters; Based on the engraving processing parameters, controlling the engraving equipment to perform jade engraving operations; During the engraving process, collect real-time engraving data; The physical simulation module is coupled to the engraving processing module and is used to: Receiving the real-time engraving data; Based on the real-time engraving data, perform finite element simulation analysis; Generate engraving process path and historical processing data; A data center module, coupled to the physical simulation module and the parameter transfer module, is used to: Receiving the engraving process path and historical processing data; Based on the carving process path and historical processing data, a dynamic adjustment model of jade material is trained; Generate engraving process parameter models under different material hardness; The engraving process parameter model is sent to the parameter transfer module for optimizing the engraving planning information.

2. The system according to claim 1, characterized in that The jade sample detection module comprises: A multispectral imaging unit, used to collect multispectral image data of the jade sample to be tested; Confocal scanning unit, used to obtain the internal three-dimensional structure information of jade; a data fusion unit, coupled to the multispectral imaging unit and the confocal scanning unit, for fusing the multispectral image data with the three-dimensional structure information to generate comprehensive feature data; A deep learning unit, coupled to the data fusion unit, is used to: Receiving the comprehensive feature data; Analyzing the comprehensive feature data using an improved U-Net++ network structure; Output the jade internal crack detection results.

3. The system according to claim 2, characterized in that The improved U-Net++ network structure includes: Shallow feature extraction module, used to extract low-level features of jade images through multi-layer convolution operations; The semantic feature fusion module is coupled to the shallow feature extraction module and is used to: receiving the low-level features; Through five feature addition modules, feature information of different resolutions is fused; Output fused multi-scale features; A global feature extraction module is coupled to the semantic feature fusion module and is used to: Receiving the fused multi-scale features; Extract global context information through serial 1×1 convolutional layers and global pooling layers; Output the final crack detection feature map.

4. The system according to claim 1, characterized in that The parameter transfer module includes: The carving planning submodule is used to: Receiving the internal crack characteristic data of the jade; Generate the initial carving path based on the improved Monte Carlo tree search algorithm; Optimizing the initial engraving path based on material hardness distribution information; The parameter conversion submodule is coupled to the carving planning submodule and is used to: Receiving the optimized engraving path; Converting the engraving path into specific engraving processing parameters, including tool trajectory, feed speed and cutting depth; The YOLOv8 target detection module is coupled to the carving planning submodule and is used to: Use the pre-trained improved U-Net++ model as the feature extractor; Perform real-time crack location based on jade image data; The crack location result is fed back to the engraving planning submodule for dynamically adjusting the engraving path.

5. The system according to claim 1, characterized in that The engraving processing module comprises: A robot control unit, used to control the movement of the engraving robot according to engraving processing parameters; A tool management unit is coupled to the robot control unit and is used to: Choose the right engraving tool; Control the replacement and adjustment of cutting tools; An adaptive vibration suppression unit is coupled to the robot control unit and the tool management unit, and is used to: Real-time detection of tool vibration signals; Based on the detected vibration signal, calculating the vibration frequency and intensity; When the vibration intensity exceeds the preset threshold, the tool speed and movement direction are adaptively adjusted; Data acquisition unit, used to collect real-time processing data during the engraving process, including tool position, cutting force and material removal rate.

6. The system according to claim 5, characterized in that The adaptive vibration suppression unit is specifically used for: Use digital filters to smooth vibration signals; Perform Fourier transform on the smoothed vibration signal to obtain the vibration fundamental frequency; Based on the fundamental frequency of vibration, calculate the vibration frequency and vibration intensity; When the vibration intensity exceeds the preset threshold V0, the following operations are performed: Change the tool movement direction parameters; Adjust tool speed parameters; The threshold V0 is smoothed using the sine fitting method; Based on the smoothed threshold, adaptive compensation of vibration is achieved.

7. The system according to claim 1, characterized in that The physical simulation module includes: a finite element analysis unit, which is used to: Construct a physical model of the jade carving process based on real-time carving data; Perform stress analysis and deformation simulation; A process path generation unit, coupled to the finite element analysis unit, is used to: Generate optimized engraving process path based on simulation results; Record the specific parameters of each engraving step; A historical data management unit is coupled to the process path generation unit and is used to: Store and manage historical processing data; Provides data query and analysis interface.

8. The system according to claim 1, characterized in that Also includes: The acoustic emission monitoring module is coupled to the engraving processing module and is used to: Acoustic emission signals during the engraving process are collected by using an acoustic emission sensor; The Hilbert-Huang transform method is used to extract the feature vector of the acoustic emission signal; Inputting the feature vector into a SVM classifier to perform crack anomaly recognition; When an abnormality is detected, an alarm signal is sent to the engraving processing module.

9. The system according to claim 1, characterized in that The data center module also includes a digital twin module, which is used to: Build a digital twin model of the jade carving process based on the carving process path and historical processing data; update the digital twin model in real time to reflect the status of the physical carving process; Provide a visual interface to display the real-time dynamics of the engraving process; A reinforcement learning submodule, coupled to the digital twin submodule, is used to: Use the simulation environment provided by the digital twin model to train reinforcement learning agents; Optimize the utilization rate of raw materials and generate the best engraving path and tool parameters; The optimization result is fed back to the parameter transfer module.

10. A method for intelligent detection of internal cracks in jade and optimization of engraving paths, based on the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1. Collect multi-angle high-resolution images of the jade sample to be tested, and obtain spectral data of the internal structure of the jade using a multi-spectral confocal scanning device; S2. Based on the high-resolution image and internal structure spectral data, the improved U-Net++ network structure is used to generate the internal crack feature data of the jade; S3. Using an improved Monte Carlo tree search algorithm, an initial engraving path is generated based on the internal crack feature data of the jade, and the initial engraving path is optimized in combination with the material hardness distribution information; S4. converting the optimized engraving path into specific engraving processing parameters, including tool trajectory, feed speed and cutting depth; S5. Control the engraving equipment to perform jade engraving operations and collect real-time engraving data during the engraving process; S6. Use adaptive vibration suppression algorithm to detect and suppress tool vibration during engraving in real time; S7. Perform finite element simulation analysis based on real-time engraving data to generate optimized engraving process path and historical processing data; S8. Use acoustic emission monitoring technology to detect abnormal conditions during the engraving process in real time; S9. Build a digital twin model of the jade carving process to update and visualize the carving status in real time; S10. Use reinforcement learning methods to optimize the utilization of raw materials based on the digital twin model and generate the optimal engraving path and tool parameters; S11. Based on the optimization results, the engraving processing parameters are dynamically adjusted to achieve continuous optimization of the engraving process.

Images