An automatic detection and cleaning system for oriented crystal mold shell

Abstract

The application discloses a kind of directional crystal mould shell automatic detection cleaning system.In the application, by transient air pressure excitation and multi-scale wavelet domain digital speckle correlation detection, the micro crack identification ability is identified from surface opening defect to sub-surface strain anomaly stage, which significantly improves the detection sensitivity of hidden cracks and wall thickness thinning defects, and replaces artificial visual experience determination with full-field strain cloud quantification grading, which significantly reduces the missed detection rate and the quality risk caused by human error. Strain-cleanness dual-factor fusion evaluation simultaneously completes the judgment of structural integrity and internal cavity cleanliness at the same station, improving the completeness and automation of defect evaluation, reducing the drag of batch offline sampling and chemical permeation operation on production rhythm, and eliminating the need for fluorescent penetrant immersion and cleaning process with transient pulse excitation method, which improves the single-piece detection rhythm and reduces the cost of hazardous chemicals procurement, storage and waste liquid treatment.

Description

Technical Field

[0001] This invention belongs to the field of crystal mold shell detection technology, specifically an automatic detection and cleaning system for oriented crystal mold shells. Background Technology

[0002] Directional crystal mold shells are specialized mold shells used in directional solidification and precision casting of single-crystal high-temperature alloys. They are the core process carriers that ensure the microstructure of the casting grows in a specific direction. Their typical structure consists of a face layer and a back layer. The face layer uses high-purity corundum and other refractory materials to resist the erosion and chemical corrosion of the high-temperature alloy liquid, ensuring a smooth blade surface. The back layer uses mullite or alumina-based materials, giving the mold shell excellent overall strength and resistance to deformation. In the manufacturing process of precision casting directional crystal mold shells, the mold shell after core removal needs to be cleaned to remove residual mold shell and loose sand, and then subjected to non-destructive testing to screen for defects such as microcracks and abnormal wall thickness. Currently, existing processes generally use manual or semi-automatic spray cleaning combined with fluorescent penetrant testing. The cleaning process largely relies on operators to move the mold shells one by one into the spray tank or ultrasonic tank. Loading, unloading, and clamping between processes are done manually. After cleaning, offline hot air drying is used, and then defects are determined by manually observing the traces of penetrant under ultraviolet light.

[0003] However, existing technologies generally rely on fluorescence penetration supplemented by manual visual inspection for defect judgment, which can only detect surface cracks and has almost no ability to detect subsurface and wall thickness anomalies. The false negative rate is high and it is difficult to give quantitative classification. The inspection of tissue structure integrity and cleanliness is usually carried out offline in steps, resulting in a long cycle and the consumption of a large amount of chemical reagents, thereby increasing the cost of hazardous waste treatment. Summary of the Invention

[0004] The purpose of this invention is to provide an automatic detection and cleaning system for oriented crystal mold shells in order to solve the problems mentioned above.

[0005] The technical solution adopted in this invention is as follows: an automatic detection and cleaning system for directional crystal mold shells, comprising an automatic feeding and positioning module, a spraying and ultrasonic cleaning module, a rinsing and hot air drying module, a transient air pressure-DIC detection module, and a sorting and unloading module;

[0006] The mold shell transfer output end of the automatic feeding and positioning module is connected to the rotating tray input end of the spray and ultrasonic cleaning module, and its positioning completion signal output end is connected to the process start control end of the spray and ultrasonic cleaning module.

[0007] The cleaning process of the spray and ultrasonic cleaning module is completed. The output end of the mold shell is connected to the inlet end of the rinsing tank of the rinsing and hot air drying module.

[0008] The drying completion output of the rinsing and hot air drying module is connected to the detection station input of the transient air pressure-DIC detection module, and its drying compliance signal output is connected to the detection trigger enable of the transient air pressure-DIC detection module.

[0009] The transient air pressure-DIC detection module's mold shell discharge end is connected to the workpiece input end of the sorting and unloading module, its judgment signal output end is connected to the sorting action control input end of the sorting and unloading module, and its network interface is connected to the host computer database. The sorting completion status feedback signal end of the sorting and unloading module is connected to the host computer database.

[0010] The transient pressure-DIC detection module includes: a transient pulse excitation submodule, a digital speckle imaging and strain calculation submodule, and a defect intelligent assessment and cleanliness detection submodule.

[0011] In a preferred embodiment, the automatic feeding and positioning module internally includes a chain conveyor, a pneumatic stop mechanism, a set of reflective photoelectric sensors, a dual-axis linear module, and a flexible end gripper. The chain conveyor uses stainless steel chain plates with a pitch of 38.1mm assembled to form a 400mm wide conveying surface. The end of the inclined material channel connects to the beginning of the conveyor, and the mold shell inside the material frame slides into the conveyor line by gravity. Guide rails are installed on both sides of the conveying channel, and the spacing is adjustable.

[0012] In a preferred embodiment, the photoelectric sensor is an Omron E3Z-R61. The transmitter and receiver are positioned on opposite sides of the conveyor line, with the installation height flush with the waist of the mold shell. Upon detecting workpiece shading, it outputs a switching signal, triggering the stop mechanism. The stop mechanism consists of a double-acting cylinder with a 20mm diameter and 30mm stroke, and a baffle with a polyurethane-coated front end. When the solenoid valve coil is energized, the cylinder piston rod extends, stopping the mold shell at the gripping reference point. The dual-axis linear module has an effective horizontal axis stroke of 600mm and an effective vertical axis stroke of 200mm, both driven by 86 stepper motors via ball screws, with a positioning repeatability of ±0.02mm. The end of the vertical axis is connected to a flexible gripper via an aluminum flange. The gripper fingertips are covered with silicone pads, and the three fingers are synchronously closed using air pressure. The gripping force is adjustable to 15N. After gripping, the linear module moves the mold shell to the rotating tray above the cleaning module station for release.

[0013] In a preferred embodiment, the spray and ultrasonic cleaning module internally includes a sealed spray chamber, a cleaning fluid spray pipeline, an immersion ultrasonic cleaning tank, an ultrasonic generating system, and a tray lifting and transferring device. The spray chamber is welded from 304 stainless steel plate, with a row of fan-shaped nozzles at the top and bottom. The nozzles are Lechler 652.404, with a spray angle of 40°. The upper and lower nozzles are arranged opposite each other, with the spacing determined by the height of the mold shell. The cleaning fluid is supplied by a vertical multistage centrifugal pump with a rated pressure of 0.5 MPa. A 10 μm bag filter and an electric heating element are connected in series in the pipeline to maintain the cleaning fluid temperature at 45 ± 2°C. The cleaning fluid is a 3% (v / v) weakly alkaline water-based cleaning agent. The spray timing is controlled by a PLC-controlled solenoid valve, set to 120 seconds. The ultrasonic cleaning tank is located behind the spray chamber, with an effective volume of 80L. Six sets of vibrating plates are installed on the inner wall, each set with 12 piezoelectric ceramic transducers. The operating frequency is 28kHz, and the total ultrasonic power is nominally 1200W, driven by an ultrasonic generator with automatic frequency tracking. The tray lifting and transfer device consists of a vertical guide column, a cylinder-driven lifting platform, and a rotatable tray. The cylinder stroke is 150mm, which can lower the tray carrying the mold shell from the spray chamber to 50mm below the ultrasonic tank liquid level. After cleaning, it is raised back to the transfer height. The tray is supported by sealed bearings and driven by a micro gear motor to rotate at a speed of 10r / min.

[0014] In a preferred embodiment, the rinsing and hot air drying module internally includes a clean water rinsing tank, a pure water spray pipeline, a six-axis transfer robotic arm, a hot air generator and blowing device, and a dew point sensor. The clean water rinsing tank has a capacity of 120L and a row of spray nozzles at the bottom. A circulating pump delivers deionized water, filtered through a 0.22μm precision filter, into the tank to create turbulence. An overflow outlet maintains a constant liquid level. A conductivity probe monitors the conductivity of the rinsing water online, automatically initiating water replenishment and wastewater discharge when it exceeds 5μS / cm. The pure water spray pipeline is located above the outlet of the rinsing tank and includes a set of needle-shaped nozzles for final rinsing of the mold shell at a pressure of 0.3MPa. The transfer robotic arm is a six-axis articulated type with a two-finger gripper at the wrist, achieving a repeatability of ±0.05mm. It is used to grasp the cleaned mold shell and transfer it from the ultrasonic tank to the rinsing tank, and then to the drying station. The hot air generation and blowing device consists of a volute-less centrifugal fan, a finned electric heater, a high-efficiency air filter, and hollow multi-directional air knives. The fan draws in ambient air, which is heated to 80°C by the heater and then passes through an H13-grade HEPA filter before being distributed to multiple rectangular flat air knives via air ducts. The air velocity at the air knife outlet is approximately 25 m / s, blowing simultaneously into the inner and outer cavities of the mold shell from four directions: top, bottom, left, and right. A SensirionSHT35 dew point sensor is installed at the exhaust vent of the drying station. The probe measures the dew point temperature of the exhaust air to determine the drying endpoint.

[0015] In a preferred embodiment, the transient pulse excitation submodule includes: an oil-free air compressor, a compressed air conditioning unit, a high-precision electro-proportional pressure regulating valve, a high-speed response two-position two-way solenoid valve, a dedicated internal cavity sealing nozzle assembly, and a timing synchronization control board. The oil-free air compressor has a rated discharge pressure of 0.8 MPa and a displacement of 40 L / min, with its outlet connected to the compressed air conditioning unit via a copper pipe. The compressed air conditioning unit consists of a pre-filter, a precision oil removal filter, an activated carbon adsorption cartridge, and a refrigerated dryer connected in series, outputting clean compressed air with a dew point of -20°C and an oil content of less than 0.003 mg / m³. The electro-proportional pressure regulating valve is model SMCITV2050, with an input pressure range of 0 to 1 MPa and an output pressure continuously adjustable within the range of 0.005 to 0.5 MPa, a linearity of ±0.5% of full scale, and accepts 0 to 10V analog voltage or RS-485 communication commands to set the target pressure. The high-speed solenoid valve is a direct-acting type with a 2mm diameter and a response time of less than 5ms. The valve body is made of 316 stainless steel, and the coil power is 8W. The outlet is connected to the sealing nozzle via a 300mm long, 4mm inner diameter nylon tube. The sealing nozzle assembly consists of an aluminum alloy base, a conical sealing head, and an O-ring fluororubber seal. The base has a flow channel machined into its inner cavity, and the sealing head has a 60° cone angle. During installation, the nozzle is pressed against the mold gate end face to form an airtight contact, and the pulsed airflow is injected into the mold cavity through the flow channel.

[0016] In a preferred embodiment, the digital speckle imaging and strain calculation submodule uses a micro-sprayer to spray random black paint speckles with an average particle size of 80 μm and a coverage of 50%. Two high-speed cameras with a pixel size of 5.5 μm and a resolution of 2048 × 2048 px simultaneously acquire 5 frames of images before and after pulse excitation under 460 nm ring blue light illumination at a frame rate of 10000 fps and an exposure time of 10 μs. The timing synchronization controller controls the shutter jitter within 50 ns. After epipolar correction, the images are sent to the embedded processing unit. This unit runs a multi-scale wavelet domain digital speckle correlation algorithm, first performing L=3 level two-dimensional discrete wavelet transform on the reference image f and the deformed image g to obtain the wavelet coefficients at each level. and Then, a coarse-to-fine search strategy is adopted, based on scale weights. After calculating the weighted normalized correlation function S and determining the most relevant integer pixel positions, the full-field displacement vector is obtained through sub-pixel interpolation. Finally, the Green-Lagrange strain components are calculated from the displacement gradient field, and the measured strain contour map is generated. The full-field solution takes less than 50ms.

[0017] The multi-scale wavelet domain weighted normalized correlation function is intuitively represented as:

[0018]

[0019] In the formula:

[0020] S is the weighted correlation coefficient, with a value ranging from [-1; 1];

[0021] l is the wavelet decomposition layer number, ranging from 1 to the total number of layers L. In this example, L=3. It is the scale weight of the l-th layer, and a higher weight is given to the coarse scale to suppress noise;

[0022] k iterates through all row and column indices of the l-th layer wavelet coefficient matrix;

[0023] and The two-dimensional discrete wavelet coefficients of the reference image f and the deformed image g at the l-th and k-th coefficient positions, respectively, are obtained by Daubechiesdb4 wavelet basis decomposition. This formula, through multi-scale fusion, suppresses speckle noise while preserving high-frequency edge details, thereby improving the detection sensitivity of small strain fields by orders of magnitude.

[0024] In a preferred embodiment, the defect intelligent assessment and cleanliness detection submodule is equipped with a deep learning inference engine. First, the measured strain cloud map (2048×2048px) is fed into a deep separable convolutional feature extraction network, outputting a 512-dimensional feature vector. The database stores a set of strain contour fingerprints from 50 qualified mold shells, and the baseline feature mean vector μ and the 512×512 covariance matrix have been pre-calculated. During each test, the engine calculates the squared Mahalanobis distance. As an indicator of strain anomaly scale. Simultaneously, a laser particulate sensor detects the gas inside the cavity at a sampling flow rate of 28.3 L / min, outputting the mass concentration (CC) of particles larger than 0.3 μm, in units of... The cleanliness limit is preset to .

[0025] The defect intelligent assessment and cleanliness detection submodule proposes a strain-cleanliness dual-factor defect severity index H, which directly multiplies the saturated Marvin contribution by the particulate matter exceedance ratio, outputting a continuous value from 0 to 1, and giving a qualified / unqualified judgment according to a threshold of 0.5. At the same time, the abnormal area is back-projected onto the cloud map to generate a defect map, and finally the judgment signal and graphic data are sent to the sorting system and the host computer through the I / O interface.

[0026] It is equipped with a strain-cleanliness dual-factor defect severity index, which is visualized as follows:

[0027]

[0028] In the formula:

[0029] H is the defect severity index, which is dimensionless.

[0030] For Mahalanobis distance, by definition

[0031] C represents the measured mass concentration of particulate matter in the cavity, in units of...

[0032] For cleanliness limits, take... ;

[0033] The formula is passed By compressing the unbounded Mahalanobis distance to the [0,1] interval and then directly multiplying it with the particulate matter exceedance ratio, equal-weighted fusion of the two abnormal signals of strain and cleanliness is achieved, thereby improving the defect detection rate.

[0034] In a preferred embodiment, the sorting and unloading module internally includes: a four-axis SCARA robot, two independent gravity slides, a qualified product palletizing and positioning platform, and a waste receiving box. The SCARA robot has a 600mm arm span and is equipped with a pneumatic gripper at its end. The gripper fingers are bonded with wear-resistant rubber, and the gripping force is adjustable. The robot controller receives the judgment signal output by the detection module through an Ethernet interface and places the mold shell into the inlet bracket of the corresponding slide according to the qualified or unqualified command.

[0035] In a preferred embodiment, the qualified product chute is made of stainless steel plate bent into a U-shaped channel with a cross-sectional width of 320mm and an inclination angle of 15°. The end is connected to a lifting palletizing and positioning platform, which consists of a ball screw lifting mechanism and cross roller guides. The platform automatically descends one workpiece height as the workpieces are stacked layer by layer. The waste receiving box is a steel mesh bin with a buffer baffle at its inlet to guide the unqualified mold shells to slide down. After sorting, the robot sends a sorting completion flag to the EtherCAT bus for the host computer to read the status.

[0036] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0037] 1. In this invention, by using transient gas pressure excitation and multi-scale wavelet domain digital speckle correlation detection, the ability to identify microcracks is advanced from surface-opening defects to the subsurface strain anomaly stage, significantly improving the detection sensitivity for defects such as hidden cracks and wall thinning. Simultaneously, full-field strain cloud map quantitative grading replaces manual visual experience judgment, greatly reducing the missed detection rate and quality risks caused by human error. The strain-cleanliness dual-factor fusion assessment simultaneously completes the assessment of structural integrity and internal cavity cleanliness at the same workstation, improving the completeness and automation of defect evaluation and reducing the delays in production cycle caused by batch offline sampling and chemical permeation operations.

[0038] 2. In this invention, the transient pulse excitation method eliminates the need for fluorescent penetrant soaking and cleaning processes, increasing the single-piece inspection cycle and reducing the costs of hazardous chemical procurement, storage, and waste liquid treatment. The entire line, from self-positioning gripping, spray ultrasonic cleaning, and dew point closed-loop drying to automatic robot sorting driven directly by judgment signals, forms a continuous flow, improving the consistency of mold shell processing and data closed-loop traceability, while reducing the risk of bumps and scratches introduced by manual handling and repeated clamping and positioning. Attached Figure Description

[0039] Figure 1 This is an overall system block diagram of the present invention;

[0040] Figure 2 This is a system block diagram of the transient air pressure-DIC detection module in this invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] Example:

[0043] Reference Figure 1-2 An automatic detection and cleaning system for directional crystal mold shells includes an automatic feeding and positioning module, a spraying and ultrasonic cleaning module, a rinsing and hot air drying module, a transient air pressure-DIC detection module, and a sorting and unloading module.

[0044] The mold shell transfer output of the automatic feeding and positioning module is connected to the rotating tray input of the spray and ultrasonic cleaning module, and its positioning completion signal output is connected to the process start control of the spray and ultrasonic cleaning module. The cleaning completion mold shell output of the spray and ultrasonic cleaning module is connected to the rinsing tank inlet of the rinsing and hot air drying module. The drying completion output of the rinsing and hot air drying module is connected to the detection station input of the transient air pressure-DIC detection module, and its drying compliance signal output is connected to the detection trigger enable of the transient air pressure-DIC detection module. The mold shell discharge end of the transient air pressure-DIC detection module is connected to the workpiece input of the sorting and unloading module, and its judgment signal output is connected to the sorting action control input of the sorting and unloading module, and its network interface is connected to the host computer database. The sorting completion status feedback signal of the sorting and unloading module is connected to the host computer database.

[0045] The transient pressure-DIC detection module includes: a transient pulse excitation submodule, a digital speckle imaging and strain calculation submodule, and a defect intelligent assessment and cleanliness detection submodule.

[0046] The automatic feeding and positioning module internally includes a chain conveyor, a pneumatic stop mechanism, a set of reflective photoelectric sensors, a dual-axis linear module, and a flexible end gripper. The chain conveyor uses stainless steel chain plates with a 38.1mm pitch to form a 400mm wide conveying surface. The inclined material channel ends at the beginning of the conveyor, and the mold shell inside the material frame slides into the conveyor line by gravity. Guide rails with adjustable spacing are installed on both sides of the conveyor channel.

[0047] The photoelectric sensor, model Omron E3Z-R61, has its transmitter and receiver positioned on opposite sides of the conveyor line, flush with the waist of the mold housing. Upon detecting workpiece shading, it outputs a switching signal, triggering the stop mechanism. The stop mechanism consists of a 20mm diameter, 30mm stroke double-acting cylinder and a polyurethane-coated baffle. When the solenoid valve coil is energized, the cylinder piston rod extends, stopping the mold housing at the gripping reference point. The dual-axis linear module has an effective horizontal axis stroke of 600mm and a vertical axis stroke of 200mm, both driven by 86 stepper motors via ball screws, achieving a positioning repeatability of ±0.02mm. The vertical axis end connects to a flexible gripper via an aluminum flange. The gripper fingers are covered with silicone pads, and pneumatic pressure drives the three fingers to converge synchronously, with a gripping force adjustable to 15N. After gripping, the linear module moves the mold housing to the rotating tray above the cleaning module station for release.

[0048] The spray and ultrasonic cleaning module internally includes a sealed spray chamber, a cleaning fluid spray pipeline, an immersion ultrasonic cleaning tank, an ultrasonic generation system, and a tray lifting and transfer device. The spray chamber is welded from 304 stainless steel plate, with a row of fan-shaped nozzles (Lechler 652.404) at the top and bottom, each with a 40° spray angle. The nozzles are arranged opposite each other, with the spacing determined by the mold shell height. The cleaning fluid is supplied by a vertical multistage centrifugal pump with a rated pressure of 0.5 MPa. A 10μm bag filter and an electric heating element are connected in series in the pipeline to maintain the cleaning fluid temperature at 45±2℃. The cleaning fluid is a 3% (v / v) weakly alkaline water-based cleaning agent. The spray timing is controlled by a PLC-controlled solenoid valve, set to 120 seconds. The ultrasonic cleaning tank is located behind the spray chamber, with an effective volume of 80L. Six sets of vibrating plates are installed on the inner wall, each set with 12 piezoelectric ceramic transducers. The operating frequency is 28kHz, and the total ultrasonic power is nominally 1200W, driven by an ultrasonic generator with automatic frequency tracking. The tray lifting and transfer device consists of a vertical guide column, a cylinder-driven lifting platform, and a rotatable tray. The cylinder stroke is 150mm, which can lower the tray carrying the mold shell from the spray chamber to 50mm below the ultrasonic tank liquid level. After cleaning, it is raised back to the transfer height. The tray is supported by sealed bearings and driven by a micro gear motor to rotate at a speed of 10r / min.

[0049] The rinsing and hot air drying module internally includes a clean water rinsing tank, a pure water spray pipeline, a six-axis transfer robotic arm, a hot air generator and blowing device, and a dew point sensor. The clean water rinsing tank has a capacity of 120L and a row of spray nozzles at the bottom. A circulating pump delivers deionized water, filtered through a 0.22μm precision filter, into the tank to create turbulence. An overflow outlet maintains a constant liquid level. A conductivity probe monitors the conductivity of the rinsing water online, automatically initiating water replenishment and wastewater discharge when it exceeds 5μS / cm. The pure water spray pipeline is located above the rinsing tank outlet and features a set of needle-shaped nozzles that perform a final rinse of the mold shell at a pressure of 0.3MPa. The transfer robotic arm is a six-axis articulated type with a two-finger gripper at the wrist, achieving a repeatability of ±0.05mm. It is used to grasp the cleaned mold shell and transfer it from the ultrasonic tank to the rinsing tank, and then to the drying station. The hot air generation and blowing device consists of a volute-less centrifugal fan, a finned electric heater, a high-efficiency air filter, and hollow multi-directional air knives. The fan draws in ambient air, which is heated to 80°C by the heater and then passes through an H13-grade HEPA filter before being distributed to multiple rectangular flat air knives via air ducts. The air velocity at the air knife outlet is approximately 25 m / s, blowing simultaneously into the inner and outer cavities of the mold shell from four directions: top, bottom, left, and right. A SensirionSHT35 dew point sensor is installed at the exhaust vent of the drying station. The probe measures the dew point temperature of the exhaust air to determine the drying endpoint.

[0050] The transient pulse excitation submodule includes: an oil-free air compressor, a compressed air conditioning unit, a high-precision electro-proportional pressure regulating valve, a high-speed response two-position two-way solenoid valve, a dedicated internal cavity sealing nozzle assembly, and a timing synchronization control board. The oil-free air compressor has a rated discharge pressure of 0.8 MPa and a displacement of 40 L / min, with its outlet connected to the compressed air conditioning unit via copper tubing. The compressed air conditioning unit consists of a pre-filter, a precision oil removal filter, an activated carbon adsorption cartridge, and a refrigerated dryer connected in series, outputting clean compressed air with a dew point of -20℃ and an oil content below 0.003 mg / m³. The electro-proportional pressure regulating valve, model SMCITV2050, has an input pressure range of 0 to 1 MPa and an output pressure continuously adjustable within the range of 0.005 to 0.5 MPa, with a linearity of ±0.5% of full scale. It accepts 0 to 10V analog voltage or RS-485 communication commands to set the target pressure. The high-speed solenoid valve is a direct-acting type with a 2mm diameter and a response time of less than 5ms. The valve body is made of 316 stainless steel, and the coil power is 8W. The outlet is connected to the sealing nozzle via a 300mm long, 4mm inner diameter nylon tube. The sealing nozzle assembly consists of an aluminum alloy base, a conical sealing head, and an O-ring fluororubber seal. The base has a flow channel machined into its inner cavity, and the sealing head has a 60° cone angle. During installation, the nozzle is pressed against the mold gate end face to form an airtight contact, and the pulsed airflow is injected into the mold cavity through the flow channel.

[0051] The digital speckle imaging and strain calculation submodule uses a micro-sprayer to spray random black paint speckles with an average particle size of 80 μm and a coverage of 50%. Two high-speed cameras with a pixel size of 5.5 μm and a resolution of 2048×2048px simultaneously acquire 5 frames of images before and after pulse excitation under 460nm ring blue light illumination at a frame rate of 10000fps and an exposure time of 10μs. The timing synchronization controller keeps shutter jitter within 50ns. After epipolar correction, the images are sent to the embedded processing unit. This unit runs a multi-scale wavelet domain digital speckle correlation algorithm, first performing L=3-level two-dimensional discrete wavelet transform on the reference image f and the deformed image g to obtain the wavelet coefficients at each level. and Then, a coarse-to-fine search strategy is adopted, based on scale weights. After calculating the weighted normalized correlation function S and determining the most relevant integer pixel positions, the full-field displacement vector is obtained through sub-pixel interpolation. Finally, the Green-Lagrange strain components are calculated from the displacement gradient field, and the measured strain contour map is generated. The full-field solution takes less than 50ms.

[0052] The multi-scale wavelet domain weighted normalized correlation function is intuitively represented as:

[0053]

[0054] In the formula:

[0055] S is the weighted correlation coefficient, with a value ranging from [-1; 1];

[0056] l is the wavelet decomposition layer number, ranging from 1 to the total number of layers L. In this example, L=3. These are the scale weights of the l-th layer, with higher weights assigned to coarse scales to suppress noise.

[0057] k iterates through all row and column indices of the l-th layer wavelet coefficient matrix;

[0058] and The two-dimensional discrete wavelet coefficients of the reference image f and the deformed image g at the l-th and k-th coefficient positions, respectively, are obtained by Daubechiesdb4 wavelet basis decomposition. This formula, through multi-scale fusion, suppresses speckle noise while preserving high-frequency edge details, thereby improving the detection sensitivity of small strain fields by orders of magnitude.

[0059] The defect intelligent assessment and cleanliness detection submodule is equipped with a deep learning inference engine. It first feeds the measured strain cloud map of 2048×2048px into a deep separable convolutional feature extraction network, and outputs a 512-dimensional feature vector. The database stores a set of strain contour fingerprints from 50 qualified mold shells, and the baseline feature mean vector μ and the 512×512 covariance matrix have been pre-calculated. During each test, the engine calculates the squared Mahalanobis distance. As an indicator of strain anomaly scale. Simultaneously, a laser particulate sensor detects the gas inside the cavity at a sampling flow rate of 28.3 L / min, outputting the mass concentration (CC) of particles larger than 0.3 μm, in units of... The cleanliness limit is preset to .

[0060] The intelligent defect assessment and cleanliness detection submodule proposes a strain-cleanliness dual-factor defect severity index H, which directly multiplies the saturated Marvin contribution by the particulate matter exceedance ratio, outputting a continuous value from 0 to 1, and giving a qualified / unqualified judgment according to a threshold of 0.5. At the same time, the abnormal area is back-projected onto the cloud map to generate a defect map, and finally the judgment signal and graphic data are sent to the sorting system and the host computer through the I / O interface.

[0061] This submodule is equipped with a strain-cleanliness dual-factor defect severity index, which is visually represented as follows:

[0062]

[0063] In the formula:

[0064] H is the defect severity index, which is dimensionless.

[0065] For Mahalanobis distance, by definition

[0066] C represents the measured mass concentration of particulate matter in the cavity, in units of...

[0067] For cleanliness limits, take... ;

[0068] The formula is passed By compressing the unbounded Mahalanobis distance to the [0,1] interval and then directly multiplying it with the particulate matter exceedance ratio, equal-weighted fusion of the two abnormal signals of strain and cleanliness is achieved, thereby improving the defect detection rate.

[0069] The sorting and unloading module is internally equipped with: a four-axis SCARA robot, two independent gravity chutes, a qualified product palletizing and positioning station, and a waste receiving box. The SCARA robot has a 600mm reach and is equipped with a pneumatic gripper at the end. The gripper fingers are bonded with wear-resistant rubber, and the gripping force is adjustable. The robot controller receives the judgment signal output by the detection module through an Ethernet interface and places the mold shell into the inlet bracket of the corresponding chute according to the qualified or unqualified command.

[0070] The qualified product chute is made of stainless steel plate bent into a U-shaped channel with a cross-sectional width of 320mm and an inclination angle of 15°. The end is connected to a lifting palletizing and positioning platform, which consists of a ball screw lifting mechanism and cross roller guides. It automatically descends one workpiece height as the workpieces are stacked layer by layer. The waste receiving bin is a steel mesh bin with a buffer baffle at its inlet to guide the unqualified mold shells to slide down. After sorting, the robot sends a sorting completion flag to the EtherCAT bus for the host computer to read the status.

[0071] From the above, we can conclude that:

[0072] In this invention, by using transient gas pressure excitation and multi-scale wavelet domain digital speckle correlation detection, the ability to identify microcracks is advanced from surface-opening defects to the subsurface strain anomaly stage, significantly improving the detection sensitivity for defects such as hidden cracks and wall thinning. Simultaneously, full-field strain cloud map quantitative grading replaces manual visual experience judgment, greatly reducing the missed detection rate and quality risks caused by human error. The strain-cleanliness dual-factor fusion assessment simultaneously completes the assessment of structural integrity and internal cavity cleanliness at the same workstation, improving the completeness and automation of defect evaluation and reducing the delays in production cycle caused by batch offline sampling and chemical permeation operations.

[0073] In this invention, the transient pulse excitation method eliminates the need for fluorescent penetrant soaking and cleaning processes, increasing the single-piece inspection cycle and reducing the costs of hazardous chemical procurement, storage, and waste liquid treatment. The entire line, from self-positioning gripping and spray ultrasonic cleaning to dew point closed-loop drying and robot-driven automatic sorting directly by judgment signals, forms a continuous flow, improving the consistency of mold shell processing and data closed-loop traceability, while reducing the risk of bumps and scratches introduced by manual handling and repeated clamping and positioning.

[0074] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0075] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An automatic detection and cleaning system for oriented crystal mold shells, characterized in that: The automatic feeding and positioning module, spraying and ultrasonic cleaning module, rinsing and hot air drying module, transient air pressure-DIC detection module and sorting and unloading module are described. The mold shell transfer output end of the automatic feeding and positioning module is connected to the rotating tray input end of the spray and ultrasonic cleaning module, and its positioning completion signal output end is connected to the process start control end of the spray and ultrasonic cleaning module. The cleaning completion mold output end of the spray and ultrasonic cleaning module is connected to the rinsing tank inlet end of the rinsing and hot air drying module. The drying completion output of the rinsing and hot air drying module is connected to the detection station input of the transient air pressure-DIC detection module, and its drying compliance signal output is connected to the detection trigger enable of the transient air pressure-DIC detection module. The mold shell discharge end of the transient air pressure-DIC detection module is connected to the workpiece input end of the sorting and unloading module, its judgment signal output end is connected to the sorting action control input end of the sorting and unloading module, and its network interface is connected to the host computer database. The sorting completion status feedback signal of the sorting and unloading module is connected to the host computer database. The transient pressure-DIC detection module includes: a transient pulse excitation submodule, a digital speckle imaging and strain calculation submodule, and a defect intelligent assessment and cleanliness detection submodule.

2. The automatic detection and cleaning system for directional crystal mold shells as described in claim 1, characterized in that: The automatic feeding and positioning module is equipped with a chain conveyor, a pneumatic stop mechanism, a set of reflective photoelectric sensors, a dual-axis linear module, and a flexible gripper at the end. The chain conveyor uses stainless steel chain plates with a pitch of 38.1mm to form a 400mm wide conveying surface. The end of the inclined material channel is connected to the beginning of the conveyor, and the mold shell inside the material frame slides into the conveyor line by gravity. Guide rails are installed on both sides of the conveying channel, and the spacing is adjustable.

3. The automatic detection and cleaning system for directional crystal mold shells as described in claim 2, characterized in that: The photoelectric sensor is an Omron E3Z-R61. The transmitter and receiver are located on opposite sides of the conveyor line, and the installation height is flush with the waist of the mold shell. After detecting that the workpiece is blocking the light, it outputs a switch signal to trigger the stop mechanism. The stop mechanism consists of a double-acting cylinder with a diameter of 20mm and a stroke of 30mm and a baffle plate with polyurethane covering the front end. After the solenoid valve coil is energized, the cylinder piston rod extends, stopping the mold shell at the gripping reference point.

4. The automatic detection and cleaning system for directional crystal mold shells as described in claim 1, characterized in that: The spray and ultrasonic cleaning module is internally equipped with a sealed spray chamber, a set of cleaning fluid spray pipelines, an immersion ultrasonic cleaning tank, an ultrasonic generating system, and a tray lifting and transfer device. The spray chamber is welded from 304 stainless steel plate, with a row of fan-shaped nozzles at the top and bottom. The nozzles are Lechler 652.404, with a spray angle of 40°. The upper and lower nozzles are arranged opposite each other, and the spacing is set according to the height of the mold shell. The cleaning fluid is supplied by a vertical multi-stage centrifugal pump with a rated pressure of 0.5MPa. A 10μm bag filter and an electric heating tube are connected in series in the pipeline to maintain the cleaning fluid temperature at 45±2℃. The cleaning fluid is a weakly alkaline water-based cleaning agent with a volume concentration of 3%.

5. The automatic detection and cleaning system for directional crystal mold shells as described in claim 1, characterized in that: The rinsing and hot air drying module is equipped with a clean water rinsing tank, a set of pure water spray pipelines, a six-axis transfer robotic arm, a set of hot air generation and blowing devices, and a dew point sensor. The clean water rinsing tank has a capacity of 120L and a row of spray agitators at the bottom. Deionized water is sent into the tank through a 0.22μm precision filter by a circulating pump to form turbulence. The overflow port maintains a constant liquid level. The conductivity probe monitors the conductivity of the rinsing water online. When it exceeds 5μS / cm, it automatically starts water replenishment and sewage discharge.

6. The automatic detection and cleaning system for directional crystal mold shells as described in claim 1, characterized in that: The transient pulse excitation submodule includes: an oil-free air compressor, a compressed air conditioning unit, a high-precision electro-proportional pressure regulating valve, a high-speed response two-position two-way solenoid valve, a dedicated internal cavity sealing nozzle assembly, and a timing synchronization control board; the oil-free air compressor has a rated discharge pressure of 0.8MPa and a discharge capacity of 40L / min, and its outlet is connected to the compressed air conditioning unit via a copper pipe.

7. The automatic detection and cleaning system for directional crystal mold shells as described in claim 1, characterized in that: The digital speckle imaging and strain calculation submodule uses a micro-sprayer to spray random black paint speckles with an average particle size of 80μm and a coverage of 50%. Two high-speed cameras with a pixel size of 5.5μm and a resolution of 2048×2048px simultaneously acquire 5 frames of images before and after pulse excitation under 460nm ring blue light illumination at a frame rate of 10000fps and an exposure time of 10μs. The timing synchronization controller controls the shutter shake within 50ns. After epipolar correction, the images are sent to the embedded processing unit. This unit runs a multi-scale wavelet domain digital speckle correlation algorithm, first performing L=3 levels of two-dimensional discrete wavelet transform on the reference image f and the deformed image g to obtain the wavelet coefficients at each level. and Then, a coarse-to-fine search strategy is adopted, based on scale weights. After calculating the weighted normalized correlation function S and determining the most relevant integer pixel positions, the full-field displacement vector is obtained through sub-pixel interpolation. Finally, the Green-Lagrange strain components are calculated from the displacement gradient field, and the measured strain contour map is generated. The full-field solution takes less than 50ms. The multi-scale wavelet domain weighted normalized correlation function is intuitively represented as: In the formula: S is the weighted correlation coefficient, with a value ranging from [-1; 1]; l is the wavelet decomposition layer number, ranging from 1 to the total number of layers L. In this example, L=3. It is the scale weight of the l-th layer, and a higher weight is given to the coarse scale to suppress noise; k iterates through all row and column indices of the l-th layer wavelet coefficient matrix; and The two-dimensional discrete wavelet coefficients of the reference image f and the deformed image g at the l-th and k-th coefficient positions are respectively obtained by Daubechiesdb4 wavelet basis decomposition.

8. The automatic detection and cleaning system for directional crystal mold shells as described in claim 1, characterized in that: The defect intelligent assessment and cleanliness detection submodule is equipped with a deep learning inference engine. It first feeds the measured strain cloud map of 2048×2048px into a deep separable convolutional feature extraction network, and outputs a 512-dimensional feature vector. ; The database stores a set of strain contour fingerprints of 50 qualified mold shells, and the baseline feature mean vector μ and the 512×512 covariance matrix have been pre-calculated. During each test, the engine calculates the squared Mahalanobis distance. As an indicator of strain anomaly scale; simultaneously, the laser particulate sensor detects the gas inside the cavity at a sampling flow rate of 28.3 L / min, outputting the mass concentration (CC) of particles larger than 0.3 μm, in units of ... The cleanliness limit is preset to ; The defect intelligent assessment and cleanliness detection submodule proposes a strain-cleanliness dual-factor defect severity index H, which directly multiplies the saturated Marvin contribution with the particulate matter exceedance ratio, outputs a continuous value from 0 to 1, and gives a qualified / unqualified judgment according to a threshold of 0.

5. At the same time, the abnormal area is back-projected onto the cloud map to generate a defect map, and finally the judgment signal and graphic data are sent to the sorting system and the host computer through the I / O interface. It is equipped with a strain-cleanliness dual-factor defect severity index, which is visualized as follows: In the formula: H is the defect severity index, which is dimensionless. For Mahalanobis distance, by definition C represents the measured mass concentration of particulate matter in the cavity, in units of... For cleanliness limits, take... ; The formula is passed The unbounded Mahalanobis distance is compressed to the [0,1] interval and then directly multiplied with the particulate matter exceedance ratio to achieve equal-weighted fusion of the two abnormal signals of strain and cleanliness.

9. The automatic detection and cleaning system for directional crystal mold shells as described in claim 1, characterized in that: The sorting and unloading module is internally equipped with: a four-axis SCARA robot, two independent gravity slides, a qualified product palletizing and positioning platform, and a waste receiving box; the SCARA robot has a 600mm arm span and is equipped with a pneumatic gripper at the end. The gripper fingers are bonded with wear-resistant rubber, and the gripping force is adjustable. The robot controller receives the judgment signal output by the detection module through an Ethernet interface and places the mold shell into the inlet bracket of the corresponding slide according to the qualified and unqualified instructions.

10. The automatic detection and cleaning system for directional crystal mold shells as described in claim 9, characterized in that: The qualified product chute is made of stainless steel plate bent into a U-shaped channel with a cross-sectional width of 320mm and an inclination angle of 15°. The end is connected to a lifting palletizing and positioning platform, which consists of a ball screw lifting mechanism and cross roller guide rails. It automatically descends one workpiece height as the workpieces are stacked layer by layer. The waste receiving box is a steel mesh box with a buffer baffle at its entrance to guide the unqualified mold shells to slide down. After sorting, the robot sends a sorting completion flag to the EtherCAT bus for the host computer to read the status.

Images