Uncertainty Information Gain-Based Modality-Adaptive Non-Destructive Robot Inspection System, Method for Generating Defect Coordinate Map thereof, and Process Control Method Interlocked with Manufacturing Execution System

Abstract

The present invention relates to a non-destructive inspection and process control system (2mons system) using autonomous adaptive closed-loop feedback of a multi-axis operating robot. The system according to the present invention includes: a target recognition unit that acquires external shape information or a reference model of an object to be inspected and sets an object coordinate system; a robot inspection unit that inspects or receives non-destructive inspection signals at different time points for an object to be inspected; a modality adapter unit that variably sets end device alignment distance, contact pressure, inspection angle, detector placement, inspection energy, pulse width, sampling timing, or coupling conditions according to the type of selected non-destructive inspection modality; a defect analysis unit that calculates a defect candidate region and an uncertainty region from primary inspection data; a trajectory reconstruction unit that recalculates a subsequent inspection position, joint angle, and inspection conditions based on a cost function including expected information gain for the uncertainty region, robot driving constraints, and physical constraints per modality; a defect map generation unit that generates a defect coordinate map corresponding to the object coordinate system using re-inspection data; and a judgment and process control linkage unit that outputs a quality judgment signal or a process correction signal based on the defect coordinate map. According to the present invention, inspection trajectories and modality conditions can be adaptively reconfigured around high-uncertainty regions, and the generated defect coordinate map can be linked with a manufacturing execution system, a monitoring and control system, or a programmable logic controller to update control parameters, set points, or actuator driving conditions of the entire process manufacturing equipment. In addition, the present invention is configured such that primary inspection data, uncertainty fields, expected information gains, physical constraints per modality, and process control parameters are combined in a closed-loop manner, thereby providing a non-linear inspection efficiency improvement and process stabilization effect that concentrates inspection resources on specific uncertainty areas and reflects subsequent inspection results back into defect coordinate maps and process correction signals, rather than a linear improvement that simply increases the number of inspections or sensors.

Description

Technology Field

[0001] The present invention relates to a non-destructive inspection system using an industrial multi-axis movable robot.

[0002] More specifically, the present invention relates to a robot non-destructive inspection and process control technology that calculates an uncertainty region inside or below the surface of an object under inspection from primary non-destructive inspection data, autonomously reconstructs the subsequent inspection trajectory and inspection conditions of a robot by comprehensively considering the expected information gain for the uncertainty region, robot driving constraints, and physical constraints for each non-destructive inspection modality, generates a defect coordinate map corresponding to the object under inspection coordinate system using re-inspection data, and updates process control parameters linked to a manufacturing execution system (MES), a supervisory control and analysis system (SCADA), or a programmable logic controller (PLC) based on the defect coordinate map. Background Technology

[0003] In advanced semiconductor packaging, battery cells, all-solid-state battery laminates, large die-casting products, aerospace composites, heterogeneous material bonded structures, and high-reliability manufactured parts, it is necessary to non-destructively inspect internal pores, microcracks, interlayer delamination, adhesion failures, foreign matter ingress, electrode folding, bump voids, shrinkage cavities, and interfacial defects that are difficult to observe directly from the outside.

[0004] Conventional automatic defect recognition systems generally have a structure that photographs an object to be inspected according to a predetermined shooting position or a fixed inspection trajectory, and determines whether there is a defect by analyzing the acquired radiographic image, ultrasonic signal, optical image, or thermal image data.

[0005] However, conventional inspection methods often have fixed inspection trajectories and sensor conditions, even though the required inspection modalities and conditions vary depending on the material, thickness, internal structure, surface curvature, and defect type of the object being inspected. As a result, reading errors due to radiation attenuation and scattering may occur in thick metal parts, subsurface defects in composites or adhesive layers may not be sufficiently identified by a single sensor alone, and in ultrasonic inspection, it is difficult to reliably obtain an effective signal if the contact pressure, surface normal alignment, and coupling conditions are improper.

[0006] Furthermore, conventional technology often limits defect detection results to simply determining pass or fail, and has limitations in that the spatial coordinates, defect type, defect distribution, and occurrence location of defects are not synchronized in real-time with the control parameters of the actual manufacturing process. Consequently, even if the same type of defect occurs repeatedly, the pressure, tension, position, temperature, speed, time, or actuator setpoint of the front-end equipment cannot be immediately corrected, which may lead to a decrease in production yield and an increase in quality costs.

[0007] In particular, fixed-trajectory-based inspection tends to inspect the entire area of ​​the object under the same resolution and conditions, which leads to a problem where inspection resources are excessively allocated to areas with a low probability of being defect candidates, while conversely, sufficient additional information is not obtained in areas where the presence of defects is ambiguous. This problem increases inspection time, cumulative energy exposure, robot movement costs, and sensor usage costs, while simultaneously limiting the reliability of judgment regarding actual defect candidate areas.

[0008] Therefore, a new robotic non-destructive inspection system is required that can calculate an area where the presence of a defect is uncertain from the results of a first inspection, non-linearly reconstruct subsequent inspection trajectories and inspection modality conditions around the area of ​​uncertainty, and update process control parameters by linking the acquired defect coordinate information with the manufacturing execution system and the field control layer. The problem to be solved

[0009] The first problem that the present invention aims to solve is to provide a robotic non-destructive inspection system capable of calculating candidate defect areas and uncertainty areas inside or below the surface of a test object from primary non-destructive inspection data, and quantifying said uncertainty areas.

[0010] The second problem that the present invention aims to solve is to provide a trajectory reconstruction structure that recalculates subsequent inspection positions, joint angles, and inspection conditions by considering the robot's collision constraints, joint torque constraints, inspection time constraints, and physical constraints of the selected non-destructive inspection modality, while maximizing the expected information gain for the uncertainty region.

[0011] The third problem that the present invention aims to solve is to provide a modality adapter unit that variably sets the alignment distance, contact pressure, irradiation angle, detector placement, irradiation energy, pulse width, sampling timing, or coupling conditions of a robot end device according to the type of selected non-destructive inspection modality.

[0012] The fourth problem that the present invention aims to solve is to provide a closed-loop process control structure that fuses primary inspection data and subsequent re-inspection data to generate a defect coordinate map corresponding to the coordinate system of the object under inspection, and links a quality judgment signal or a process correction signal with a manufacturing execution system, a monitoring and control system, or a programmable logic controller based on the defect coordinate map.

[0013] The fifth problem that the present invention aims to solve is to provide an adaptive manufacturing inspection platform in which primary inspection data, uncertainty fields, expected information gain, physical constraints by modality, and process control parameters are combined to induce non-linear inspection efficiency improvement and defect detection reliability improvement, rather than simple linear summation. means of solving the problem

[0014] The uncertainty information gain-based modality adaptive robot non-destructive inspection system according to the present invention for solving the above problem is,

[0015] A target recognition unit that acquires external shape information or a reference model of a test object and sets the coordinate system of the test object;

[0016] A robot inspection unit configured to variably drive multiple axes with respect to the above-mentioned test object to investigate or receive non-destructive inspection signals at specific points in time;

[0017] A modality adapter unit that variably sets at least one of the end device alignment distance, contact pressure, irradiation angle, detector placement, irradiation energy, pulse width, sampling timing, or coupling conditions of the robot inspection unit according to the type of selected non-destructive inspection modality;

[0018] A defect analysis unit that calculates candidate defect regions and uncertainty regions inside or beneath the surface of the inspected object from primary inspection data;

[0019] A trajectory reconstruction unit that recalculates subsequent inspection positions, joint angles, and inspection conditions based on a cost function including expected information gain for the above-mentioned uncertainty region and collision constraints, driving torque constraints, inspection time constraints, and signal quality constraints per modality of the robot inspection unit;

[0020] A defect map generation unit that generates a defect coordinate map corresponding to the coordinate system of the inspected body by fusing the re-inspection data obtained by the above subsequent inspection with the above first inspection data; and

[0021] It includes a judgment and process control linkage unit that outputs a quality judgment signal or a process correction signal based on the above defect coordinate map, and links with at least one of an external manufacturing execution system, a monitoring and control system, or a programmable logic controller to update control parameters, set points, or actuator driving conditions of the entire process manufacturing equipment.

[0022] The above non-destructive testing modality may include at least one of a transmitted radiation modality, an elastic wave modality, a terahertz electromagnetic wave modality, and an active thermal response modality. Effects of the invention

[0023] According to the present invention, since subsequent precision inspections can be performed focusing on the uncertainty area calculated from primary inspection data without collectively inspecting the entire area of ​​the object under inspection at high resolution, the use of inspection resources can be reduced while increasing the reliability of identifying candidate defect areas.

[0024] In addition, the present invention allows the modality adapter to reflect the physical requirements of the selected non-destructive testing modality, so that radiation testing, seismic testing, terahertz testing, and active thermal response testing can be adaptively operated on a single robot platform according to the purpose.

[0025] In addition, since the present invention recalculates subsequent inspection positions and inspection conditions using uncertainty fields and expected information gains, it is possible to perform adaptive inspection corresponding to the shape, material, and defect candidate distribution of the inspection target compared to the conventional fixed trajectory inspection method.

[0026] In addition, since the present invention can update process control parameters by linking the defect coordinate map with a manufacturing execution system, a monitoring and control system, or a programmable logic controller, the defect detection results can be utilized for adjusting pre-process control conditions rather than being limited to simple post-process screening.

[0027] In addition, the present invention is configured such that primary inspection data, uncertainty index, expected information gain, physical constraints by modality, and process control parameters influence each other in a closed-loop manner. Accordingly, rather than a linear improvement that simply increases the number of inspections, it is possible to provide a non-linear effect in which inspection resources are concentrated in a specific uncertainty area, and additional information obtained from that area is reflected back into the defect coordinate map and process correction signal.

[0028] Here, the nonlinear effect refers to the effect where a small difference in signal uncertainty in the first inspection sequentially changes the subsequent inspection trajectory, modality selection, inspection conditions, and process correction parameters, thereby improving the judgment reliability of defect candidate regions and process feedback sensitivity beyond a proportional increase, even at the same inspection time or energy conditions. Specific details for implementing the invention

[0029] Preferred embodiments of the present invention are described below. However, the following embodiments are merely examples to aid in understanding the present invention, and the scope of the present invention is not limited to the following embodiments.

[0030] The frequency, voltage, angle, pressure, error range, inspection cycle, threshold value, etc. described in this specification are exemplary ranges applicable in one embodiment and may be changed depending on the material, thickness, type of internal defect, inspection modality, inspection equipment specifications, and process conditions of the manufacturing line of the object under inspection.

[0031] The present invention specifies the physical layout of the system, data flow, control signal flow, inspection modality switching structure, defect coordinate map generation structure, and manufacturing process feedback flow as follows in a text-based structural description, so that a person skilled in the art can understand the overall structure and operation flow without referring to separate drawings.

[0032] Physical and logical layout of the entire system

[0033] The system according to the present invention can be broadly divided into an object recognition area, a robot inspection area, a central control area, a defect map generation area, and a process control linkage area.

[0034] A test object and a target recognition unit are placed in the target recognition area. The test object may be a battery cell, a semiconductor package, a composite part, a large casting, a heterogeneous material laminated structure, or a high-reliability manufactured part. The target recognition unit acquires external shape information, reference marking, manufacturing lot information, or reference model information of the test object using at least one of a 3D optical camera, a laser scanner, a structured light sensor, LiDAR, a stereo camera, a barcode reader, or an RFID reader.

[0035] The object recognition unit establishes a coordinate system of the object under inspection based on the above-mentioned acquired information. The above-mentioned object coordinate system functions as a reference coordinate system in which the primary inspection data, subsequent re-inspection data, and defect coordinate map acquired thereafter are aligned with each other.

[0036] A robot inspection unit and a modality adapter unit are positioned in the robot inspection area. The robot inspection unit may be composed of a multi-axis manipulator, a gantry robot, a mobile robot platform, a dual robot arm structure, or a combination thereof. A modality adapter unit is coupled to the end device of the robot inspection unit. Depending on the selected non-destructive inspection modality, the modality adapter unit mounts or aligns at least one of a radiation source, a detector, an ultrasonic probe, a terahertz antenna, a thermal stimulus source, or a thermal imaging detector.

[0037] A defect analysis unit and a trajectory reconstruction unit are logically arranged in the central control area. The defect analysis unit calculates candidate defect regions and uncertainty regions using the object coordinate system transmitted from the object recognition unit and primary inspection data transmitted from the robot inspection unit. The trajectory reconstruction unit recalculates subsequent inspection positions, joint angles, and inspection conditions by considering the expected information gain for the uncertainty region, robot driving constraints, and physical constraints by modality.

[0038] A defect map generation unit is positioned in the defect map generation area. The defect map generation unit fuses primary inspection data and subsequent re-inspection data to generate a defect coordinate map corresponding to the coordinate system of the object under inspection. The defect coordinate map may include defect center coordinates, defect volume, defect type, defect reliability, defect distribution density, and manufacturing lot information.

[0039] A judgment and process control linkage unit is disposed in the process control linkage area. The judgment and process control linkage unit receives a defect coordinate map and generates a quality judgment signal or a process correction signal. The process correction signal may be transmitted to at least one of a manufacturing execution system, a monitoring and control system, or a programmable logic controller, and at least one of the pressure, tension, temperature, position, speed, time, flow rate, voltage, current, torque, or actuator setpoint of the front-end manufacturing equipment may be updated.

[0040] Therefore, the present invention has a structure in which object recognition, primary inspection, uncertainty calculation, subsequent trajectory reconstruction, defect coordinate map generation, and process control linkage are combined into a single closed-loop system.

[0041] Calculation of uncertainty region based on first test data

[0042] The robotic inspection unit performs a primary non-destructive inspection on the object under inspection. The primary inspection may be a high-speed inspection to rapidly explore the entire area of ​​the object under inspection, and may be performed by radiographic inspection, seismic inspection, terahertz inspection, active thermal response inspection, or a combination thereof.

[0043] The data obtained from the above first inspection is stored as a first inspection dataset corresponding to the spatial coordinates, surface coordinates, or subsurface coordinates of the object under inspection. Each coordinate x may correspond to at least one of signal attenuation, reflection intensity, scattering characteristics, time delay, thermal response change, ultrasonic echo, or defect class probability.

[0044] The defect analysis unit maps the above primary inspection dataset to a spatial voxel or a subsurface grid. Subsequently, the defect analysis unit calculates a probability density function p(f(x)|D) or a defect class probability pk(x) representing the physical state or the possibility of defect existence at each coordinate. Here, D represents the primary inspection dataset.

[0045] The uncertainty index U(x) at each spatial coordinate x can be calculated by the following mathematical formula 1.

[0046] [Mathematical Formula 1]

[0047] U(x) = - ∫ p(f(x)|D) log p(f(x)|D) df

[0048] In the above equation, x is a spatial coordinate vector inside or below the surface of the object under inspection, D is a primary inspection dataset, and p(f(x)|D) is a probability density function representing the physical state, density, possibility of defect existence, or signal characteristics of the corresponding coordinate when the primary inspection dataset is given.

[0049] In addition, if the defect type is classified into discrete class k, the uncertainty index U(x) can be calculated by the following mathematical formula 2.

[0050] [Mathematical Formula 2]

[0051] U(x) = - Σ(k=1 to K) pk(x) log pk(x)

[0052] In the above formula, K is the number of defect type classes, and pk(x) is the probability that the corresponding coordinate belongs to the k-th defect type.

[0053] The defect analysis unit may calculate a coordinate region where the uncertainty index U(x) exceeds a preset threshold as an uncertainty region. The preset threshold may be, for example, a region with an uncertainty value of 0.6 or higher based on maximum entropy scaling criteria, or a region with an uncertainty value corresponding to the top 10% to 30% of the total spatial coordinates. However, the above values ​​are merely examples of one embodiment and may be changed depending on the inspection target and process conditions.

[0054] The above uncertainty region refers not to simple defect candidate coordinates, but to a spatial region where there is a high probability of obtaining additional information in subsequent inspections. Accordingly, the primary inspection data is transformed not into a simple reading result, but into a mathematical input that determines subsequent inspection actions.

[0055] Reconstruction of follow-up inspection trajectories based on expected information gain

[0056] The trajectory reconstruction unit recalculates the subsequent inspection position, joint angle, and inspection conditions for the above-mentioned uncertainty region.

[0057] The primary inspection trajectory of the robot inspection unit may be a circular, semicircular, linear, spiral, grid, multi-axis free curve, or any path corresponding to the external shape of the object being inspected. However, in the present invention, the subsequent inspection trajectory is not simply a repetition of the primary inspection trajectory, but is reconstructed according to the location, size, density, and information gain directionality of the uncertainty region calculated by the defect analysis unit.

[0058] The trajectory reconstruction unit can calculate the robot's joint angle vector θ and modality driving parameter e using the objective function J(θ,e) according to the following mathematical formula 3.

[0059] [Mathematical Formula 3]

[0060] J(θ,e) = IG(θ,e) - λ1Ccollision - λ2Cexposure - λ3Ctime - λ4Ctorque - λ5Csignal

[0061] In the above equation, IG(θ,e) is the expected information gain function, Ccollision is the collision constraint cost, Cexposure is the signal investigation or contact exposure cost according to the selected modality, Ctime is the inspection time cost, Ctorque is the robot joint torque cost, Csignal is the effective signal-to-noise ratio constraint cost, and λ1 to λ5 are normalized weights for each cost term.

[0062] The above Cexposure may be defined differently depending on the modality. For example, in a transmissive radiation modality, it may correspond to cumulative dose, irradiation energy, or source usage; in an elastic wave modality, it may correspond to contact pressure, amplitude, coupling amount, or contact holding time; in a terahertz electromagnetic wave modality, it may correspond to irradiation output, scan time, or signal attenuation amount; and in an active thermal response modality, it may correspond to thermal stimulation amount, stimulation time, or surface temperature rise amount.

[0063] The above expected information gain function IG(θ,e) can be defined by the following mathematical equation 4.

[0064] [Mathematical Formula 4]

[0065] IG(θ,e) = ∫Ω |∇U(x) · Vgain(x;θ,e)| dx

[0066] In the above equation, Ω is the spatial region to be examined, ∇U(x) is the spatial gradient of the uncertainty index, and Vgain(x;θ,e) is the valid information acquisition weight vector defined in joint angle and modality conditions.

[0067] The trajectory reconstruction unit can recalculate at least one of the subsequent inspection position, robot joint angle, irradiation angle, detector position, source-detector distance, contact pressure, coupling supply amount, sampling timing, irradiation energy, or thermal stimulation conditions in a direction that maximizes the objective function.

[0068] In fixed trajectory inspection, the robot repeats a predetermined path, but in the present invention, the subsequent inspection trajectory is changed non-linearly according to the location, size, density, information gain directionality, and robot driving constraints of the uncertainty region.

[0069] For example, even if the uncertainty region is small, if the region is located at a position of high process importance or has a high information acquisition weight in a specific modality, the trajectory reconstruction unit can concentrate subsequent inspection resources on said region. Conversely, even if the uncertainty region is wide, if the information gain is low or there is a high risk of collision, torque burden, or signal quality degradation, a different angle or a different modality can be selected.

[0070] Physical operating structure of the modality adapter part

[0071] The modality adapter unit is coupled to the end device of the robot inspection unit. The modality adapter unit is not a single sensor fixing unit, but is composed of a variable physical interface to meet the physical requirements of the selected non-destructive inspection modality.

[0072] When a transmissive radiation modality is selected, the modality adapter controls at least one of the opposing alignment distance between the radiation source and the detector, the focus-detector distance, the irradiation angle, the irradiation energy, and the shielding collimator aperture ratio. In this case, the radiation source and the detector may be positioned facing each other with the object under inspection in between, and the modality adapter changes the irradiation conditions according to the thickness, density, inspection target area, and uncertainty distribution of the object under inspection.

[0073] When an elastic wave modality is selected, the modality adapter may include an ultrasonic transducer, a force / torque sensor, a coupling supply unit, and a contact pressure control unit. In this case, the transducer at the robot end is aligned with the normal direction of the surface of the object under inspection, a coupling is supplied, and the contact pressure is feedback controlled. Unlike free-space transmissive sensors, the above structure ensures the feasibility of ultrasonic inspections requiring surface contact and acoustic impedance matching.

[0074] When a terahertz electromagnetic wave modality is selected, the modality adapter unit controls at least one of the incident angle of the terahertz antenna, polarization conditions, time domain sampling timing, reflective inspection mode, transmissive inspection mode, and separation distance from the surface of the object to be inspected. The terahertz signal may be incident toward a non-metallic layer, a dielectric layer, an adhesive layer, a coating layer, or a protective layer, and a reflected or time-delayed signal may be received.

[0075] When the active thermal response modality is selected, the modality adapter controls the thermal stimulus source and the thermal image detector. Thermal stimulus may be applied to the surface of the object under inspection through a flash lamp, laser, hot air, or induction heating source, and the thermal image detector may collect surface temperature attenuation, phase delay, or thermal diffusion patterns over time.

[0076] Therefore, the modality adapter of the present invention does not simply list a plurality of inspection modalities, but provides a structure that dynamically matches the physical conditions required by each modality at the robot end.

[0077] Structure of non-linear effect generation

[0078] In the present invention, the nonlinear effect is not simply caused by using multiple inspection modalities in parallel. The nonlinear effect of the present invention is caused by a closed-loop structure in which the uncertainty index U(x) calculated from primary inspection data, the spatial gradient ∇U(x) of the uncertainty index, the weight Vgain for valid information acquisition per modality, the physical constraint cost of the robot, and the process control parameter are combined in a chain.

[0079] For example, even if the uncertainty index of a specific area in the primary inspection data slightly exceeds a preset threshold, if the area is a process-sensitive location with a high probability of defect occurrence or if the information acquisition weight of a specific modality directed toward that area is high, the trajectory reconstruction unit can significantly change the subsequent inspection location and inspection conditions. In this case, inspection resources are not simply distributed evenly, but are locally redistributed according to the spatial concentration of uncertainty and the directionality of expected information gain.

[0080] In addition, data obtained from a subsequent re-inspection changes the reliability of the defect coordinate map, and the defect coordinate map can be used again to update the setpoint of the entire process equipment through a manufacturing execution system, a monitoring and control system, or a programmable logic controller. Accordingly, the present invention provides a process inspection platform in which inspection area selection, inspection modality selection, robot trajectory control, and process parameter updating are non-linearly combined.

[0081] Accordingly, the present invention is distinguished from linear improvements that simply increase the number of sensors or the number of shots. In the present invention, small differences in uncertainty change subsequent inspection trajectories and inspection conditions, and additional information obtained as a result is transmitted back as a defect coordinate map and process correction signal, thereby simultaneously improving inspection reliability and process stability.

[0082] Generation of defect coordinate maps

[0083] The defect map generation unit fuses the primary inspection data and the re-inspection data obtained by the subsequent inspection to generate a defect coordinate map corresponding to the coordinate system of the object under inspection.

[0084] The above defect coordinate map may include defect center coordinates, defect volume, defect area, defect length, defect depth, defect type label, reliability, defect distribution density, and manufacturing lot information.

[0085] The above defect coordinate map can be implemented as a 3D voxel map, a surface sub-depth map, a CAD coordinate system alignment map, a defect probability map, a defect type classification map, or a combination thereof.

[0086] The defect map generation unit may use backprojection, algebraic iterative reconstruction, probabilistic reconstruction, multiview fusion, sensor fusion, or a combination thereof.

[0087] When radiation data is used, projection data from multiple time points can be reconstructed into defect voxel information. When seismic data is used, time delay or echo signals can be converted into depth information. When terahertz data is used, reflection time or phase change can be converted into interfacial defect information. When active thermal response data is used, surface temperature attenuation or thermal diffusion phase difference can be converted into subsurface defect candidate information.

[0088] As such, inspection data does not end as a simple read-out image but is structured into a defect coordinate map connected to the coordinate system of the object under inspection.

[0089] Process control interlock

[0090] The judgment and process control linkage unit outputs a quality judgment signal or a process correction signal based on the above defect coordinate map.

[0091] The above quality judgment signal may include at least one of pass, fail, re-inspection, hold, repair required, or process condition verification required.

[0092] The above process correction signal can be transmitted to at least one of a manufacturing execution system (MES), a supervisory control and analysis system (SCADA), or a programmable logic controller (PLC).

[0093] Based on the above process correction signal, at least one of the pressure, tension, temperature, position, speed, time, flow rate, voltage, current, torque, or actuator setpoint of the entire process manufacturing equipment can be updated.

[0094] The above communication can be performed via MQTT, OPC-UA, industrial Ethernet, fieldbus, real-time control network, or a combination thereof.

[0095] The judgment and process control linkage unit can generate a process correction signal to update the control parameters of the upstream process equipment associated with a defect when a specific defect type is repeatedly detected in the defect coordinate map or when the defect distribution density at a specific location exceeds a reference value. For example, if electrode folding defects are repeatedly detected in a battery cell, winding tension or stacking pressure may be updated, and if shrinkage cavities are repeatedly detected in a casting, injection pressure, cooling water flow rate, or mold temperature conditions may be updated.

[0096] Accordingly, the present invention provides a closed-loop manufacturing control structure in which the results of non-destructive inspection go beyond simple post-process quality judgment and are connected to the updating of pre-process control parameters.

[0098] Examples

[0099] Example 1: Application of a radiation modality for inspecting internal defects in battery cells

[0100] In this embodiment, the test subject may be a pouch-type battery cell, a cylindrical battery cell, a prismatic battery cell, or an all-solid-state battery stacked cell.

[0101] The target recognition unit acquires the outer dimensions of the battery cell, the electrode tab position, the cell edge, reference marking, or manufacturing lot information, and aligns it to a three-dimensional coordinate system. The alignment tolerance can be set, for example, in the range of 0.01 mm to 0.5 mm.

[0102] The robotic inspection unit selects a penetrating radiation modality to perform a primary high-speed scan. In the primary high-speed scan, the tube voltage of the radiation source can be selected in a range of, for example, 40 kV to 225 kV, depending on the thickness of the battery cell, the density of the outer material, and the type of defect to be inspected. The shooting angle pitch can be set in a range of, for example, 10 degrees to 30 degrees.

[0103] The defect analysis unit calculates candidate regions for electrode folding, foreign matter ingress, electrode overlap defects, tab welding defects, or internal short circuits from the primary scan data. If the uncertainty index U(x) of a specific spatial voxel exceeds a preset threshold, the trajectory reconstruction unit reduces the subsequent inspection angle pitch centered on the region to a range of, for example, 1 to 5 degrees, and performs a local re-inspection by adjusting the focus-detector distance, collimator aperture ratio, or irradiation energy through the modality adapter unit.

[0104] The defect map generation unit fuses primary inspection data and re-inspection data to generate a defect coordinate map corresponding to the battery cell coordinate system. The defect coordinate map may include defect center coordinates, defect volume, defect type, and judgment reliability.

[0105] The judgment and process control linkage unit transmits the defect coordinate map to an MES or SCADA and enables at least one of the electrode coating die gap, winding tension, stacking pressure, tap welding time, or actuator setpoint to be updated through a PLC linked thereto. The feedback cycle may be set, for example, in a range of 1 second to 60 seconds or per lot.

[0106] In the above embodiment, if only some areas have a high uncertainty index in the first scan, subsequent inspection resources are concentrated on those areas. Accordingly, compared to a method of repeatedly photographing the entire battery cell at the same resolution, the information density regarding candidate defect areas can increase non-linearly. That is, the increase in the reliability of judgment regarding candidate defect areas can be greater than the simple increase in the number of photographs.

[0108] Example 2: Application example of a composite modality for semiconductor package micro-void inspection

[0109] In this embodiment, the test specimen may be a 2.5D package, a 3D silicon interposer, a high-bandwidth memory package with microbumps mounted thereon, a flip-chip ball grid array substrate, or a semiconductor package with underfill applied.

[0110] The target recognition unit aligns the package on the inspection stage with the reference coordinate system using the reference mark, outer shape, CAD pattern data, or substrate coordinate information of the semiconductor package. The alignment tolerance can be set, for example, in the range of 0.01 mm to 0.05 mm, i.e., to a level of 10 micrometers to 50 micrometers.

[0111] The robotic inspection unit performs a primary high-speed inspection using a transmission radiation modality, and the defect analysis unit calculates candidate areas for microbump voids, underfill voids, bump cracks, interfacial delamination, or foreign matter.

[0112] In cases where the transmission of terahertz signals is restricted by a metal layer or silicon core, the terahertz electromagnetic wave modality may be selected as an auxiliary modality for defect candidates in dielectric or nonmetallic regions such as EMC, underfill, adhesive layer, coating layer, or nonmetallic protective layer.

[0113] The modality adapter unit controls the incident angle, polarization conditions, separation distance, and sampling timing of the terahertz antenna, and can select a frequency band in the range of, for example, 0.1 THz to 3 THz or 0.1 THz to 10 THz.

[0114] The trajectory reconstruction unit adjusts the subsequent inspection incidence angle for the uncertainty region to a range of, for example, 1 to 3 degrees, and the defect map generation unit fuses the radiation inspection data with the terahertz reflection or time domain data to generate a defect coordinate map corresponding to the package coordinate system.

[0115] The judgment and process control linkage unit can transmit process correction signals to a PLC or SCADA linked to a die bonder, underfill dispensing device, epoxy dispensing device, reflow equipment, or thermal compression equipment based on the defect type and defect location. Accordingly, at least one of the dispensing pressure, dispensing time, bonding head position, thermal pressurization conditions, or cooling conditions can be updated.

[0116] In the above embodiment, for interface defect candidates that appear ambiguous in the radiation inspection, a dielectric layer or a non-metallic protective layer to which a terahertz modality can be applied is selectively re-inspected. Accordingly, a candidate region that had low confidence in one modality can be combined with the information acquisition weight of another modality to non-linearly increase the judgment confidence.

[0118] Example 3: Application of seismic wave and active thermal response modality for internal interlayer delamination inspection in composites

[0119] In this embodiment, the test subject may be a carbon fiber reinforced plastic part, an aerospace composite part, an automotive composite structure, a laminated structure including an adhesive layer, or a non-metallic heterogeneous material composite.

[0120] The robot inspection unit can select an active thermal response modality as the primary inspection condition. The modality adapter unit applies thermal stimulation to the surface of the object under inspection using a flash lamp, laser, induction heating source, or hot air stimulation source, and the infrared thermal imaging detector unit collects surface temperature attenuation, phase delay, or thermal diffusion patterns over time.

[0121] The defect analysis unit identifies areas showing a difference exceeding a preset threshold compared to the thermal decay time constant or phase delay of the normal area as candidate areas for subsurface delamination or adhesion failure.

[0122] If the uncertainty index for the above candidate region exceeds a preset threshold, the trajectory reconstruction unit may select an elastic wave modality as a subsequent inspection modality. In this case, the modality adapter unit selects the frequency of the phased array ultrasonic transducer from a range of, for example, 0.5 MHz to 20 MHz, and aligns the transducer in the direction of the surface normal according to the force / torque sensor feedback of the robot end.

[0123] The modality adapter unit supplies an ultrasonic coupling agent to the inspection surface and can maintain a contact pressure in a range of, for example, 5 N to 50 N. The trajectory reconstruction unit can perform a local re-inspection by adjusting the subsequent ultrasonic irradiation angle in a range of, for example, 1 degree to 5 degrees.

[0124] The defect map generation unit fuses primary thermal response data and subsequent ultrasonic echo data to generate a defect coordinate map including the location, depth, area, and reliability of interlayer delamination.

[0125] The judgment and process control linkage unit can transmit a process correction signal to a SCADA or PLC linked to an autoclave, a thermal compression facility, an adhesive application device, or a curing process facility based on the above defect coordinate map. Accordingly, at least one of the temperature profile, pressure profile, curing time, adhesive application amount, or cooling conditions can be updated.

[0126] In the above embodiment, an active thermal response modality extracts uncertainty candidates over a wide area, and an elastic wave modality reinforces depth direction information of the corresponding candidate region. As such, inspection modalities having different physical principles are sequentially combined via an uncertainty field, thereby non-linearly improving the depth resolution and reliability of the defect coordinate map compared to when only a single inspection modality is used.

[0128] Example 4: Application of Radiation Modality for Internal Shrinkage Cavity Inspection in Large Castings

[0129] In this embodiment, the test object may be an engine block for automobiles, a large aluminum casting, a magnesium casting, a gigacasting structure, a reduction gear housing, or a metal part having an internal hollow structure.

[0130] The object recognition unit establishes a reference coordinate system of the object under inspection using a barcode, RFID, external scan data, or CAD model. The alignment tolerance can be set, for example, in the range of 0.1 mm to 0.5 mm.

[0131] The robotic inspection unit can perform a primary high-speed scan by selecting a transmission radiation modality. Under primary inspection conditions, the tube voltage of the radiation source can be set in the range of, for example, 40 kV to 160 kV, and if the thickness and density of the object under inspection are high, a range of 160 kV to 450 kV or a separate high-energy radiation source may be applied. The shooting angle pitch can be set in the range of, for example, 20 degrees to 30 degrees.

[0132] The defect analysis unit calculates internal pores, shrinkage pores, cracks, or foreign matter candidate areas, and if the uncertainty index of said area exceeds a preset threshold, the trajectory reconstruction unit can reduce the subsequent inspection angle pitch to a range of, for example, 1 to 2 degrees.

[0133] The modality adapter section performs re-inspection by variably controlling the opposing alignment distance between the radiation source and the detector, the collimator aperture ratio, the irradiation angle, or the irradiation energy conditions.

[0134] The defect map generation unit generates an internal defect voxel map corresponding to the casting CAD coordinate system. The judgment and process control linkage unit can send a control signal to a line classification gate or a rejecter if the total volume, distribution density, or location of defects exceeds quality standards.

[0135] In addition, the judgment and process control linkage unit can transmit a process correction signal to the MES, SCADA, or PLC of the casting process so that at least one of the molten metal injection pressure, plunger speed, mold temperature, cooling water flow rate, cooling timing, or pressurization holding time is updated.

[0136] In the above embodiment, when the boundary of an internal deep defect candidate appears ambiguous in the first high-speed scan, the trajectory reconstruction unit resets the direction of subsequent inspection by simultaneously considering collision constraints and expected information gain. Accordingly, the coordinate reliability of the internal shrinkage cavity can be increased with fewer subsequent inspection resources than simply re-photographing from all angles, and the increased coordinate reliability can be linked to the correction of pressure, cooling, and injection conditions of the casting process.

Claims

Claim 1 A target recognition unit that acquires external shape information or a reference model of an object to be inspected and establishes a coordinate system of said object to be inspected; a robot inspection unit configured to variably drive multiple axes with respect to said object to be inspected or received non-destructive inspection signals at each point in time; a modality adapter unit that variably sets at least one of the end device alignment distance, contact pressure, irradiation angle, detector placement, irradiation energy, pulse width, sampling timing, or coupling condition of said robot inspection unit according to the type of selected non-destructive inspection modality; a defect analysis unit that calculates defect candidate regions and uncertainty regions inside or below the surface of said object to be inspected from primary inspection data; a trajectory reconstruction unit that recalculates subsequent inspection positions, joint angles, and inspection conditions based on a cost function including expected information gain for said uncertainty region and collision constraints, driving torque constraints, inspection time constraints, and signal quality constraints per modality of said robot inspection unit; and a defect map generation unit that fuses re-inspection data acquired by said subsequent inspection with said primary inspection data to generate a defect coordinate map corresponding to the coordinate system of said object to be inspected. An uncertainty information gain-based modality adaptive robot non-destructive inspection system comprising a judgment and process control linkage unit that outputs a quality judgment signal or a process correction signal based on the defect coordinate map and links with at least one of an external manufacturing execution system, a monitoring and control system, or a programmable logic controller to update control parameters, set points, or actuator driving conditions of a full-process manufacturing facility. Claim 2 A robot non-destructive testing system according to claim 1, wherein the non-destructive testing modality comprises at least one of a transmission radiation modality, an elastic wave modality, a terahertz electromagnetic wave modality, and an active thermal response modality, and wherein the modality adapter unit physically variably controls the driving conditions of the end device or sensor of the robot inspection unit according to the selected non-destructive testing modality. Claim 3 A robotic non-destructive inspection system according to claim 1, wherein the defect analysis unit calculates the uncertainty index U(x) at each spatial coordinate x by the following mathematical formula. [Mathematical Formula 1] U(x) = - ∫ p(f(x)|D) log p(f(x)|D) df In the above formula, x is a spatial coordinate vector inside or below the surface of the object under inspection, D is a primary inspection dataset, and p(f(x)|D) is a probability density function representing the physical state, density, possibility of defect existence, or signal characteristics of the corresponding coordinate when the primary inspection dataset is given. Claim 4 A robotic non-destructive inspection system according to claim 3, wherein the defect analysis unit calculates the uncertainty index U(x) by the following mathematical formula when the defect type is classified into discrete class k. [Mathematical Formula 2] U(x) = - Σ(k=1 to K) pk(x) log pk(x) A robotic non-destructive inspection system characterized in that, in the above formula, K is the number of defect type classes and pk(x) is the probability that the corresponding coordinate belongs to the k-th defect type. Claim 5 A robot non-destructive inspection system according to claim 1, wherein the trajectory reconstruction unit calculates the robot's joint angle vector θ and modality driving parameter e using an objective function J(θ,e) according to the following mathematical formula. [Mathematical Formula 3] J(θ,e) = IG(θ,e) - λ1Ccollision - λ2Cexposure - λ3Ctime - λ4Ctorque - λ5Csignal In the above formula, IG(θ,e) is an expected information gain function, Ccollision is a collision constraint cost, Cexposure is a signal irradiation or contact exposure cost according to the selected modality, Ctime is an inspection time cost, Ctorque is a robot joint torque cost, Csignal is an effective signal-to-noise ratio constraint cost, λ1 to λ5 are normalized weights, and the robot non-destructive inspection system is characterized in that the expected information gain function IG(θ,e) is calculated by the following mathematical formula. Claim 6 A robotic non-destructive inspection system according to claim 2, wherein when the above-mentioned transmissive radiation modality is selected, the modality adapter part is characterized by variably controlling at least one of the opposing alignment distance between the radiation source and the sensor detector, the focus-detector distance, the irradiation angle, the irradiation energy, or the shielding collimator aperture ratio. Claim 7 A robot non-destructive inspection system according to claim 2, wherein when the elastic wave modality is selected, the modality adapter unit controls the physical contact pressure between the probe at the end of the robot inspection unit and the surface of the object to be inspected in real-time feedback control, and variably controls at least one of the acoustic medium coupling supply amount, probe normal alignment state, probe frequency, or ultrasonic irradiation angle. Claim 8 A robotic non-destructive inspection system according to claim 2, wherein when the terahertz electromagnetic wave modality is selected, the modality adapter unit variably controls at least one of the incident angle of the terahertz signal, polarization conditions, time domain sampling timing, reflective inspection mode, transmissive inspection mode, or separation distance from the surface of the object to be inspected, and when the active thermal response modality is selected, the modality adapter unit variably controls at least one of the stimulation intensity of the thermal stimulation source, stimulation time, stimulation position, thermal image detection timing, or thermal response sampling conditions. Claim 9 A robot non-destructive inspection system according to claim 1, wherein the defect coordinate map includes at least one of defect center coordinates, defect volume, defect area, defect length, defect depth, defect type label, defect reliability, defect distribution density, and manufacturing lot information, and the judgment and process control linkage unit transmits a process correction command to at least one of a manufacturing execution system, a monitoring and control system, or a programmable logic controller via an industrial communication protocol interface when a defect factor derived from the defect coordinate map exceeds a reference threshold quality value, and wherein at least one of the pressure, tension, temperature, position, speed, time, flow rate, voltage, current, torque, or actuator setpoint of the front-end manufacturing equipment is updated based on the process correction command. Claim 10 A method for generating a defect coordinate map and controlling a process using a robotic non-destructive inspection system according to any one of claims 1 to 9, comprising: a step of establishing a coordinate system for an object to be inspected by acquiring external shape information or a reference model of the object to be inspected; a step of acquiring primary inspection data using a non-destructive inspection modality; a step of calculating a defect candidate region and an uncertainty region from the primary inspection data; a step of recalculating a subsequent inspection position, joint angle, or inspection condition based on a cost function including expected information gain for the uncertainty region, robot driving constraints, and physical constraints per modality; a step of acquiring re-inspection data according to the recalculated subsequent inspection position, joint angle, or inspection condition; and a step of generating a defect coordinate map corresponding to the coordinate system for the object to be inspected by fusing the primary inspection data and the re-inspection data. A method for generating a defect coordinate map and controlling a process, comprising the step of generating a quality judgment signal or a process correction signal based on the defect coordinate map, and transmitting the process correction signal to at least one of a manufacturing execution system, a monitoring and control system, or a programmable logic controller to update at least one of a control parameter, a set point, or an actuator driving condition of a manufacturing facility.

Images