Machine vision-based shock absorber connecting rod surface defect online detection system

Abstract

The application discloses a machine vision-based shock absorber connecting rod surface defect online detection system and relates to the technical field of surface detection, and comprises the following contents: firstly, the high-reflective piston rod after cleaning is subjected to the same field double-condition line scanning collection and the collection quality mark is established; then, the actual surface phase is calculated according to the nominal motion data and the phase reference original data, the double-condition line data is corrected line by line, the expansion image under the unified coordinate, the pollution state diagram and the phase position reliability are generated; subsequently, the pollution gate defect extraction is carried out according to the pollution state diagram, and the budget arbitration is performed on the uncertain area; finally, the defect position, size, severity and defect digital map are output, and the threshold value and compensation parameter are written back. The scheme can stably distinguish the pollution, mirror surface reflection and real micro-defects, improve the cylindrical surface expansion consistency and online detection continuity, reduce the false positives and false negatives, and give consideration to the detection beat and batch adaptability.

Description

Technical Field

[0001] This invention relates to the field of surface inspection technology, specifically to an online inspection system for surface defects on shock absorber connecting rods based on machine vision. Background Technology

[0002] This equipment is primarily used for the electroplating, cleaning, and assembly of rod-shaped metal parts such as piston rods and shock absorber bars in automotive shock absorbers. It involves full-circumferential inspection of the outer cylindrical surface after chrome plating, cleaning, and before assembly. This station requires continuous workpiece flow with a fixed cycle time for release, rejection, and traceability. Currently, this industry has evolved from manual visual inspection to machine vision, primarily using line-scan cameras in conjunction with workpiece rotation to acquire images of the cylindrical surface, and then using difference, texture, or template matching to identify defects.

[0003] Chinese patent document CN110658203A discloses an optical detection system and method for microscopic defects on the surface of a piston rod. It includes a pair of opposing, coaxially arranged tip mechanisms, a first linear drive mechanism for driving the tip mechanisms to clamp or release the piston rod, a line scan camera for acquiring images of the piston rod surface, a second linear drive mechanism for driving the line scan camera to move axially along the piston rod, a third linear drive mechanism for driving the line scan camera to move radially closer to or away from the piston rod, and a rotation drive mechanism for driving the piston rod to rotate. During operation, the tip mechanisms clamp the piston rod and cause it to rotate, while the line scan camera acquires images of the outer cylindrical surface segment by segment during axial movement, achieving scanning imaging of the piston rod's outer surface. Chinese patent document CN118396980A discloses a shock absorber rod detection method based on multi-scale semantic differences of reference images, and Chinese patent document CN120125577B discloses a piston defect detection method based on full-angle annular images, sliding window texture regularity factors, and template matching.

[0004] However, there are still many shortcomings in the technology for the high-reflectivity chrome-plated piston rod scene after cleaning and before assembly.

[0005] First, the piston rod, after fine grinding, electroplating, and polishing, has a smooth surface that easily forms a specular reflection under strong light. Background technology CN110658203A already indicates that the piston rod has microscopic defects such as chromium pits and pinholes ranging from 10 to 500 micrometers, which can easily overlap with local bright spots and weakly reflective areas under high light. Second, the reference image difference route CN118396980A and the texture rule factor and template matching route CN120125577B are based on the premise of image consistency and comparability. On actual production lines, residual dust and other surface disturbances exist on the workpiece surface after cleaning. Changes in the curved surface reflection state and local grayscale distribution with circumferential position can also cause image differences, which can originate from both actual defects and surface condition changes. Finally, when using encoders or mechanical motion synchronous scanning methods, the unfolded image mainly reflects the nominal motion information of the mechanism. When small slippage, eccentricity, or circular runout exists, the unfolded image may show local misalignment and seam drift. If not controlled, this can easily lead to false alarms, missed alarms, repeated inspections, and cycle time fluctuations.

[0006] Therefore, based on the above, the technical problem that the existing technology needs to solve is: how to improve the stability of distinguishing between contamination, reflection and real micro-defects in the scenario of full-circumference online inspection of highly reflective rods after cleaning and before assembly, and improve the consistency between cylindrical surface development and defect judgment. Summary of the Invention

[0007] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an online detection system for surface defects on shock absorber connecting rods based on machine vision. The system calculates the actual surface phase based on nominal motion data and original phase reference data, performs line-by-line unfolding correction on the dual-conditional line data, and generates an unfolded image, a contamination state map, and phase position confidence level in a unified coordinate system. Subsequently, it extracts contamination-gated defects based on the contamination state map and performs budgeted arbitration on uncertain regions. Finally, it outputs the defect location, size, severity, and a digital map of defects, and writes back the threshold and compensation parameters. This system stably distinguishes between contamination, specular reflection, and real micro-defects, improves the consistency of cylindrical surface unfolding and the continuity of online detection, and balances detection cycle time and batch adaptability; thus solving the technical problems described in the background art.

[0008] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: The machine vision-based online detection system for surface defects of shock absorber connecting rods includes: in response to the cleaned workpiece entering the detection position, performing line scanning on the same surface area under first and second optical conditions, and simultaneously acquiring nominal motion data and phase reference raw data. The actual surface phase is obtained based on the nominal motion data and the original phase reference data. The acquired dual-condition line data is then expanded and corrected line by line based on the actual surface phase to obtain the first and second expanded images under the unified axial-circumferential coordinates. Contamination state map and phase position confidence are constructed based on the first and second unfolded images. The re-blowing, re-shooting, or defect determination is gated based on the contamination state map. The second detection condition arbitration is triggered only for uncertain areas, areas with phase position confidence below the preset value, or preset high-risk areas. The defect location, size, and severity are output based on the main determination result and the arbitration result.

[0009] Furthermore, before online scanning and acquisition, a local pre-liquid removal treatment is performed on the surface of the workpiece to be measured, and the first condition line image, the second condition line image, nominal motion data and phase reference raw data are acquired synchronously under unified hardware triggering. These are then bound with the workpiece identification, timestamp and acquisition quality mark to form a raw co-domain data packet. The raw co-domain data packet serves as the input object for subsequent determination of the actual surface phase, generation of unfolded image and construction of contamination state map.

[0010] Furthermore, the acquisition quality is judged on the original intra-domain data packets. When any of the following situations are detected, such as the saturated pixel ratio exceeding the preset upper limit, the loss or continuous distortion of the phase reference original data, the continuous appearance of local high humidity areas, or the time alignment error of the dual-condition line image exceeding the preset value, local re-blowing and local re-shooting are performed on the corresponding spatial window, and a re-sampling version number is written to the re-sampling result. The re-sampling version number increases with the number of re-sampling times.

[0011] Furthermore, the nominal phase is obtained based on the nominal motion data, and the phase correction is obtained by combining the reference center position and the reference position in the original phase reference data; When the system is configured with a radial displacement signal, the nominal phase is corrected by combining the radial displacement and the reference radial displacement to obtain the actual surface phase corresponding to each line image. The actual surface phase is used as the writing reference for the unified circumferential coordinates.

[0012] Furthermore, based on the actual surface phase and the axial position in the nominal motion data, each first condition line image and second condition line image is written into a unified axial-circumferential grid to generate a first unfolded image and a second unfolded image; shape-preserving piecewise cubic interpolation is used for sections with sufficient samples, and bilinear interpolation is used for sections with insufficient boundary samples to complete line-by-line unfolding correction.

[0013] Furthermore, based on the local response difference between the first and second unfolded images, the response change of the second unfolded image before and after local reblowing, and the continuous length of the abnormal connected domain along the principal axis, a pollution state map under a unified coordinate system is constructed, and the pollution state map is kept in positional correspondence with the first and second unfolded images.

[0014] Furthermore, based on the same-domain misalignment, phase continuity index, and acquisition saturation ratio, a phase position confidence level corresponding to the unified coordinates is generated, and the phase position confidence level and the contamination status map are synchronously written into the unified expansion cache; wherein, the phase position confidence level is used to limit the segments that can enter the defect candidate extraction path, and serves as the input for subsequent arbitration triggering conditions.

[0015] Furthermore, in areas where the contamination status map is below the first threshold, defect candidate values ​​are generated based on the linear continuous response of the second unfolded image, the mirror interruption response of the first unfolded image, and the pit-like edge response of the first and second unfolded images. The defect candidate values ​​are then mapped one-to-one with uniform coordinates and used as the basic input for the main decision path.

[0016] Furthermore, the maintenance state is maintained based on the confirmation threshold, release threshold, and phase lower limit threshold; when the defect candidate value reaches the confirmation threshold and the phase position confidence reaches the phase lower limit threshold, the defect candidate segment is confirmed; when the contamination state map is higher than the second threshold, the maintenance state is released and local re-blowing and local re-shooting are triggered for the corresponding segment.

[0017] Furthermore, when a segment is in a threshold zone, the phase position confidence is lower than a preset value, it is located in a preset high-risk area, or the main judgment state is inconsistent before and after reblowing, an arbitration request package is generated and sent to the arbitration request queue; the arbitration request package includes at least the workpiece identification, axial boundary, circumferential boundary, triggering reason and request mode.

[0018] Furthermore, the arbitration request queue is configured with an arbitration trigger rate cap, a queue warning threshold, and a queue hard threshold. When the queue reaches the queue warning threshold, newly entering arbitration request packets are downgraded to fast second-view re-judgment. When the queue reaches the queue hard threshold, new 3D re-judgment is stopped and new request packets are transferred to bypass re-examination requests.

[0019] Furthermore, the abnormal segments corresponding to the maintained state are spatially fused with the abnormal windows in the arbitration result record according to the unified axial-circumferential coordinates to form a defect segment number. The severity value of each defect segment number is generated based on the average defect candidate value of the segment, the arbitration measurement, the preset high-risk area marker, the axial length, and the circumferential span.

[0020] Furthermore, a defect digital map is generated based on the axial range, circumferential range, severity value, and evidence index of each defect segment number, and release instructions, rejection instructions, or bypass re-inspection instructions are output according to the handling status mark; the confirmed results are written back to the contamination baseline, phase compensation offset, and arbitration budget baseline, and take effect from the next workpiece or the next batch of workpieces.

[0021] (III) Beneficial Effects This invention provides an online detection system for surface defects of shock absorber connecting rods based on machine vision, which has the following advantages: The system employs dual-condition line scan acquisition within the same domain, pre-liquid removal processing, and acquisition quality archiving. This locks the first-condition image stream, the second-condition image stream, nominal motion data, and the original phase reference data onto the same workpiece, the same surface region, and the same acquisition cycle, preventing input object drift and providing a unified data foundation for subsequent unified unfolding and state determination. The first and second unfolded images are corrected line-by-line using the actual surface phase instead of the nominal angle, unifying axial-circumferential coordinates and suppressing suture drift, cross-channel mismatch, and abnormal thread breakage caused by micro-slippage, circular runout, and local velocity fluctuations. This ensures that subsequent determinations always point to the actual surface position.

[0022] A contamination state map is created based on the first unfolded image, the second unfolded image, the second unfolded image after reblowing, and the local continuous length. Residual liquid, watermark, oil film, and dust are first placed in the pending state before entering the subsequent gated path. Contamination identification and defect identification are unfolded in layers to enhance the stability of anomaly attribution under wet conditions.

[0023] Defect candidates are extracted only in areas where the pollution state map is below the first threshold. The confirmation threshold, release threshold, and phase lower limit threshold remain unchanged. Meanwhile, uncertain areas, low phase position confidence areas, high-risk areas, and re-blowing inconsistency areas are sent to budgeted arbitration to achieve mutual complementarity between the main decision path and the verification path.

[0024] The main judgment result and the arbitration result are integrated into a defect digital map and severity value. The confirmed results are written back to the contamination baseline, phase compensation bias and arbitration budget baseline, so as to realize single-item release, rejection and bypass re-inspection, while enhancing the adaptability in batch switching and continuous operation scenarios. Attached Figure Description

[0025] Figure 1 This is a diagram showing the overall architecture of the detection system of the present invention; Figure 2 This is a schematic diagram of the structural arrangement of the detection positions in this invention; Figure 3 This is a flowchart illustrating the formation of the original intra-domain data packet in this invention. Figure 4 This is a schematic diagram illustrating the unified deployment and pollution state generation of the present invention; Figure 5 This is a schematic diagram of the pollution gating, defect extraction, and arbitration chain of the present invention; Figure 6 This is a schematic diagram of the defect digital map and parameter write-back closed loop of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figures 1-6 This invention provides an online detection system for surface defects of vibration damper connecting rods based on machine vision, including: Step 1, without pre-drying the rod to be tested completely, firstly, the dual-condition line image of the same surface area and the surface motion reference signal are stably bound into an original co-domain data packet, and then the original co-domain data packet is attached to a unique workpiece identification and acquisition quality mark, so that subsequent steps can distinguish wet film interference, specular reflection and real surface anomalies under the same coordinates, the same object and the same time sequence.

[0028] This step is executed in a coordinated manner by the detection controller. The pre-drying unit, main imaging mechanism, motion mechanism, and phase reference mechanism each perform different but interlocking field actions. The pre-drying unit does not aim to dry the test rod to absolute dryness, but rather removes large droplets that would obstruct imaging and compresses irregular drips into narrow strips of wet film or discrete water droplets that can be identified in subsequent steps. The main imaging mechanism does not take just one image, but forms a first optical condition line image and a second optical condition line image on the same surface area. The first optical condition line image retains more specular components, while the second optical condition line image suppresses specular components, making the appearance of the same local surface comparable in the two image streams. The motion mechanism does not directly provide the true position of the surface; it only provides a nominal motion signal. Therefore, the phase reference mechanism needs to supplement it with the original phase reference signal. The detection controller encapsulates these pieces of information on the same cycle to form the original co-domain data packet that can be used in the next step.

[0029] When the cleaned rod enters the testing position from the cleaning and conveying section, the most prominent problem is not whether there is any abnormality on the surface, but that the surface state has not yet stabilized: local residual liquid spreads along the axial or circumferential direction, and specular reflection will form bright bands at different circumferential positions. If only a single optical condition is used for image acquisition at this time, the subsequent steps will not obtain the surface body information, but the superposition of the surface body information and the transient liquid film information at the station.

[0030] Furthermore, if the dual-condition line images are not obtained within the same surface area, but rather exhibit temporal, angular, or axial misalignment, then subsequent steps in creating the contamination state map will separate the two representation objects that should correspond to the same location. After reading the workpiece identification, the detection controller first sends an entry shaping command to the pre-drying unit, then a line scan start command to the motion mechanism, and finally a dual-condition synchronization trigger command to the main imaging mechanism.

[0031] If image acquisition precedes liquid removal, the main imaging mechanism will initially perceive random sludge and large droplets. Furthermore, if the test rod is rotated at high speed before dual-condition synchronization is activated, the first and second optical condition line images are more likely to lose their co-domain relationship. Following the logic of this step, the test rod should first enter a dimensionally and orientationally stable entry window, then the dual-condition line scan window, and finally the phase reference window. Since the main imaging mechanism and the phase reference mechanism are fixed on the same frame, and the nominal motion signal of the motion mechanism is acquired by the same detection controller, this step forms a single-chain motion transmission relationship, rather than a stacking of multiple parallel and isolated actions.

[0032] In one embodiment, the pre-liquid removal unit is located upstream of the detection inlet. The pre-liquid removal unit includes a set of slit air knives, a set of negative pressure suction ports, and a liquid guide shroud. The long axis of the slit air knife nozzle is parallel to the axial direction of the rod under test. The nozzle slit width is preferably one-twentieth to one-fiftieth of the maximum diameter of the rod under test. The distance from the nozzle to the outer surface of the rod under test is preferably one to three times the maximum diameter of the rod under test. The air source for the air knife is filtered compressed air. The liquid guide shroud uses a stainless steel plate with a quartz observation window. The negative pressure suction ports are located downstream of the air knife to remove large droplets and water mist that have been stripped away from the line of sight of the main imaging mechanism. The purpose of this arrangement is not to blow the rod under test into a dry rod, but to confine large, random liquid layers into a locally wet film with a limited distribution and visible boundaries.

[0033] Taking a test rod delivered by a cleaning machine as an example, before the test rod enters the detection position, there are three types of visible liquid forms on its outer surface: large droplets converging along the lower side, oblique flow marks formed by the backflow from the cleaning nozzle, and a thin wet film adhering to the chrome-plated surface. After the pre-removal unit operates, the first type of large droplets is peeled off by the air knife, the second type of oblique flow marks are cut off and drawn away by negative pressure, and the third type of thin wet film is not forcibly eliminated, but is confined to a narrower, more continuous surface band. Thus, the main imaging mechanism subsequently sees not arbitrarily shaped liquid clumps, but rather wet films or discrete water droplets that are closer to boundaries, which lays the foundation for subsequent steps to identify the dominant contamination area.

[0034] In parallel embodiments, the slit air knife can be replaced with an annular microporous air sleeve, the negative pressure suction port can be replaced with an adsorption cavity surrounding the circumference of the rod to be tested, and the liquid guide cover can be replaced with a transparent anti-fog cover. If the surface of the material of the rod to be tested is highly hydrophilic, it is preferable to shorten the action time of the air knife and increase the suction path length. If the surface of the material of the rod to be tested is highly hydrophobic, it is preferable to reduce the distance between the nozzle and the rod to be tested to avoid the droplets being stretched into long tails and re-attached to the downstream surface.

[0035] In use, the surface state faced by the main imaging mechanism before operation is focused to a relatively stable wet film boundary state, eliminating the need to deal with large-area random drips in subsequent steps. Furthermore, the pre-dehydration unit only changes the liquid state and does not directly alter the texture of the chrome-plated surface, thus preventing premature masking of actual surface anomalies.

[0036] The main imaging mechanism is mounted on the same high-rigidity crossbeam, the length of which covers the detection section of the rod to be tested. The crossbeam is preferably made of aluminum alloy and fitted with a steel mounting base. A quartz protective window is installed in front of the line scanning lens.

[0037] The imaging branch corresponding to the first optical condition is used to preserve the specular component, and the imaging branch corresponding to the second optical condition is used to suppress the specular component. In one embodiment, the first optical condition uses a line light source arranged along the specular reflection direction, and the second optical condition uses a cross-polarized line light source. In another embodiment, the first optical condition uses visible light illumination, and the second optical condition uses near-infrared illumination. The two imaging branches share the same hardware trigger reference, and the detection controller drives the two branches to expose synchronously at a fixed line frequency, so that each first optical condition line image and each second optical condition line image correspond to the same sampling sequence number under the same workpiece identification.

[0038] Furthermore, in order to translate the requirement of co-domain into a verifiable field rule, the detection controller introduces a co-domain misalignment factor for dual-condition image acquisition. : In the formula, the first sampling axis coordinate The first optical condition line image corresponds to the axial position, with a value range from the start point to the end point of the detection segment of the rod under test; the second sampling axial coordinate. The axial position corresponding to the second optical condition line image, with a value range relative to the first sampling axial coordinate. Consistent; Axial mapping pitch The actual length corresponding to each pixel along the axis, with values ​​greater than 0 and less than the minimum discriminative length of the defect to be tested; first sampling angular coordinates. The nominal circumferential angle position corresponding to the first optical condition line image, with a value ranging from 0 to 2π; the second sampling angular coordinate. The nominal circumferential angle position corresponding to the second optical condition line image, with a value ranging from 0 to 2π; circumferential mapping pitch. The angle corresponding to each pixel in the circumferential direction, with values ​​greater than 0 and less than the single-row circumferential resolution angle; intra-domain misalignment. The value range is from zero to positive infinity, and it is used to determine whether two line images still belong to the same surface region.

[0039] When the same domain misalignment When the misalignment is below the preset maximum in-domain misalignment, the detection controller encapsulates the paired line image into in-domain double-condition line data; when the in-domain misalignment is below the preset maximum in-domain misalignment, the detection controller encapsulates the paired line image into in-domain double-condition line data. When the misalignment exceeds the preset maximum in-domain misalignment, the detection controller does not directly allow the image of the line pair to pass through, but instead records it as an abnormal acquisition quality. In other words, the in-domain misalignment... It's not for compensation in subsequent steps, but rather to keep out dual-condition image acquisition that doesn't meet the same domain condition in step one.

[0040] In practice, the first and second optical condition line images already possess co-domain constraints before proceeding to subsequent steps, eliminating the need for extensive blind registration in later steps. Furthermore, the co-domain misalignment is... Moving it to step one helps to expose the effects of hardware installation errors, trigger timing differences, and axial jitter during the acquisition phase, rather than letting the sources of anomalies be introduced only during the contamination state map calculation phase.

[0041] Furthermore, during controlled rotation of the rod under test, minute slippage at the clamping end, slight circular runout of the rod body, elastic clearance of the support, and changes in friction conditions caused by localized wet film all contribute to deviations between the angle the mechanism rotates and the actual angle the surface moves. Therefore, it is crucial to acquire the original phase reference signal in parallel, and to simultaneously complete the acquisition quality documentation and anomaly diversion. Otherwise, the next step of determining the actual surface phase will lack reliable input, and it will be impossible to determine which inputs should be rejected or re-acquired.

[0042] Upon receiving the co-domain dual-condition line data, the detection controller does not immediately begin the expansion calculation. Instead, it waits for the original phase reference signal corresponding to the sampling sequence number from the phase reference mechanism to arrive before timestamping the nominal motion signal, the original phase reference signal, and the dual-condition line data. The detection controller only calculates the acquisition quality marker when all three types of data share the same workpiece identification and the same sampling sequence number. The acquisition quality marker is not a single switch but a set of hierarchical labels established around saturation, co-domain, phase continuity, and wet coverage status. After this processing, the next step obtains not only the image and motion reference but also the boundary information determining whether this batch of inputs is suitable for participating in the actual surface phase determination.

[0043] The phase reference mechanism is installed downstream of the main imaging mechanism, and the two are fixed on the same crossbeam or the same substrate to ensure that the mechanical relative position remains unchanged.

[0044] In one embodiment, the phase reference mechanism includes a stripe projector and a phase reference camera. The stripe projector projects narrow reference stripes onto the surface of the rod under test, with the stripes extending axially. The phase reference camera observes the center position and brightness envelope of the stripes from a fixed side. Thus, when the rod under test rotates circumferentially or experiences slight jerking, the center position and brightness envelope of the stripes undergo measurable changes, which constitute the original phase reference signal.

[0045] If space is limited, the phase reference camera can be replaced by a combination of a displacement sensor and a reflected light spot. As long as the output is still a reference quantity that changes synchronously with the actual movement of the surface, the terminology remains unchanged.

[0046] To determine whether the original phase reference signal is continuous, the detection controller introduces a phase continuity index. : In the formula, the phase continuity index The phase reference original signal is in continuous The stability of the line, ranging from 0 to 1; the number of consecutive lines : The number of adjacent samples participating in the continuous calculation, with values ​​ranging from integers greater than or equal to 2; reference center coordinates : No. The center position of the reference fringe corresponding to the line is within the effective pixel range of the phase reference camera; reference amplitude. : No. The reference fringe brightness envelope amplitude corresponding to the stripe, ranging from 0 to 1, is used to characterize whether the stripe is disrupted by water mist, droplets, or local shadows; positional release constant. Reference center coordinates Allowable natural fluctuation scale, with values ​​ranging from the number of pixels greater than 0; amplitude buffering constant. Reference amplitude The permissible natural fluctuation scale, with values ​​greater than 0 and less than 1, is used to limit the impact of sudden amplitude drops on the phase continuity index. The intensity of inhibition.

[0047] Phase continuity index A value close to 1 indicates that the fringe center position and fringe brightness envelope remain smooth across multiple consecutive sampling lines; the phase continuity index... A drop indicates that the surface reference information has been obscured by water droplets, the light spot is distorted, or vibration has disturbed it. The detection controller adjusts the phase continuity index. It is written into the original local data packet as part of the acquisition quality tag, instead of dragging it to the next step for interpretation.

[0048] Furthermore, the phase continuity index Simultaneously sensing both position and amplitude dimensions ensures that states where stripes still exist but are distorted, or where stripe positions are still stable but brightness has severely collapsed, will not be mistaken for reliable inputs that can be used directly.

[0049] In phase continuity index After generation, the detection controller continues to read the saturation statistics of the first optical condition line image and the second optical condition line image, and compares them with the in-domain misalignment. Phase continuity index The acquisition quality marker is written together. First, the detection controller obtains the acquisition saturation ratio by counting saturated pixels. : In the formula, the sampling saturation ratio : The proportion of near-full-frame grayscale area in dual-condition sampling, with a value range of 0 to 1; First saturation count The first optical condition line image contains the number of pixels that reach the saturation threshold, with values ​​ranging from 0 to the total number of pixels in the first optical condition line image; the second saturation count... The number of pixels in the second optical condition line image that reach the saturation threshold, with a value range from 0 to the total number of pixels in the second optical condition line image; total effective pixels. The total number of effective pixels in the two-line image used for quality assessment, with values ​​ranging from positive integers.

[0050] Subsequently, the detection controller generates the acquisition quality index. The overall availability of the current original intra-domain data packet before it enters the next step is preferably within the range of 0 to 1; weighting coefficient Weighting coefficients and weighting coefficients They represent the sampling saturation ratio, respectively. , co-domain misalignment and phase continuity index The proportion in the comprehensive evaluation, with a value ranging from 0 to 1 and satisfying a sum of 1; maximum intra-domain misalignment. : The upper limit of allowed same-domain values, with a range of values ​​greater than 0, used to account for the misalignment of same-domain values. Normalize to a comparable scale.

[0051] The detection controller uses the data acquisition quality index When, if the misalignment quantity in the same region Greater than the maximum co-domain misalignment The second term in the equation is then truncated to zero to prevent severely misaligned data from passing through the high-phase continuity exponent. It was incorrectly retained.

[0052] Abnormal traffic diversion on-site actions follow a fixed link: when the acquisition quality index When the threshold is exceeded, the detection controller encapsulates the dual-condition line image, nominal motion signal, phase reference original signal, and acquisition quality marker corresponding to the current sampling sequence number into a raw in-domain data packet; when the acquisition quality index... When the image falls within the repeat shooting threshold range, the detection controller first commands the pre-cleaning unit to perform local re-blowing, and then commands the main imaging mechanism to repeat shooting within the same sampling window; when the acquisition quality index of several consecutive sampling sequences... When all values ​​are below the rejection threshold, the detection controller will not repeat the re-blowing process, but will instead divert the test member to the bypass re-inspection channel. From the perspective of on-site operation, the operator can intuitively see that a test member briefly pauses when passing through the main imaging window, undergoes a partial re-blowing, and then re-crosses the line; if high humidity occlusion or reference stripe distortion continuously occurs on the same surface, the test member will no longer repeatedly cycle within the main line, but will directly enter the bypass re-inspection.

[0053] In practice, the original intra-domain data packets are not a bunch of unfiltered raw inputs, but rather a usable input that has already undergone inbound quality stratification. Furthermore, anomaly triage relies on the acquisition quality index. Execution ensures that problems in the data collection phase are closed at that phase, preventing unusable inputs from being passed to the next step.

[0054] In one implementation, the test member is a cleaned chrome-plated vibration damper connecting rod. The pre-drying unit uses a slit air knife with a negative pressure suction port. The main imaging mechanism adopts a dual-branch scanning arrangement on a shared frame. The first optical condition is linear illumination that retains the specular component, and the second optical condition is cross-polarized line illumination. The phase reference mechanism uses a stripe projector and a phase reference camera. The detection controller uses an industrial computer in conjunction with an image acquisition card. All raw signals are timestamped and bound under the same industrial bus clock. With this arrangement, the action chain visible on-site is continuous and clear: test member entry, localized dehydration, dual-condition synchronous image acquisition, stripe reference acquisition, quality mark generation, release or re-shooting or bypass diversion. There are no issues of unclear subjects or unclear sequence relationships between actions.

[0055] In the parallel extension approach, the imaging realization of the first and second optical conditions is not limited to the combination of cross-polarization and specular-sensitive illumination; it can also employ a dual-branch system of visible and near-infrared light, or a single-lens beam splitting path to form two band branches. The phase reference mechanism is not limited to a combination of a stripe projector and a phase reference camera; it can also employ a spot displacement scheme, or a reflection spot tracking scheme coupled with a radial displacement sensor. The pre-removal unit is not limited to an air knife; it can also employ a combination of narrow-slit hot air and negative pressure, as long as its function remains to converge large droplets into a wet film with finite boundaries, rather than rewriting the surface bulk information in advance. Through this parallel extension, the inventive concept of step one remains singular: regardless of the specific component used, the purpose is to organize the dual-condition line image, nominal motion signal, and original phase reference signal into a raw co-domain data packet that can be directly invoked in the next step.

[0056] Step 2: Convert the original intra-domain data packets output in Step 1 into a first unfolded image, a second unfolded image, a contamination state diagram, and a phase position confidence under a unified axial-circumferential coordinate system, thereby providing directly executable state inputs for the contamination gating defect extraction and budgeted arbitration in Step 3.

[0057] The detection controller still serves as the unified execution entity, receiving the same-domain dual-condition line data, nominal motion data, phase reference raw data, and acquisition quality markers output from step one.

[0058] Phase acquisition is the process of restoring nominal motion data to nominal phase. Then, the nominal phase is corrected using the original phase reference data. This yields the actual surface phase, which is closer to the true circumferential position of the surface. Coordinate reconstruction, in this context, refers to reconstructing the first and second optical condition line images based on the actual surface phase. Projected onto the same axial-circumferential grid, forming the first unfolded image. Second unfolded image The so-called state attribution is to take the first unfolded image... Second unfolded image Second unfolded image after reblowing The pollution state value is composed of localized continuous spreading characteristics. And combined with the same-domain misalignment amount retained in step one Phase continuity index and sampling saturation ratio Calculate the phase position confidence. .

[0059] Therefore, image processing and state determination are not separated, nor are motion compensation and contamination attribution. Instead, these two lines are mutually constrained: the more stable the coordinate reconstruction, the lower the contamination state value. The closer to the true state of the surface; the more contamination state value The clearer the boundary, the less likely the gating boundary in step three will mistake the wet film for a defect.

[0060] As a supplement, the original intra-domain data packet includes at least: workpiece identification. Sampling sequence number timestamp First condition line data Second condition line data Accumulated encoder pulses Axial position Reference center location radial displacement (Optional) Co-domain misalignment Phase continuity index Saturation ratio Data collection quality index and re-collected version number Among them, the first version number The version number of the re-sampling after partial re-blowing / re-tapping increases sequentially.

[0061] The nominal motion data output in Step 1 only represents the rotational and axial travel states reported by the motion mechanism at the current sampling cycle. However, the test rod is also affected by slight slippage at the clamping end, slight circular runout of the rod body, and changes in surface friction conditions after pre-drying during clamping, rotation, and station transition. Therefore, the actual circumferential position of the surface corresponding to the same line image is not equal to the theoretical position corresponding to the nominal motion data. If the cylindrical surface unfolding diagram is directly constructed using the nominal motion data, although the first and second optical condition line images have satisfied the co-domain constraint in Step 1, seam offset, local stretching, or local compression will still occur during the accumulation of the entire unfolding diagram.

[0062] The detection controller first converts the nominal motion data from step one into nominal phase for each sample number. and axial position Subsequently, the reference position offset and radial displacement offset are extracted from the original phase reference data, and the nominal phase is then corrected using these two types of offsets. , obtained the Actual surface phase of the line Based on this, the detection controller writes the first optical condition line image and the second optical condition line image into the circular buffer one by one, and then calculates the actual surface phase. and axial position The cached pixels are mapped to a uniform axial-circumferential mesh. To avoid edge overshoot during interpolation, the detection controller preferably uses shape-preserving piecewise cubic interpolation; in implementations with limited hardware computing power, bilinear interpolation, i.e., the first unfolded image... Second unfolded image All are based on actual surface phase Driven by.

[0063] The detection controller already possesses the nominal motion data, phase reference raw data, and co-domain misalignment amount under the same sampling sequence number. Phase continuity index and sampling saturation ratio Therefore, to avoid directly treating the instantaneous brightness fluctuations of the reference fringe as an angle correction factor, the detection controller does not use a single brightness threshold, but instead first extracts the reference center position from the raw phase reference data. Then, in an optional implementation, the radial displacement is extracted. Finally, the reference position is obtained through continuous sampling. and reference radial displacement The phase shift is obtained. Where: actual surface phase : No. The true circumferential phase of the surface corresponding to the line takes values ​​from 0 to 2π, which are the equivalent circumferential positions after cyclic expansion; the nominal phase The first step is obtained by converting the nominal motion data output from step one. The nominal circumferential phase of the line, with a value range from 0 to 2π, is the equivalent circumferential position after cyclic expansion. Accumulated encoder pulses Indicates the first The cumulative encoder pulse count corresponding to the sampling time of each line, with a value range of non-negative integers, originates from the nominal motion data output in step one, and its function is to represent the nominal rotational progress of the mechanism. (Number of encoder pulses per revolution) This represents the total number of calibration pulses corresponding to one revolution, with values ​​ranging from positive integers. The values ​​are derived from the factory calibration or field calibration of the motion mechanism, and its function is to accumulate encoder pulses. Converted to nominal phase .

[0064] Position correction factor : The scaling factor for the circumferential phase correction due to the reference center position offset; the value range is a calibration factor greater than 0; reference center position : No. The center position of the reference pattern corresponding to the stripe, with a value range equal to the effective pixel range of the imaging surface of the phase reference mechanism; if the phase reference mechanism uses a stripe projector and a reference camera, then The gray-scale centroid position of the reference stripe within the region of interest of the reference camera, or the center position obtained by template matching; Reference position The center position of the reference pattern recorded under calibration conditions, with a value range relative to the reference center position. Consistent; Reference pitch : Spatial pitch of the phase reference pattern on the imaging plane of the phase reference mechanism, with values ​​ranging from the pixel length greater than 0; Radial correction coefficient : The proportional coefficient for radial displacement change to correct circumferential phase, with values ​​greater than or equal to 0, serving as a calibration coefficient; radial displacement : No. The current radial position of the rod under test corresponding to the line is within the effective measurement range of the phase reference mechanism or displacement sensor; if the system is equipped with a displacement sensor, the radial displacement is... The radial displacement is obtained directly from the sampled values ​​of the displacement sensor; if the system is not equipped with an independent displacement sensor, the radial displacement... This can be obtained from variations in the width of a reference drawing or from dual-view geometric reconstruction; if radial displacement compensation is not used, it should be explicitly stated in the embodiments. This means that the correction option is not enabled. Reference radial displacement Radial position under calibration conditions, value range and radial displacement Consistent; nominal radius : Nominal radius of the member under test, the value range is the design radius range of the member under test of this model, used to convert the radial displacement difference into a circumferential correction ratio; In one implementation, the phase reference raw data is formed by a fringe projector and a phase reference camera, with the fringe width preferably less than one-tenth of the field of view of the phase reference mechanism, in order to maintain the reference center position. The positioning boundary is clear; if a displacement sensor is used instead of a phase reference camera, the detection controller retains the same formula structure, only requiring the reference center position to be changed. Replace with the center position of the reflection spot.

[0065] Taking on-site operation as an example, after a cleaned rod with two thin watermarks is placed in the testing position, the operator can see the reference stripes slightly swaying left and right in the phase reference camera image as the rod rotates. The detection controller does not delete this sway as noise, but instead incorporates it into the actual surface phase. .

[0066] During use, the nominal motion data is continuously corrected by the phase reference original data, and the actual surface phase... No longer equivalent to the nominal rotation state on the motor side; the phase correction simultaneously absorbs the reference center position offset and radial displacement offset, making the subsequent unfolded coordinates less prone to cumulative misalignment due to circular runout and micro-slippage; calibration parameters are used. , , and nominal radius Afterwards, for different models of test rods, only the calibration file needs to be changed without rewriting the step logic.

[0067] The detection controller obtains the actual surface phase Then, the first and second optical condition line images are first written into the circular buffer, and then the axial position is used. As a vertical index, based on the actual surface phase Resampling is performed as a lateral index. Preferably, the detection controller uses shape-preserving piecewise cubic interpolation for local blocks consisting of four adjacent lines, and bilinear interpolation for local blocks with fewer than four boundary lines, thus avoiding overshoot at the boundaries.

[0068] At the same time, the detection controller does not assume that all resampling results are reliable, but instead considers the same-domain misalignment output from step one as reliable. Phase continuity index and sampling saturation ratio Recoupling yields phase position confidence. : Where: Phase position confidence : No. The degree to which a line can be relied upon by subsequent steps after uniform expansion, with a value ranging from 0 to 1; Phase continuity index. The continuity calculated from the original phase reference data in step one has a value between 0 and 1; the same-domain misalignment amount. The degree of intra-domain offset in the dual-condition image acquisition in step one, with a value range from 0 to the upper limit of the positive value; the maximum intra-domain misalignment. This indicates the upper limit of the allowed intra-domain range, with values ​​greater than 0 (the calibration upper limit); sampling saturation ratio. This indicates the saturation level of the dual-condition sampling in step one, with a value range from 0 to 1; the maximum saturation value... This indicates the maximum allowed data collection saturation level of the system, with a value range of greater than 0 and less than 1. In one implementation, the detection controller measures the phase position confidence. Together with the first unfolded image Second unfolded image Line-by-line writing to a unified expanded cache; in another implementation, the detection controller divides the phase position confidence into fixed-length axial segments. After segment averaging, the data is written to a unified expanded buffer to reduce the impact of localized single-line jitter. For example, when a test member passes through the main imaging window, if a narrow strip of wet film still obscures the reference fringe on one segment, the phase continuity index of that segment... It will decrease, thus reducing the confidence level of phase position. Synchronous descent; at this time, the first unfolded image Second unfolded image Although it is still generated, subsequent steps will not treat this segment the same as high-confidence segments.

[0069] When using it, the coordinate system is uniformly expanded from the actual surface phase. Drive, first unfolded image Second unfolded image Possessing the same circumferential reference; phase position confidence Reintroducing the quality information from step one into step two makes the expansion result inherently carry usability markers; the resampling process and phase position reliability The synchronous generation allows step three to determine which sections require more careful processing without having to revisit the original line image.

[0070] After generating the unified unfolded coordinates, the first unfolded image With the second unfolded image Although they fall within the same axial-circumferential grid, this is not enough to directly determine whether the surface anomaly is defect-driven or contamination-driven.

[0071] The same watermark in the first unfolded image This may appear as a highlight band in the second expanded image. The middle part may appear as a low-contrast boundary, while a shallow pit in the first unfolded image Second unfolded image The differences in these areas are different from those in watermarks. If a single difference or a single grayscale threshold is still used, the wet film edge, the oil film smear area, and the real micro-pits are easily confused with each other.

[0072] The detection controller uses the first expanded image in the unified expanded buffer. Second unfolded image Based on this, the dual-conditional response difference is first calculated, and then the second unfolded image after re-blowing is obtained from the re-blowing image obtained after local re-blowing. The re-blowing response difference is calculated, and then the spreading characteristics are extracted by combining this with the local continuous length. The dual-condition response difference reflects the apparent difference at the same location under two conditions: mirror retention and mirror suppression. The re-blowing response difference reflects whether the local surface response has changed before and after re-blowing. The local continuous length reflects whether the anomalous area is closer to spreading along the liquid flow direction or closer to a local isolated depression. After combining the three, the detection controller generates a contamination status value. And based on the confidence level of the phase position The pollution status value Perform confidence modulation to create a state layer that can be directly passed to step three.

[0073] Furthermore, the detection controller does not display the first unfolded image. With the second unfolded image Instead of directly using the absolute grayscale value for the conclusion, a normalized difference term is constructed to ensure that overall brightness drift does not mask local state differences. For poor re-blowing response, the detection controller does not immediately force re-blowing for every pixel; instead, it only performs local re-shooting on the segments marked as locally re-sampling in step one. The second unfolded image after the re-shooting is recorded as the second unfolded image after re-blowing. .

[0074] For locally continuous lengths, the detection controller tracks adjacent abnormal pixels along the axial and circumferential directions in the unified expanded buffer to obtain the main continuous direction length of the connected band where the current pixel is located, where: Where: pollution state value Axial coordinates Circumferential coordinates The anomaly is closer to a pollution-dominated state, and the value range is non-negative; First unfolded image The pixel response obtained under the first optical conditions and after uniform unfolding has a value range corresponding to the grayscale range of the image acquisition bit depth; the second unfolded image This represents the pixel response obtained under the second optical condition after uniform unfolding, with a value range similar to that of the first unfolded image. Consistent; when step three triggers local re-blowing and local re-sampling for a certain local area, step one generates a re-sampling version number for the same workpiece identification and the same spatial window. The original same-domain data packets; Step 2: This re-collected version number The second conditional line data is then subjected to line-by-line unfolding correction again, and the second unfolded image after reblowing is written at the same axial-circumferential coordinate position. .

[0075] Second unfolded image after reblowing This represents the pixel response obtained under the second optical condition after local reblowing and resampling, with a value range similar to the second unfolded image. Consistent; tiny positive numbers This represents the lower limit for preventing the denominator from being zero during normalization; its value range is positive values ​​greater than 0 and much smaller than the principal grayscale level of the image; weighting coefficients. This indicates the difference in the two-conditional response under pollution conditions. The proportion in, with a value between 0 and 1; weighting coefficient This indicates the difference in re-blowing response under contamination conditions. The proportion in, with a value between 0 and 1; weighting coefficient This indicates the continuity of the spread in the pollution state value. The proportion in the value ranges from 0 to 1, and is used to characterize the importance of the liquid's spreading pattern. Local continuous length This represents the length of the abnormal connected region at the current location along the main continuous direction, with a value range of pixel lengths greater than or equal to 0, used to distinguish between strip-shaped wet films and local isolated anomalies. The detection controller first performs threshold segmentation on the normalized difference map to obtain candidate abnormal connected regions; then, it marks each connected region; finally, it calculates the projection length of the connected region on the main axis based on the main axis direction of the connected region, and records this projection length as the local continuous length. .

[0076] Continuous length reference Represents the local continuous length used for normalization The length reference is a preset length with a value range greater than 0, used to map anomaly bands of different sizes to a uniform scale; In one implementation, the detection controller uses a second unfolded image. As a reference before reblowing, the second unfolded image after reblowing. As a reference after reblowing, the second optical condition branch is less sensitive to the specular bright band, making it more conducive to observing whether the wet film has been blown away. For example, if there is a thin water mark extending axially on the outer surface of a test rod, then the first unfolded image... In the second expanded image, a continuous bright band will appear in this area. The middle section displays dark lines with blurred boundaries; the second unfolded image after local reblowing. The dark lines in the image shift backward, at which point the contamination state value... It will rise; if the other area is a shallow pit in the coating, then the second unfolded image after reblowing... It basically remains in its original position, and the local continuous length Shorter, pollution status value It will not rise due to re-blowing.

[0077] During use, the pollution status value It does not rely on a single grayscale or a single difference, but is formed by the combined effects of dual-conditional response difference, re-blowing response difference, and spreading continuity; local re-blowing is confined to locally re-sampling sections, preventing the entire piece from being dragged into repeated cycles; local continuous length After its introduction, the morphological differences between banded liquids and isolated surface defects are written into the same state variable, providing a more explicit gating basis for step three.

[0078] The detection controller does not display the pollution status value. Instead of directly sending it to step three, we first check the phase position reliability corresponding to the same position. When the pollution status value of a certain area High and phase position reliability When the pollution state value is also high, the detection controller outputs it as a high-confidence pollution-dominant region; when the pollution state value is high... The confidence level is in the middle interval and the phase position. When the level is low, the detection controller does not rush to conclusions, but instead outputs a low-confidence region to be judged; when the pollution state value... Low phase position reliability When the contamination level is high, the detection controller outputs a non-contamination-dominated region for step three to extract defect candidates. This process aims to prevent high-contamination state values ​​from being displayed. From error expansion or low contamination state value Cases of failed highlight suppression are indiscriminately sent to the defect extraction stage.

[0079] On an exemplary production line, a detection controller receives multiple consecutively numbered test rods. One of the test rods has a thin, wet film near its end support area, as shown in the first unfolded image. A ring of bright lines appears in the middle, the second unfolded image The signal appears as a discontinuous dark band, and the phase position confidence level at this location is... Simultaneously, the signal decreases. Based on this, the detection controller outputs this section as a low-confidence area to be judged, rather than immediately classifying it as a defective section. Immediately afterwards, a spot-like indentation is found in the middle of another test member, its first unfolded image... Second unfolded image The position is stable under a unified coordinate system; the second unfolded image after reblowing. Almost unchanged, phase position reliability If the level remains high, the detection controller will output that section as a non-contamination-dominated area and hand it over to step three to continue extracting defect candidates. For on-site personnel, what they can see intuitively is that even if there is a difference on the surface, the system does not follow the same path. Instead, it has already distinguished the subsequent paths based on the status and confidence level at the end of step two.

[0080] In the parallel expansion method, if the second optical conditional branch uses near-infrared illumination, then the second unfolded image after reblowing... The same terminology will still be used; if a second-view re-image is used as the re-sampling image source after local re-blowing, the detection controller will first map the second-view image back to a unified axial-circumferential grid through geometric transformation, and then write it into the second unfolded image after re-blowing. If a stripe projector is not available on site, then the phase position confidence level... The phase continuity index can still be retained from step one. , co-domain misalignment and sampling saturation ratio The calculation yields only the phase continuity index. The continuity evaluation is replaced by the output of a displacement sensor.

[0081] When using this method, avoid mistaking unstable areas as high-confidence contaminated areas or high-confidence uncontaminated areas. Step 2 directly outputs three types of layers: high-confidence contaminated dominant area, low-confidence undetermined area, and non-contaminated dominant area, so that step 3 does not require reinterpreting the physical source of the image.

[0082] Step 3: Based on the pollution status value output in Step 2 and phase position reliability First, the areas that can be directly identified as defective and the areas to be observed are separated in the unified unfolded diagram. Then, the confirmation, re-inspection, re-shooting and arbitration paths of the defect candidate segments are constrained by the state and arbitration budget, thereby breaking down whether it belongs to a defect and whether further verification is needed into single-chain actions that are connected one after the other.

[0083] The detection controller acts as the main execution unit, and the local re-blowing unit, arbitration detection mechanism, and diversion execution mechanism are scheduled by the detection controller. The detection controller first reads the first unfolded image output from step two. Second unfolded image Second unfolded image after reblowing Pollution status value and phase position reliability Then, the entire unified unfolded drawing is cut into continuous processing windows according to axial and circumferential sections.

[0084] For each processing window, the detection controller first determines the contamination status value. The system determines whether the processing window belongs to the clean priority zone, the contamination priority zone, or the pending zone, and then generates defect candidate values ​​within the clean priority zone. Then use the state maintenance To suppress repeated flipping between adjacent lines, arbitration demand values ​​are finally generated for the pending and high-risk areas. When the arbitration demand value Exceeding the arbitration trigger threshold At this time, the detection controller does not directly send the entire piece to the second detection condition. Instead, it generates an arbitration request package containing only the workpiece identification, axial start and end points, circumferential start and end points, trigger reason, and request mode, which is then handed over to the arbitration inspection agency for execution in the queue order. Thus, step three does not replace the existing logic with a more complex classifier, but rather links contamination gating, state maintenance, and arbitration budget into a continuous control chain, enabling the main detection path and the slow verification path to work collaboratively under the same terminology system.

[0085] Although step two has already provided the pollution status value. and phase position reliability However, if the entire unified unfolded diagram is directly sent to the defect judgment at this time, the contamination-dominant section, the low-phase position information section, and the actual defect section will still be mixed together. Especially in the scenario where there are thin water marks in the middle of the shock absorber connecting rod and local scratches at the end, if contamination gating is not performed first, step three will submit the boundary of the long wet film and the boundary of the short scratch to the subsequent logic processing together, causing too many state inversions, and thus dragging the arbitration and testing agency into a large number of invalid requests.

[0086] The detection controller outputs the pollution status value in step two. As the primary gating condition, with the first threshold Second threshold Three zones were defined: pollution status value Below the first threshold The area is now on a clean-priority path, with a pollution status value of [value missing]. Above the second threshold The area enters the pollution priority path, pollution status value Between the first threshold With the second threshold The area between them enters the path to be judged. Subsequently, the detection controller only calls the first unfolded image in the cleaning-priority path. Second unfolded image Extracting Defect Candidate Values Furthermore, local context is preserved within the path to be judged, and no negative or positive conclusions are made in advance. In this way, all subsequent actions originate from the already contaminated state value. The process starts with preliminary, stratified processing windows, rather than from a uniform, undifferentiated overall view. Specifically, the detection controller first processes each processing window based on its contamination status value. Generate cleaning gating factors Cleaning gate factor The closer the value is to 1, the closer the current location is to the cleaning priority zone; cleaning gating factor The closer the value is to zero, the closer the current location is to the contamination priority zone. Then, the detection controller checks the first unfolded image... Second unfolded image The linear continuous response, mirror discontinuity response, and pitted edge response are extracted, and then a clean gating factor is used. These responses are uniformly modulated. The linear continuous response is used to identify scratches extending axially or diagonally, and the specular interruption response is used to identify the first unfolded image. The phenomenon that the normally continuous specular bright band is partially interrupted, and the pitted edge response is used to identify the second unfolded image. Locally isolated annular or semi-annular boundaries.

[0087] In one implementation, the detection controller uses a combination of directional derivative and refined tracking for linear continuous responses, a local bright band length truncation method for specular interruption responses, and a ring sampling difference method for pit-like edge responses. For example, when a fine scratch exists in the middle section of a test member, the second unfolded image... A continuous, clearly oriented, thin dark line will form in the first unfolded image. In the middle, a local bright specular band breaks, and the detection controller will simultaneously read both a continuous linear response and a specular interruption response; however, if only watermark boundaries exist at the same location, then although the second unfolded image... Edges will also appear in the middle, but the pollution state value of this area is given in step two. High, cleaning gate factor It will inhibit it from entering the defect candidate path. Where: Cleaning gating factor Axial coordinates Circumferential coordinates The degree to which a candidate is allowed to enter the defect extraction path, with a value range of 0 to 1; the first threshold. Cleaning priority boundary, with values ​​greater than 0 and less than the second threshold. The default value; Second threshold Pollution priority boundary, with values ​​greater than the first threshold. The preset value is used to define the upper limit of whether a re-blowing should be prioritized or pending judgment; pollution state value The pollution dominance level at the current location output in step two, with a non-negative value range, is used as a cleaning gating factor. Input; in obtaining the cleaning gating factor Then, the detection controller continues to construct candidate defect values. : Where: Defect candidate value Axial coordinates Circumferential coordinates It more closely approximates the degree of real surface anomalies, with a value range of non-negative; linear continuous response. According to the second unfolded image The directional derivative and the linear anomaly response obtained by continuous tracking have a non-negative value range; in the implementation of the linear continuous response, the detection controller is in the second unfolded image. The directional gradients are calculated along the axial, circumferential, and two diagonal directions, respectively. Pixel bands with consistent gradient directions that extend continuously in the same direction are selected, and a linear continuous response is formed by the consistency of continuous length and direction. For long and narrow scratches, the response is higher; for the edges of watermarks, the response does not increase continuously due to their weaker directional consistency and persistence. Mirror interrupt response According to the first unfolded image The response obtained by measuring the degree to which the bright band of the mid-specular surface is cut off has a non-negative value range; the detection controller detects the first unfolded image. The center line of the local specular bright band is extracted, and the continuity and width changes of the bright band within two observation windows before and after the current position are calculated. When the current position causes the continuity of the specular bright band to be interrupted, and the interruption position is repeated in the continuous sampling line, a specular interruption response is formed. .

[0088] Pit-shaped edge response According to the first unfolded image With the second unfolded image The response obtained from local edge structure differences, with a non-negative value range, represents anomalies such as pitting, pinholes, and shallow pits; the detection controller is used in the second unfolded image. A ring-shaped sampling band is constructed with the current location as the center. The radial symmetry and local closure of the edge intensity within the ring-shaped sampling band are compared. When the edge is locally isolated, closed, or semi-closed, a pit-shaped edge response is formed. .

[0089] Weighting coefficient Weighting coefficients and weighting coefficients Representing linear continuous response Mirror interrupt response and pit-shaped edge response In defect candidate values The proportion of each factor is between 0 and 1, and the sum of the three factors is preferably 1, so that different defect morphologies can be represented under the same candidate value framework. When using, clean the gate control factor. Prior to defect candidate values This mechanism ensures that polluted areas do not mistakenly enter defect candidate paths based on strong local edges or bright bands; defect candidate values Simultaneously absorbing three types of responses—linear, mirror-interrupted, and pitted—ensures that the subsequent state retention in step three does not depend on a single shape; First unfolded image With the second unfolded image By assigning different responsibilities, the loss of physical origin is avoided by simply subtracting two image streams. The detection controller obtains defect candidate values. Instead of immediately labeling each pixel as defective or non-defective, a hold state is introduced for the same position on consecutive sampling lines. Maintain the status. The design motivation lies in the fact that real scratches, pits, and pinholes typically exhibit a positionally stable response across several consecutive sampling lines, while wet film edges and highlight boundaries are more prone to vertical jitter with slight positional changes. Therefore, the detection controller uses a confirmation threshold. and release threshold A hysteresis window is constructed, and a lower phase threshold is introduced simultaneously. Constraining low-phase position signal segments ensures that confirmed defect candidate segments do not immediately disappear due to fluctuations in a single line, and unconfirmed segments do not immediately become established due to spikes in a single line.

[0090]

[0091] In the formula: maintain state : No. Whether the corresponding position of the sampling line maintains the defect candidate confirmation state at the current time, with a value range of 0 or 1; defect candidate value : No. The candidate defect values ​​at the corresponding locations of the sampling lines are non-negative and are used to determine whether a candidate segment enters the confirmation process; the confirmation threshold... Maintain state The upper threshold for switching from 0 to 1, with values ​​greater than the release threshold. Preset value; release threshold Maintain state The lower threshold for switching from 1 to 0 is a value within the range of less than the confirmation threshold. The preset value prevents confirmed defect candidate segments from being released prematurely due to slight fluctuations; Phase position reliability : No. The phase confidence level at the corresponding position of each sampling line, ranging from 0 to 1, prevents low-confidence expansion segments from being directly confirmed; phase lower limit threshold. : The lower bound of the phase position confidence level for confirming defect candidates, ranging from 0 to 1, serves as a phase gate; Contamination state value : No. The degree of pollution dominance at the location corresponding to each sampling line, with a non-negative value, is used to directly release and maintain the state in the pollution-preferred area; the second threshold. Pollution priority boundary, with values ​​greater than the first threshold. The preset value is used to ensure that obviously contaminated sections do not retain the defect confirmation status; In actual workstations, if a component under test has localized scratches at its end, then the defect candidate values ​​on several consecutive sampling lines will be... It will continue to remain at the confirmation threshold. Above, the confidence level of simultaneous phase position Maintain a high level, maintain the status It will stabilize at one; if there is a narrow watermark in the middle section of the other test member, then the defect candidate value will be... The value will fluctuate at the boundary, while the pollution state value... It is close to or exceeds the second threshold. The detection controller will maintain the state. The pressure is lowered to zero, and the local re-blowing unit is invoked to re-blow the segment. This local re-blowing does not cover the entire test member, but is performed only on the local area corresponding to the current processing window based on the axial-circumferential position recorded in step two.

[0092] Keep in good condition during use. The candidate segment is transformed from a pixel-by-pixel instantaneous response to a stable state on a continuous line, reducing the impact of local noise and boundary jitter; the confirmation path and release path will not oscillate back and forth on the same threshold; the local re-blowing loop is embedded in the release logic of the holding state to avoid the wet film boundary occupying the main decision path for a long time.

[0093] The detection controller has successfully identified or released a portion of the area, but three types of sections remain that cannot be directly terminated: firstly, defect candidate values. Falling at the confirmation threshold and release threshold Maintain the state between. The first is the segment that is in a continuous but not yet definitively determined state; the second is the phase position reliability. The first category includes sections with low reliability, where local responses show anomalies but the reliability of the analysis is insufficient. The second category includes high-risk sections such as sealing sensitive areas and end-fitting areas, which require more careful handling even if the main judgment is not fully confirmed. Performing a full rescan on these sections would consume a large amount of slow resources; ignoring them all would allow truly high-risk anomalies to be missed downstream.

[0094] The detection controller first categorizes the sources of defects in the section to be arbitrated into four types: threshold band sources, low-phase sources, high-risk sources, and re-blowing inconsistencies. Threshold band sources refer to defect candidate values. Between the confirmation threshold With release threshold The segment between; low phase source refers to phase position confidence. Below the lower phase threshold The four categories of sources are: high-risk sections, high-risk sources, and inconsistencies in reblowing. High-risk sources refer to sections where the main judgment path differs before and after localized reblowing. The detection controller compresses these four types of sources into a single arbitration requirement value. Then, based on the current length of the arbitration request queue, the warning threshold, and the hard threshold, it determines which arbitration detection conditions to invoke. In this way, the arbitration detection agency does not receive the entire image, but only individual arbitration request packets with spatial boundaries and triggering reasons.

[0095] The detection controller first generates threshold band markings on the unified unfolded diagram. High-risk marker Inconsistent markings with re-blowing Threshold marking As a binary quantity, when the defect candidate value Located at the release threshold With confirmation threshold Set to 1 if the risk level is between 1 and 0 otherwise; High-risk marker. Generated from workpiece design documents and assembly sensitive area annotations; reblowing inconsistency mark. If the main judgment states before and after the local re-blowing contradict each other, then one of them shall be selected.

[0096] Subsequently, the detection controller uses these markers and phase position confidence. Calculate the arbitration demand value : Where: Arbitration demand value Axial coordinates Circumferential coordinates The threshold determines the necessity of entering the second detection condition, with a non-negative value range, and is used as the arbitration trigger criterion; the threshold is marked. Is the current location within the confirmation threshold? and release threshold The uncertain band between these values ​​is 0 or 1; when The value is 1 if the condition is met, otherwise it is 0. Phase position reliability : The confidence level of the current position's expanded coordinates, ranging from 0 to 1, giving higher arbitration priority to low-confidence segments; represents the position in the unified axial-circumferential coordinate system. The reliability of the unfolding result is calculated by combining the same-domain misalignment amount, phase continuity index and acquisition saturation ratio obtained in step one in step two, with a value range of 0 to 1. The larger the value, the more stable the actual surface phase at that location is and the more reliable the correspondence between the first unfolded image and the second unfolded image. When the value is lower than the preset threshold, the location is marked as a low-confidence region and used as the basis for arbitration triggering in step three.

[0097] High-risk marker : Whether the current position belongs to the sealing sensitive area, end mating area, or pre-specified area of ​​interest, with a value range of 0 or 1; indicates the position in the unified axial-circumferential coordinate system. Whether it belongs to a pre-defined high-risk area is generated by mapping the structural reference, sealing sensitive zone, end mating zone or preset interest zone of the workpiece under test, and the value is 0 or 1; where a value of 1 indicates that the position belongs to a high-risk area, and a value of 0 indicates that the position belongs to a normal area; when the position is marked as a high-risk area, step three prioritizes including it in the arbitration path, and step four uses it as a risk position input in the severity determination.

[0098] Inconsistent markings during reblowing : Whether the main judgment result before and after the local re-blowing is inconsistent at the current position. The value range is 0 or 1, used to capture unstable segments; 1 is taken when the main judgment state before and after the re-blowing is inconsistent, otherwise 0 is taken; weight coefficient Weighting coefficients Weighting coefficients and weighting coefficients These represent the threshold band source, low phase source, high risk source, and re-blow inconsistency source in the arbitration demand value, respectively. The percentage in the value ranges from 0 to 1 and is used to adjust the arbitration priority according to the on-site working conditions. In one implementation, the detection controller will arbitrate the demand value. Above the arbitration trigger threshold The sections are merged into rectangular or arc-shaped arbitration windows, and then an arbitration request package is generated. The arbitration request package includes workpiece identification, axial start point, axial end point, circumferential start point, circumferential end point, trigger cause word, request mode word, and timestamp. If the trigger cause word only contains threshold band sources, the request mode word is preferably second-view imaging; if it contains both high-risk sources and re-blowing inconsistency sources, the request mode word is preferably laser profile or mode projection. The result that can be observed on-site is that the entire rod under test is not moved to the arbitration testing agency, but only a small area in the middle section or a ring area at the end is reviewed separately.

[0099] When used, the arbitration requirement value Multiple sources are unified into a single, sortable arbitration entry point, avoiding the need for secondary detection conditions to switch rules between different sources; the arbitration request package clearly records spatial boundaries and triggering reasons, eliminating the need for arbitration detection agencies to reinterpret the main judgment logic; high-risk sources and sources with inconsistent re-evaluation are written into the same arbitration demand value. Subsequently, process risks and image uncertainties can be coordinated on the same scale.

[0100] The detection controller maintains an arbitration request queue and two queue thresholds: a warning threshold and a hard threshold. Each request packet in the arbitration request queue carries a workpiece identification, spatial boundary, and request mode word. Under normal circumstances, when the length of the arbitration request queue is lower than the warning threshold, the detection controller calls the corresponding arbitration detection agency according to the request mode word; when the length of the arbitration request queue reaches the warning threshold, the detection controller reduces the new request mode to fast second-view imaging, allowing the queue to clear the backlog first; when the length of the arbitration request queue reaches the hard threshold, the detection controller stops adding new 3D arbitration requests, marks new request packets as bypass re-inspection requests or pending verification requests, and fixes the current workpiece's main judgment status as pending confirmation.

[0101] On a typical production line, the main station continuously processes the components to be tested, while the arbitration testing mechanism is positioned to the side of the main line. For local strip-shaped areas in the middle, a second-view camera takes supplementary photos at a fixed angle; for sensitive sections with end seals, a laser profilometer reads local depth variations; if the station allows for a brief stop, a pattern projection device projects stripes at the arbitration station and acquires the local three-dimensional shape. The testing controller uses the same calling protocol for the three types of arbitration testing mechanisms, differing only in the request pattern word and the data type of the returned data. The returned results are uniformly compiled into an arbitration result record, which includes at least the workpiece identification, arbitration window number, arbitration conclusion, return timestamp, and original trigger reason word.

[0102] In the parallel expansion method, if a laser profilometer is not available on-site, the request pattern word for high-risk sections remains unchanged, but is executed by a high-resolution second-view camera. If the on-site line speed is low and a stop station is available, the pattern projection device takes over most high-risk requests. If the arbitration testing agency and the main station are not on the same rack, the testing controller still completes the call using the same request packet format, only requiring the addition of the arbitration station number to the request packet. Regardless of the arbitration testing agency used, the terminology system remains consistent, i.e., the arbitration requirement value. Step four involves deciding whether to join the queue, the arbitration request queue determining when to execute, and the arbitration result record determining when to send it back.

[0103] When in use, the arbitration testing agency is explicitly defined as a budget-constrained slow resource, rather than part of the main decision path; the early warning threshold and hard threshold make queue congestion explicit in advance, so that the main station will not be held up by a few difficult cases; through unified request packets and unified return records, the second testing condition maintains a clear coupling relationship with the main decision path, avoiding the output of step three from remaining at the level of abstract suggestions for re-inspection.

[0104] Step 4: Based on the main judgment result and arbitration result record output in Step 3, perform severity judgment, defect digital map generation, handling status output and parameter write-back on abnormal sections within the same workpiece, so that the single-piece conclusion and batch correction share the same set of spatial coordinates and status terms.

[0105] The detection controller first merges scattered abnormal pixels into defect segments according to a unified axial-circumferential coordinate system, and then calculates the severity value for each defect segment. Then based on the severity value and high-risk markers The system generates a disposal status marker and finally outputs the defect digital map, disposal status marker, and structured result package of the workpiece to the rejection actuator, manufacturing execution system interface, statistical process control interface, and evidence storage.

[0106] At the same time, the detection controller also extracts contaminated-dominant samples, non-contaminated-dominant samples, and high-confidence clean samples from the confirmed results to establish a batch contamination baseline. Phase compensation bias The arbitration budget baseline is written back. After this process, step four is not an isolated result display end, but rather the judgment results of steps one to three at the single-piece level are solidified into executable actions, and these actions are applied in reverse to the detection parameters of the next workpiece.

[0107] Among them, the hold state output in step three Arbitration result records are essentially still segment-level or local window-level information. If each segment is directly output as a defect at this point, situations will arise where the same scratch is divided into multiple segments, the same pit is counted repeatedly, and the same end anomaly is repeatedly reported in the main decision path and the arbitration path. On the other hand, simply using whether arbitration is triggered as the basis for determining the severity of a defect cannot distinguish between a shallow scratch and a deep pit located in a sealing sensitive area.

[0108] Among them, the detection controller first maintains the state. The main judgment segment is spatially overlapped with the arbitration window identified as abnormal in the arbitration result record. If the two segments meet the overlap condition in both the axial and circumferential directions, they are merged into the same defect segment number. If the two do not intersect axially and only pass through a high-risk marker circumferentially. If associated, the two defect segment numbers will remain the same. Each defective section is numbered. Once generated, its axial start point, axial end point, circumferential start point, and circumferential end point are fixed. Subsequently, the detection controller extracts the segment average defect candidate value, segment arbitration measurement, axial length, circumferential span, and high-risk marker from the corresponding segment, and then forms a severity value. In this process, the output of the arbitration testing agency does not replace the main determination path, but is included in the severity calculation along with the main determination path.

[0109] The detection controller assigns a number to each defective section. Extracting candidate values ​​of average defects in sections Candidate value of average defect in section All states within this section are maintained. Defect candidate value corresponding to a pixel that is 1 It is aggregated. If the defective section number... For at least one arbitration window, the detection controller reads the arbitration measurement corresponding to that arbitration window. If the defective section number If the arbitration path is not entered, the detection controller will use that segment in the first expanded image. Second unfolded image The principal morphological response is used to calculate the alternative measurement, which is then recorded as the arbitration measurement. This ensures that the terminology remains unchanged. The arbitration measurement here... In second-view imaging mode, the distortion originates from boundary sharpness and local light and dark breaks; in laser contour mode, it originates from local height undulations; and in pattern projection mode, it originates from the amplitude of stripe deformation. Subsequently, the detection controller calculates the axial length. and circumferential span This constitutes the severity score. .

[0110] In one implementation, defect segment fusion is performed using eight-neighbor connectivity within a unified axial-circumferential grid; if a scratch is briefly interrupted in the main decision path, but both ends are confirmed by the same arbitration window, the detection controller treats it as having the same defect segment number. .

[0111] If two pits are close together axially but clearly separated circumferentially, they should be kept as two separate defect sections. This allows the defective digital map to reflect real spatial objects, rather than intermediate fragments from the algorithm. Specifically: Where: Severity value Defect section number The overall severity, with a non-negative value range, is used as the core criterion for marking the subsequent handling status in step four; the average defect candidate value of the section. Defect section number The average defect candidate degree of the internal main decision path, with a value range of non-negative; arbitration measurement quantity. Defect section number Measurement results obtained from arbitration testing institutions or alternative measurement pathways have non-negative values; high-risk marker. Defect section number Whether it falls into a sealed sensitive area or a preset high-risk area, the value range is 0 or 1; Axial length Defect section number The geometric length along the axial direction, taking values ​​greater than or equal to 0, is the actual length value; axial reference length. : Axial length The normalized benchmark, with a value range of a preset length value greater than 0, maps different length levels to a unified scale; circumferential span measurement. Defect section number The geometric span in the circumferential direction takes values ​​greater than or equal to 0, which are either the actual angular span or the equivalent arc length; the circumferential reference length. Circumferential span The normalized benchmark takes a value range of a preset length value greater than 0, mapping different circumferential span levels to a unified scale; Weighting coefficient Weighting coefficients Weighting coefficients Weighting coefficients and weighting coefficients These represent the candidate values ​​of the average defect in the section. Arbitration measurement High-risk marker Axial length and circumferential span In severity value The proportion in the value ranges from 0 to 1, with a preferred sum of 1, to unify the scales from different sources; Arbitration Measurement Unified normalization to the closed interval of 0 to 1. When the arbitration mode is second-view imaging. Calculated based on the degree of boundary fracturing and boundary sharpness; when the arbitration mode is laser profiling, the arbitration measurement... Calculated from the normalized value of the local height difference relative to the reference height; when the arbitration mode is mode projection, the arbitration measurement... The normalized value is calculated from the local stripe deformation amplitude relative to the reference deformation amplitude. All three normalization results are denoted as arbitration measurements. and enter the same severity value. formula.

[0112] As a further supplement: when the arbitration mode is second-view imaging, the definition is: Boundary fracture amount Indicates the degree of abnormal boundary breakage in the arbitration window; boundary reference value. Indicates the boundary reference value for clean samples; sustained-release constant. This indicates a positive value that prevents the denominator from being zero.

[0113] When the arbitration mode is laser contour, the definition is: Local height difference This indicates the height difference between the abnormal area within the arbitration window and the reference plane; height reference value. This indicates the height reference value for a clean sample.

[0114] When the arbitration mode is a mode projection, the following definition is defined: stripe deformation This indicates the offset of the stripes within the arbitration window relative to the reference shape; stripe reference value. The stripe reference deformation represents the clean sample.

[0115] The outputs of the three arbitration modes, after being normalized as described above, are uniformly denoted as the arbitration measurement quantity. The range of values ​​is all non-negative.

[0116] For example, if there is an axially extending scratch in the middle of a member under test, then the candidate value of the average defect value of the section is... Higher, axial length Larger circumferential span If the risk level is low and the area is not located in a high-risk zone, then a high-risk marker will be used. Zero, severity value Mainly composed of the average candidate value of defects in the section Arbitration measurement and axial length The decision is made jointly. If a localized deep pit exists in the sealing sensitive area at the end of another member under test, the axial length measurement... and circumferential span Not necessarily large, but a high-risk marker. As one, arbitration measurement It will also increase the severity value. .

[0117] When using, the severity value Generate a defect digital map record, which must include at least the workpiece identification and the defect section number. Generate a defect digital map record, which must include at least the workpiece identification and the defect section number. Axial start point, axial end point, circumferential start point, circumferential end point, severity value High-risk marker Source markers and evidence indexes. Source markers indicate whether the defective section primarily originates from the main decision path, the arbitration path, or a fusion of both; the evidence index points to the first unfolded image. Second unfolded image Second unfolded image after reblowing The system stores the images or curves from the arbitration testing agency. Subsequently, the testing controller generates a disposal status marker based on severity thresholds and high-risk location rules. The disposal status markers include at least four types: release, rejection, bypass re-inspection, and pending confirmation.

[0118] In one implementation, the rejection mechanism is positioned at a fixed distance downstream of the detection position. The rejection mechanism employs a cylinder push rod, a rotary fork, or a robotic gripper. The manufacturing execution system interface and the statistical process control interface are connected to the detection controller via an industrial Ethernet network. The evidence storage is located in the detection controller's local solid-state memory or a host computer array. After reading the transport position signal, the detection controller does not immediately act upon the generation of the defect digital map. Instead, it binds the current workpiece's handling status marker to its current position, triggering the corresponding execution only when the workpiece reaches the rejection mechanism's operating area. If the handling status marker is "bypass re-inspection," the diversion baffle opens; if the handling status marker is "rejection," the rejection mechanism operates; if the handling status marker is "pending confirmation," the workpiece is sent to a buffer position and not released prematurely.

[0119] Taking on-site action as an example, a test member has two defective sections in step three, with the middle section showing the severity of the scratch. It falls into the general abnormality level, while the severity value of the shallow pit in the sensitive area of ​​the end seal is... Although it did not reach the level of a mid-section scratch, it was marked as high-risk. The detection controller generates two defect digital map records and marks the overall handling status of the workpiece as either bypass re-inspection or rejection. The operator sees two areas with coordinates and type labels on the workpiece outline on the host computer interface, instead of a string of isolated values.

[0120] In the parallel implementation, if the manufacturing execution system interface is not configured on site, the defect digital map record can still be written to the local evidence storage and exported manually; if the bypass re-inspection channel is not configured on site, the bypass re-inspection in the handling status mark can be replaced by the pending confirmation cache bit; if the removal execution mechanism uses a robot, the detection controller adds a grasping coordinate field to the result package, but the basic terminology of the defect digital map record remains unchanged.

[0121] When in use, the defect digital map record transforms the algorithm's judgment results into spatialized data objects that can be executed on-site, enabling the elimination execution mechanism, bypass re-inspection execution mechanism, and the upper-level system to use the same data source; the handling status mark is bound to the workpiece position to avoid the result being disconnected from the execution time; the evidence index attaches the key image objects from steps one to three to the current workpiece, providing a clear sample source for the subsequent parameter writing in step four.

[0122] If step four only outputs the single-piece disposal status to the actuator without applying the confirmed sample back to steps one, two, and three, then the entire system will still require frequent manual intervention during batch switching, light aging, humidity changes, and clamping condition changes. Especially when a batch of test components has excessive cleaning residue but the actual defects have not increased, if the contamination baseline is not written back, the first threshold in step three will be affected. With the second threshold It will remain in a tight state, resulting in too many bypass re-inspections; conversely, when the phase reference condition drifts slowly, if the phase compensation bias is not written back, the actual surface phase in step two will gradually deviate from the real surface motion.

[0123] The detection controller only extracts write-back samples from workpieces whose handling status has been confirmed, and not from workpieces awaiting confirmation. For write-back samples, the detection controller further distinguishes between contamination-dominant samples, non-contamination-dominant clean samples, and defect confirmation samples. Contamination-dominant samples are used to update the batch contamination baseline. Non-contaminated clean samples were used to update the phase compensation bias. Defect confirmation samples are used to update the arbitration budget baseline and maintain alarm counts. Subsequently, the detection controller is processed in batches. Execute parameter writeback and update the batch contamination baseline. Phase compensation bias The arbitration budget baseline is written into the detection context of the next workpiece. In this way, the write-back is not to make additional judgments for the current workpiece, but to modify parameters for subsequent workpieces.

[0124] The inspection controller scans the defect digital map record of each confirmed workpiece. If a workpiece's handling status is marked as "released," and some sections of it were identified as contamination-dominant in step three and disappeared after local reblowing, then these sections are included in the contamination-dominant sample set. If a workpiece's handling status is marked as "released," and its sections showed a phase position confidence level in step two... If no arbitration was triggered in step three, these segments are included in the non-contamination-dominated clean sample set. Then, the detection controller performs batch updates for both types of samples.

[0125] Batch contamination baseline The update involves using a residual weighted write-back method to update the new pollution state value. Baseline changes were driven only on high-confidence contamination-dominant samples, where: Where: Batch contamination baseline : No. The contamination baseline used in this batch, with a non-negative value range, is used as the first threshold in step three. Second threshold Reference center; batch contamination baseline The contamination baseline for the next batch, with a range similar to the batch contamination baseline. Consistent; pollution status value : No. The pollution status value corresponding to each dominant pollution sample segment, with a non-negative value range, is used to provide information about this batch of samples; the number of polluted samples. : The number of sample segments participating in the contaminated baseline rewriting, with values ​​ranging from positive integers, used to provide an upper bound for the summation range; Pollution writeback weight : No. The effective weights for each polluted dominant sample segment, ranging from 0 to 1, are used to ensure that high-confidence, disappeared polluted segments contribute more to the baseline; the pollutant write-back step size. Batch contamination baseline The update step size, ranging from 0 to 1, is used to limit the impact of a single batch on subsequent batches; the release constant. The denominator protection factor during baseline contamination updates is a small constant with a value range greater than 0, used to prevent denominator instability when the sample size is extremely low. Phase compensation bias The update involves selecting high-confidence, non-contaminated, predominantly clean samples for the detection controller, and then analyzing the actual surface phase of these samples. With nominal phase The difference is used to calculate the compensation bias required in step two; Where: Phase compensation bias : No. The phase compensation offset used in this batch is a real number within the allowable range of circumferential angular deviation, used to provide a reference correction for determining the actual surface phase in step two; phase compensation offset The value range of the phase compensation bias used in the next batch is the same as the phase compensation bias. Consistent, used to receive write-back results; actual surface phase : No. The actual surface phase of a high-confidence clean sample segment, with values ​​ranging from the circumferential position within a uniform unfolding period, is used to provide a true motion reference; the nominal phase... : No. The nominal phase of a high-confidence clean sample segment, with a value range matching the actual surface phase. Consistency is used to provide a reference for the motion of the mechanism; Number of phase samples Participating in phase compensation bias The number of sample segments to be written back, with values ​​ranging from positive integers, is used to provide an upper bound for the summation range; phase write-back weights. : No. The effective weights for high-confidence clean sample segments, ranging from 0 to 1, are used to determine the relative confidence levels. High-risk segments that do not enter the high-risk zone contribute more to bias updates; phase write-back step size Phase compensation bias The update step size, ranging from 0 to 1, is used to limit the phase correction amplitude of a single batch; the release constant. The denominator protection value during phase compensation bias update is a small constant with a value range greater than 0, used to prevent update instability caused by too small a sample size. In one implementation, the detection controller performs a batch contamination baseline test after each batch of a fixed number of workpieces. and phase compensation bias Write back and update the batch contamination baseline. Mapped to the first threshold in step three Second threshold The updated phase compensation bias The initial term is obtained by mapping to the actual surface phase in step two. If the number of samples in the same batch is insufficient, the original value is maintained.

[0126] When using this method, parameter write-back should only use confirmed result samples to prevent pending samples from contaminating the baseline; batch contamination of the baseline is also prevented. With phase compensation bias These address two types of problems: wet-state changes and mechanical drift, allowing steps two and three to be corrected accordingly. The introduction of step size and weight terms ensures that batch parameter changes remain continuous, preventing parameter jumps caused by a single extreme workpiece.

[0127] Furthermore, after reading the arbitration result record, the detection controller will identify high-risk markers within this batch. This includes the number of sections confirmed by arbitration, the number of bypass re-inspection workpieces, and the number of workpieces pending confirmation. If high-risk confirmation sections appear consecutively within a fixed axial band, the detection controller not only adjusts the arbitration budget baseline for the next batch but also outputs a maintenance alarm. This maintenance alarm is process-oriented, indicating potential bias in cleaning nozzles, clamps, or electroplating stations, rather than changing the current workpiece handling status. The maximum allowed length of the arbitration request queue. The threshold for arbitration request queue warnings is determined jointly by the average processing time of the second detection condition, the allowed waiting time on the main line, and the bypass buffer capacity. Take as the maximum allowed length The preset ratio, the hard threshold of the arbitration request queue Take as the maximum allowed length .

[0128] In one implementation, the arbitration budget baseline includes an arbitration request queue warning threshold, an arbitration request queue hard threshold, and a high-risk area priority quota. The detection controller adjusts the high-risk area priority quota based on the source composition of confirmed defects: if the proportion of actual defects in a high-risk area increases, the priority quota for the next batch of high-risk areas is increased; if the vast majority of arbitration requests are ultimately confirmed to be contamination-driven, the arbitration quota for the next batch of general sections is reduced, while the triggering of local re-blowing is advanced. Simultaneously, the detection controller performs an exit determination for the current workpiece: when all defective sections of the workpiece have been marked with a handling status and there are no pending arbitration windows, the detection controller marks the current workpiece as closed; when there are still pending arbitration windows, the detection controller marks the current workpiece as pending confirmation, keeping its result package in an unreleased state.

[0129] For example, if the laser profilometer confirms the presence of localized deep pits in the end-sealed sensitive areas of several consecutive test members, the detection controller will simultaneously generate a rejection or bypass re-inspection command for the current workpiece and write the event of continuous high-risk end-sealing confirmation to the maintenance alarm channel. If, in another batch of workpieces, most arbitration requests are determined to be watermark boundaries after second-view imaging, the detection controller will not amplify the arbitration path but will instead increase the pre-emptive level of localized re-blowing and tighten the arbitration budget for general sections. What can be seen on-site is an alarm prompt on the maintenance interface targeting specific axial bands and workstations, rather than just a series of statistical values.

[0130] In the parallel implementation, if there is no statistical process control interface on site, maintenance alarms can be directly written to the local workstation touch screen and audible and visual alarm unit; if there is only one arbitration testing agency, the arbitration budget baseline is simplified to two items: request queue early warning threshold and request queue hard threshold; if the production cycle is fixed and buffering of workpieces to be confirmed is not allowed, the workpieces to be confirmed are received by the bypass buffer area, and the process exit is still based on whether the result packet is closed, without changing the terminology system.

[0131] When in use, the arbitration budget baseline write-back prevents slow detection resources from being occupied for a long time due to occasional occurrences of a batch of workpieces in the past; maintenance alarms elevate continuously occurring high-risk confirmation sections to process-level signals, preventing the system from simply rejecting the workpieces without indicating workstation offsets; exit judgment uses whether the result package is closed as the sole criterion, making the termination conditions of the current workpiece process clear and executable.

[0132] Ultimately, all the above steps are uniformly scheduled and executed by the detection controller. The detection controller receives and manages three types of data objects: raw intra-domain data packets, arbitration request packets, and structured result packets. The raw intra-domain data packets contain workpiece identification, sampling sequence number, timestamp, first condition line data, second condition line data, cumulative encoder pulses, axial position, reference center position, radial displacement, intra-domain misalignment, phase continuity index, acquisition saturation ratio, acquisition quality index, and re-acquisition version number. The detection controller first converts the nominal phase from the cumulative encoder pulses, then corrects it with the reference center position and radial displacement to obtain the actual surface phase, and writes the line data into a unified axial-circumferential mesh through line-by-line cylindrical unfolding mapping. For the re-acquisition version number data formed by local re-blowing / re-sampling, the detection controller updates the second unfolded image after re-blowing within the same spatial window. The detection controller then generates a contamination state value based on the dual-condition response difference, re-blowing response difference, and the projection length of the principal axis of the connected domain; it generates candidate values ​​for defects based on the cleanliness gating factor constraint; it maintains the state based on the confirmation threshold, release threshold, and phase position confidence; and it generates arbitration requirement values ​​based on threshold band markers, low phase areas, high-risk areas, and re-blowing inconsistency markers. Measurement results from different modes returned by the arbitration testing agency are all normalized into arbitration measurements and included in the severity value calculation. Once the current workpiece's handling status is closed, it does not change due to the writing back of the current workpiece's parameters; the batch contamination baseline, phase compensation bias, and arbitration budget baseline take effect from the next workpiece or the next batch. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0133] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0134] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0135] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0136] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A machine vision-based online detection system for surface defects in shock absorber connecting rods, characterized in that: include, In response to the cleaned workpiece entering the detection position, line scanning is performed on the same surface area under the first and second optical conditions, and nominal motion data and phase reference raw data are acquired simultaneously. The actual surface phase is obtained based on the nominal motion data and the original phase reference data. The acquired dual-condition line data is then expanded and corrected line by line based on the actual surface phase to obtain the first and second expanded images under the unified axial-circumferential coordinates. Contamination state map and phase position confidence are constructed based on the first and second unfolded images, and the determination of re-blowing, re-shooting or defects is gated based on the contamination state map; The second detection condition arbitration is triggered only for uncertain areas, areas where the phase position confidence is lower than the preset value, or preset high-risk areas, and the defect location, size, and severity are output based on the main judgment result and the arbitration result.

2. The online detection system for surface defects of the shock absorber connecting rod according to claim 1, characterized in that: Before online scanning and acquisition, local pre-liquid removal treatment is performed on the surface of the workpiece to be measured. Under unified hardware triggering, the first condition line image, the second condition line image, nominal motion data and phase reference raw data are acquired synchronously. These are then bound with the workpiece identification, timestamp and acquisition quality mark to form the original co-domain data packet. The original co-domain data packet serves as the input object for subsequent determination of the actual surface phase, generation of unfolded image and construction of contamination state map.

3. The online detection system for surface defects of the shock absorber connecting rod according to claim 2, characterized in that: The acquisition quality is judged for the original intra-domain data packets. When any of the following situations are detected, such as the saturation pixel ratio exceeding the preset upper limit, the loss or continuous distortion of the original phase reference data, the continuous appearance of local high humidity areas, or the time alignment error of the dual condition line image exceeding the preset value, local re-blowing and local re-shooting are performed on the corresponding spatial window, and a re-sampling version number is written to the re-sampling result. The re-sampling version number increases with the number of re-samplings.

4. The online detection system for surface defects of the shock absorber connecting rod according to claim 3, characterized in that: The nominal phase is obtained based on the nominal motion data, and the phase correction is obtained by combining the reference center position and the reference position in the original phase reference data; When the system is configured with a radial displacement signal, the nominal phase is corrected by combining the radial displacement and the reference radial displacement to obtain the actual surface phase corresponding to each line image. The actual surface phase is used as the writing reference for the unified circumferential coordinates.

5. The online detection system for surface defects of the shock absorber connecting rod according to claim 4, characterized in that: Based on the actual surface phase and the axial position in the nominal motion data, each first condition line image and second condition line image is written into a unified axial-circumferential grid to generate a first unfolded image and a second unfolded image. For sections with sufficient samples, shape-preserving piecewise cubic interpolation is used, and for sections with insufficient boundary samples, bilinear interpolation is used to complete the line-by-line unfolding correction.

6. The online detection system for surface defects of the shock absorber connecting rod according to claim 5, characterized in that: Based on the local response difference between the first and second unfolded images, the response change of the second unfolded image before and after local reblowing, and the continuous length of the abnormal connected domain along the principal axis, a pollution state map under a unified coordinate system is constructed, and the pollution state map is kept in positional correspondence with the first and second unfolded images.

7. The online detection system for surface defects of the shock absorber connecting rod according to claim 6, characterized in that: The phase position confidence is generated based on the same-domain misalignment, phase continuity index, and acquisition saturation ratio, and is written into the unified expansion cache synchronously with the pollution state map. The phase position confidence is used to limit the segments that can enter the defect candidate extraction path and serves as the input for subsequent arbitration triggering conditions.

8. The online detection system for surface defects of the shock absorber connecting rod according to claim 7, characterized in that: Defect candidate values ​​are generated only in areas where the pollution state map is below the first threshold, based on the linear continuous response of the second unfolded image, the mirror interruption response of the first unfolded image, and the pit-like edge response of the first unfolded image and the second unfolded image. The defect candidate values ​​are then mapped one-to-one with uniform coordinates and used as the basic input for the main decision path.

9. The online detection system for surface defects of the shock absorber connecting rod according to claim 8, characterized in that: The maintenance state is maintained based on the confirmation threshold, release threshold, and phase lower limit threshold. When the defect candidate value reaches the confirmation threshold and the phase position confidence reaches the phase lower limit threshold, the defect candidate segment is confirmed. When the contamination state map is higher than the second threshold, the maintenance state is released and local re-blowing and local re-shooting are triggered for the corresponding segment.

10. The online detection system for surface defects of the shock absorber connecting rod according to claim 9, characterized in that: When the segment is in the threshold band segment, the phase position confidence is lower than the preset value, it is located in the preset high-risk area, or the main judgment status is inconsistent before and after the re-blow, an arbitration request packet is generated and sent to the arbitration request queue. An arbitration request package should include at least the workpiece identification, axial boundary, circumferential boundary, triggering reason, and request mode.

11. The online detection system for surface defects of a shock absorber connecting rod according to claim 10, characterized in that: The arbitration request queue is configured with an arbitration trigger rate cap, a queue warning threshold, and a queue hard threshold. When the queue reaches the queue warning threshold, newly incoming arbitration request packets are downgraded to fast second-view re-evaluation. When the queue reaches the queue hard threshold, new 3D re-evaluation is stopped and new request packets are transferred to bypass re-examination requests.

12. The online detection system for surface defects of a shock absorber connecting rod according to claim 11, characterized in that: The abnormal segments corresponding to the maintained state are spatially fused with the abnormal windows in the arbitration result record according to the unified axial-circumferential coordinates to form a defect segment number. The severity value of each defect segment number is generated based on the average defect candidate value of the segment, the arbitration measurement, the preset high-risk area marker, the axial length, and the circumferential span.

13. The online detection system for surface defects of the shock absorber connecting rod according to claim 12, characterized in that: A defect digital map is generated based on the axial range, circumferential range, severity value, and evidence index of each defect section number, and release instructions, rejection instructions, or bypass re-inspection instructions are output according to the handling status mark. The confirmed results are written back to the contamination baseline, phase compensation bias, and arbitration budget baseline, and take effect from the next workpiece or the next batch of workpieces.

Images