New energy tire forming quality defect detection method based on active infrared thermal imaging

Abstract

The present application relates to tire forming detection technical field, more particularly, the present application relates to a kind of new energy tire forming quality defect detection method based on active infrared thermal imaging, comprising: obtaining the thermal imaging sequence collected after the pulse thermal excitation of the new energy tire to be detected is applied;Phase map of each frequency is obtained by performing discrete fourier transform on thermal imaging sequence pixel by pixel.The present application objectively divides shallow and deep dominant frequency by constructing phase space stationary index;Quantify the shallow layer spatial stability of each pixel to extract adaptive masking strength, and then quantitatively strip shallow penetration interference in deep phase map;Finally, based on the dynamic weighted fusion of each frequency defect response intensity deep layer feature.The present application completely breaks the limitation of fixed frequency phase analysis, greatly enhances the detection contrast of deep weak defects, and provides reliable guarantee for new energy tire factory quality and driving safety.

Description

Technical Field

[0001] This invention relates to the field of tire forming inspection technology. More specifically, this invention relates to a method for detecting defects in the forming quality of new energy tires based on active infrared thermal imaging. Background Technology

[0002] Pulse-phase thermal imaging is an active infrared non-destructive testing method that combines pulsed thermal excitation with frequency domain analysis. It performs a pixel-by-pixel discrete Fourier transform on the thermal imaging sequence acquired after excitation, using the phase map corresponding to each frequency component as a defect criterion. It has already been applied in fields such as aerospace composite materials and electronic packaging. With the rapid popularization of new energy vehicles, higher requirements are placed on the structural strength and driving safety of tires. New energy tires are composed of multiple heterogeneous materials such as ply, steel wire, carcass rubber, and tread rubber during the molding process. Uneven rubber flow and local pressure deviations can easily introduce internal defects such as bubbles, delamination, and inclusions at the interlayer interfaces. Introducing pulse-phase thermal imaging into the factory inspection of new energy tire molding quality aims to non-destructively detect these hidden defects within the multi-layered structure before tire installation, preventing them from gradually expanding under high-speed alternating loads and ultimately causing tire blowouts.

[0003] In related technologies, for example, Chinese patent application CN114219765A provides a method for adaptively extracting defects from infrared thermal images based on phase features. First, infrared thermal image sequences during the heating phase are acquired online in real time. Then, the infrared thermal image sequences are preprocessed. Next, a suitable number of frames of infrared thermal images are selected for spatial high-pass filtering. Specific phase information is extracted from the filtered image sequence to obtain a phase map at a specific frequency. Finally, the phase map is converted into a grayscale image and subjected to median filtering. The filtered image clearly reflects the size and location of the defects.

[0004] However, in the actual testing of tires, a multi-layered composite structure, heat waves are reflected at the interfaces of various layers, such as the ply, steel wire, and carcass rubber. The time-domain temperature curve at the same location is actually superimposed with multiple reflection components from interfaces at different depths. This results in the phase diagrams of each frequency containing both the background phase of the shallow, continuous interfaces and the phase response of deep defects. Because the shallow interfaces are continuously distributed throughout the entire testing area and have strong reflections, their phase components dominate in the frequency domain, masking the phase of deep defects located near the steel wire and inside the carcass rubber. This leads to the fact that, during the segmental inspection of real tires, deeply buried air bubbles and delamination defects are almost invisible on the phase diagram, with severely reduced contrast. Quality inspection often judges them as qualified and releases them. After these tires with deep-seated hidden dangers enter the vehicle assembly and are put on the road, the defects continue to expand during long-term driving, becoming a potential safety hazard. This makes it difficult for factory testing to truly guarantee quality and safety, and fails to meet the actual needs of quality testing for new energy tire molding. Summary of the Invention

[0005] This invention provides a method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging. It aims to solve the problem in related technologies where the strong background phase of the shallow interface in the multi-layer composite structure of tires severely obscures the signal of deep defects, resulting in low detection contrast, high false negative rate, and difficulty in accurately identifying hidden defects inside the tire body.

[0006] In a first aspect, the present invention provides a method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging, comprising: S1, acquiring a thermal imaging sequence of the new energy tire to be inspected after applying pulsed thermal excitation; S2, performing discrete Fourier transform on the thermal imaging sequence pixel by pixel to obtain phase maps of each frequency, calculating the local phase standard deviation of each phase map using a sliding window, and constructing a phase space stationarity index reflecting the degree of spatial consistency of the phase maps of each frequency; S3, dividing all frequencies into a shallow dominant frequency set and a deep dominant frequency set based on the phase space stationarity index, and extracting the phase maps pixel by pixel. The shallow background phase reference value of the phase value within the shallow dominant frequency set is obtained by weighting and correcting it with local spatial stability to obtain the shallow phase occlusion intensity of each pixel; S4, the shallow phase occlusion intensity is used as the weight to perform pixel-by-pixel stripping on the phase map of each frequency in the deep dominant frequency set to obtain the stripped phase map, and then the defect response intensity of each frequency at each pixel position is used as the fusion weight to perform weighted fusion on the stripped phase map to obtain the deep frequency adaptive fusion index; S5, adaptive threshold segmentation and connected component analysis are performed on the deep frequency adaptive fusion index to output the spatial location of the defect region and the defect confidence.

[0007] By constructing a phase space stability index, the complex frequency domain is adaptively divided into two sets: shallow-dominant and deep-dominant. The shallow background phase reference is accurately extracted and combined with local spatial stability quantification into pixel-by-pixel occlusion intensity. Then, using this as weight, shallow interference components are stripped pixel by pixel in the deep phase map, completely eliminating the masking of deep features by strong reflections from shallow structures. Finally, the stripped deep features are adaptively weighted and fused based on the defect response intensity at each frequency and location. This allows hidden defects that were originally buried deep in the steel wire layer or tire carcass rubber and had extremely weak contrast to be highlighted in the final fused index map. Compared with the traditional static output of a single phase map, this significantly improves the detection contrast and recognition reliability of defects across the entire depth range, fundamentally ensuring the effective interception of deep structural hazards before new energy tires leave the factory.

[0008] Furthermore, the formula for calculating the phase space stationarity index includes: ;in, For frequency The corresponding phase space stationarity index, For frequency The mean of the corresponding local phase standard deviation map across all pixels in the entire image. This represents the maximum value in the local phase standard deviation plot corresponding to all frequencies.

[0009] By normalizing and inverting the mean of the local phase standard deviation of the entire image with the global maximum standard deviation, a phase space stationarity index with unified dimensions is constructed. Compared with the conventional operation of subjectively selecting frequency intervals based on human experience, this index can objectively and quantitatively map the smoothness and consistency of each frequency phase map in the global space, providing a solid data basis for subsequent accurate extraction of frequency intervals dominated by shallow background.

[0010] Furthermore, frequency The corresponding method for obtaining the local phase standard deviation plot includes: for frequency The corresponding phase diagram, with side length as A sliding window of pixels is used to traverse the entire phase map. Within each window, the standard deviation of the phase values ​​of all pixels within the window is calculated, and the local phase standard deviation corresponding to the center pixel of that window is obtained. After traversing all windows, the frequency is obtained. The corresponding local phase standard deviation plot.

[0011] Local neighborhood scanning can precisely capture the spatial fluctuation difference between the weak phase abrupt changes caused by deep defects and the smooth background caused by shallow continuous interfaces, thereby accurately locating and quantifying the local consistency of the phase map, eliminating the interference of isolated noise for subsequent accurate calculation of the stationarity index of the whole map.

[0012] Furthermore, the method for obtaining the shallow background phase reference value includes: classifying frequencies with a phase space stationarity index greater than or equal to a consistency threshold into the shallow dominant frequency set; and at each pixel location, calculating the average of the phase values ​​of all frequencies in the shallow dominant frequency set along the frequency axis to obtain the shallow background phase reference value of that pixel.

[0013] By setting a consistency threshold, the set of dominant shallow frequencies is automatically defined, and the average phase value of all frequencies in the set is calculated at the pixel level. Compared with the approach of directly using a single low-frequency phase to represent the shallow background, this cross-frequency mean calculation can more stably and realistically restore the inherent shallow structural background phase characteristics of each spatial location, avoiding the reference distortion caused by occasional heating disturbances at specific frequencies.

[0014] Furthermore, the shallow phase occlusion intensity needs to satisfy the following relationship: ;in, For pixels The shallow phase masking intensity at that location, For pixels The shallow background phase reference value at that location, In pixels The normalized standard deviation of the shallow background phase reference values ​​of all pixels within the centered window.

[0015] Based on the shallow background phase reference value, a local spatial normalized standard deviation is cleverly introduced for negative attenuation correction. This allows for the assignment of high occlusion weights to areas with uniform structure and strong reflection to increase subsequent stripping force, while automatically reducing occlusion weights in areas with large phase fluctuations and potential defect information. Compared to using a fixed global coefficient, this achieves pixel-level adaptive stripping force control, effectively preventing over-stripping caused by misjudging deep defects as strong background.

[0016] Further, the stripped phase map is obtained, including: for each frequency in the deep dominant frequency set, the phase value of its original phase map at each pixel is subtracted from the product of the stripping coefficient and the shallow phase occlusion intensity to obtain the stripped phase map corresponding to that frequency; the stripping coefficient is adaptively determined by minimizing the spatial standard deviation of the stripped phase map within a known defect-free reference area.

[0017] This method uses mathematical analysis to precisely remove the residual shallow strong reflection components that permeate the deep frequency map, ensuring that the deep phase map after stripping retains only the pure defect anomaly response, thus maintaining the integrity of the original deep features to the greatest extent.

[0018] Furthermore, the method for obtaining the defect response intensity includes: for each frequency in the stripped phase map set, taking each pixel as the center and with a side length of... The difference between the maximum and mean values ​​of the phase values ​​after stripping within the window is taken as the defect response intensity at that pixel location at that frequency.

[0019] Compared to directly using the original pixel absolute value, this method can effectively filter out the influence of residual low-frequency background drift, and keenly and dynamically highlight the phase change amplitude of the defect core area relative to the surrounding normal area, providing a highly representative differentiated weight index for subsequent multi-frequency fusion.

[0020] Furthermore, the fusion weights are based on the defect response intensity of each frequency at each pixel location, including: summing and normalizing the defect response intensity of all frequencies in the deep dominant frequency set at each pixel location based on the defect response intensity, to obtain the fusion weights of each frequency at each pixel location.

[0021] Furthermore, it also includes preprocessing the thermal imaging sequence: the preprocessing specifically includes performing bad pixel repair on each frame of the thermal imaging sequence, performing mean filtering on the thermal imaging sequence along the time axis, and performing spatial alignment correction on each frame of the thermal imaging sequence.

[0022] Furthermore, the method for applying pulsed thermal excitation includes: applying it by a symmetrically arranged array of halogen lamps; dividing the tire under inspection into several detection sections along the circumference, with each detection section independently completing pulsed thermal excitation and thermal imaging sequence acquisition, and then stitching together the acquisition results of each section to obtain full-circumference thermal imaging data covering the entire circumference of the tire.

[0023] Beneficial effects: By constructing a phase space stability index, the dominant frequencies of shallow and deep layers are objectively divided; the shallow spatial stability is quantified pixel by pixel to extract adaptive occlusion intensity, thereby quantitatively stripping shallow penetration interference from the deep phase map; finally, deep features are dynamically weighted and fused based on the defect response intensity of each frequency. This invention completely breaks through the limitations of fixed-frequency phase analysis, greatly enhances the detection contrast of weak deep defects, and provides a reliable guarantee for the quality of new energy tires at the factory and for driving safety. Attached Figure Description

[0024] Figure 1 This is a schematic flowchart illustrating a defect detection method according to an embodiment of the present invention. Detailed Implementation

[0025] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0026] like Figure 1 As shown, S1: Data acquisition and preprocessing.

[0027] In this embodiment, the new energy vehicle tire to be inspected is fixed on the inspection fixture, with the tire tread facing the excitation source and the infrared thermal imager. Two symmetrically arranged halogen lamp arrays are used as the active excitation source, applying a pulsed thermal excitation to the tire surface for a duration of 3 to 5 seconds. Immediately after the excitation ends, the infrared thermal imager is activated for continuous acquisition. The resolution of the infrared thermal imager is no less than 640×480 pixels, the acquisition frame rate is set to 25 frames / second, and the acquisition duration is 60 seconds, resulting in a thermal imaging sequence containing 1500 frames of temperature images. Each pixel value in each frame of the thermal imaging sequence corresponds to the surface radiation temperature at the corresponding spatial location at the corresponding time. To cover the inspection requirements of the entire tire circumference, the tire is divided into several inspection segments along the circumference. Each segment independently completes the above excitation and acquisition process, and finally, the data is stitched together to obtain complete full-circumference thermal imaging data.

[0028] The following preprocessing operations were performed sequentially on the acquired thermal imaging sequence. First, bad pixel repair was performed on each frame of the thermal imaging sequence, replacing pixels with abnormal responses with the mean of the effective pixels in their spatial neighborhood, thus eliminating fixed noise introduced by defects in the thermal imager's own detection unit. Second, mean filtering was performed along the time axis of the thermal imaging sequence, smoothing the temporal temperature curve of each pixel using a sliding window of five consecutive frames to suppress random thermal noise during acquisition. Third, spatial alignment correction was performed on each frame of the thermal imaging sequence, using the excitation start frame as a reference, and image registration methods were employed to eliminate inter-frame displacement deviations introduced by tire tool vibration or slight camera jitter, ensuring that the same pixel coordinates always correspond to the same physical location on the tire surface throughout the entire time series. After preprocessing, the preprocessed thermal imaging sequence was obtained, serving as input for subsequent frequency domain analysis.

[0029] S2: Construct the phase space stationarity index.

[0030] In this embodiment, the new energy tire is composed of multiple heterogeneous materials such as cord layer, steel wire layer, carcass rubber, and tread rubber. The thermal conductivity and heat capacity of each layer of materials are significantly different. During the process of the heat wave generated by the active pulse excitation propagating into the interior of the tire, it will be reflected at the interface of each material layer. After the reflected heat wave returns to the surface, it is superimposed on the time-domain temperature curve of the same pixel. After performing a discrete Fourier transform on the time-domain temperature curve, the phase components of different frequencies actually contain the superposition contribution from interfaces of different depths.

[0031] In the high-frequency band, the thermal wave penetration depth is relatively shallow, and the reflection component corresponding to the shallow interface dominates in the high-frequency phase map. Since the shallow interface is a continuously distributed structural interface throughout the entire tire detection area, its corresponding phase response spatially exhibits a large-scale, smoothly transitioning background phase distribution, meaning that the phase value difference between adjacent pixels is small, and the spatial consistency is strong. In contrast, internal defects exist only in local areas, and the phase anomalies they cause spatially manifest as local abrupt changes relative to the background phase, with a larger phase difference between adjacent pixels and weaker spatial consistency.

[0032] For example: Assume the thermal wave frequency corresponding to the interface between the tread rubber and the carcass rubber is... ,exist In the corresponding phase maps, the phase difference between adjacent pixels in defect-free areas is only on the order of 0.02 radians, while the phase difference between adjacent pixels at deep bubble defect locations can reach the order of 0.15 radians. By constructing an index reflecting the spatial consistency of phase maps at each frequency, we can effectively distinguish between frequency ranges dominated by background phase and those dominated by defect phase, providing a quantitative basis for subsequent layer-by-layer stripping of shallow occlusions. If this index is not constructed and stripping is directly performed on phase maps of all frequencies, it will be impossible to distinguish which frequency ranges' phase distribution belongs to the shallow interface background and which belongs to the deep defect response. The stripping operation will lose its targeting, and the phase enhancement effect of deep defects cannot be guaranteed.

[0033] Perform a discrete Fourier transform on each pixel along the time axis of the preprocessed thermal imaging sequence in step S1 to obtain the frequency range of The set of phase diagrams, denoted as ,in For frequency The corresponding phase diagram, Indicates spatial location In frequency The phase values ​​below. For each phase map, with a side length of... A sliding window of pixels is used to traverse the entire phase map. Within each window, the standard deviation of the phase values ​​of all pixels within the window is calculated, and the local phase standard deviation corresponding to the center pixel of that window is obtained. After traversing all windows, the frequency is obtained. Corresponding local phase standard deviation plot This reflects the spatial fluctuation of the phase value within the neighborhood of that location. In this embodiment, the window side length... The value can be 5, 7 or 9.

[0034] For frequency Corresponding local phase standard deviation plot Calculate the mean of the local phase standard deviation of all pixels in the entire image to obtain the frequency. Corresponding phase space stationarity index : ;in, For frequency The mean of the corresponding local phase standard deviation map across all pixels in the entire image. The maximum value in the local phase standard deviation plot corresponding to all frequencies is used for normalization. Calculate the phase space stationarity index for all frequencies within the range. This yields a uniformity curve with frequency as the independent variable. .

[0035] The shallow interface background phase is spatially continuous and smooth, corresponding to the local phase standard deviation. The size is generally smaller across the entire map. Smaller, phase space stationarity index A value approaching 1 indicates that the phase pattern at this frequency is dominated by the background phase. Within the frequency range corresponding to deep defects, the local phase abrupt changes caused by the defects affect the corresponding local phase standard deviation. The defect location is significantly enlarged. Increase accordingly A decrease indicates a strong local phase anomaly in the phase diagram at this frequency. This can be determined using the phase space stationarity index. This method can quantitatively distinguish between the frequency range dominated by the shallow interface background and the frequency range containing the response of deep defects, providing an accurate basis for the extraction of shallow phase occlusion intensity in the following steps, and avoiding the loss of deep defect information due to performing the stripping operation in the wrong frequency range.

[0036] S3: Construct shallow phase occlusion intensity.

[0037] In this embodiment, during the analysis in step S2, the frequency range dominated by the shallow interface background phase has been identified using the phase space stationarity index. However, simply knowing which frequency ranges are dominated by the shallow background is not enough; it is also necessary to further quantify the specific degree of occlusion of the deep phase map by the shallow background phase at each pixel location in order to perform targeted pixel-by-pixel stripping of the deep phase map in the following step S4.

[0038] In the multi-layer composite structure of new energy tires, the thermal wave reflection intensity at the shallow interface is not uniformly distributed spatially. The material layer thicknesses differ between the sidewall and crown areas, and the distribution density of the steel wire layer varies locally in the circumferential direction. These structural factors lead to differences in the thermal wave reflection capability of the shallow interface at different spatial locations. At locations with strong reflectivity, the shallow phase component accounts for a larger proportion in the frequency domain decomposition of the pixel's time-domain temperature curve, resulting in a stronger degree of masking for the phase of deeper defects. Conversely, at locations with weak reflectivity, the shallow phase component accounts for a smaller proportion, resulting in weaker masking.

[0039] For example, in areas with a dense distribution of steel wire layers, the thermal impedance difference between the steel wire and rubber interface is large, and the amplitude of the shallow thermal wave reflection component can be 3 to 5 times that of the deep defect reflection component. The corresponding shallow phase occlusion intensity is significantly higher than in areas with sparse steel wires. If a uniform fixed occlusion intensity is applied to the entire image for stripping, insufficient stripping in dense steel wire areas will result in deep defects still being occluded; conversely, excessive stripping in sparse steel wire areas will over-reduced normal deep phase information, introducing new false detections. Therefore, it is necessary to construct a pixel-by-pixel shallow phase occlusion intensity to provide spatially adaptive occlusion weights for the stripping operation in step S4.

[0040] Based on the phase space stationarity index obtained in step S2 Set a consistency threshold , will satisfy The set of frequencies is denoted as the shallow dominant frequency set. Not satisfied The set of frequencies is denoted as the deep dominant frequency set, where, The value is 0.75. At each pixel location, for example, at pixel location... Above, the average phase value of all frequencies within the shallow dominant frequency set is calculated along the frequency axis to obtain the pixel. Shallow background phase reference value : ;in, shallow dominant frequency set The number of frequencies contained therein For frequency The corresponding phase map at the pixel position The phase value at each pixel location. Above, the intensity of shallow phase occlusion depends not only on the shallow background phase reference value. The absolute magnitude of the occlusion strength also depends on the local spatial stability of the shallow phase component at that location within its local neighborhood. Local spatial stability refers to the uniformity of the shallow background phase reference value within the neighborhood of that pixel. Stronger spatial stability indicates a purer shallow background phase at that pixel location, resulting in more definitive occlusion of the deeper phase. Weaker spatial stability indicates that the pixel location has been mixed with some deep defect phase components, and directly using this value as the occlusion weight would lead to over-stripping. Therefore, the shallow phase occlusion strength is constructed as follows: ;in, For pixels The shallow phase masking intensity at that location, For pixels The shallow background phase reference value at that location, In pixels Centered on, with side length as The normalized standard deviation of the shallow background phase reference values ​​of all pixels within the window is calculated by dividing by the mean of the shallow background phase reference values ​​within the window. This standard deviation reflects the spatial stability of the shallow phase components at that location. The window side length is... The value is consistent with that of step S2.

[0041] In areas with dense steel wire layers and strong shallow thermal wave reflection, the shallow background phase reference value is relatively large, and the spatial stability of the shallow phase distribution in this region is relatively strong, with a normalized standard deviation. The shallow background phase reference value is relatively small, and the shallow phase occlusion intensity is relatively large. Therefore, a stronger stripping weight is applied to this location in subsequent step S4. In areas with relatively uniform shallow structures and weak reflection, the shallow background phase reference value is small, and the shallow phase occlusion intensity is small, resulting in a weaker stripping weight being applied subsequently. When a location has been mixed with phase components from deep defects, the local phase space stability decreases, and the normalized standard deviation... The shallow phase occlusion intensity is automatically reduced to avoid misjudging deep defect phases as shallow background and over-peeling. The shallow phase occlusion intensity achieves pixel-by-pixel spatial adaptive occlusion quantization, laying the foundation for the precise peeling operation in step S4.

[0042] S4: Construct a deep frequency adaptive fusion index.

[0043] In this embodiment, after constructing the shallow phase shielding intensity in step S3, it is necessary to further utilize the phase diagrams of each frequency in the deep dominant frequency set to identify deep defects. However, the deep dominant frequency set contains multiple frequency components, each frequency corresponding to a different thermal wave penetration depth. Defects at different depths exhibit different levels of response intensity in the phase diagrams of different frequencies.

[0044] The depth distribution of internal defects in new energy tires is not fixed. Defects such as bubbles and delamination may exist at different depths, such as between the ply and the carcass, near the steel wire layer, or inside the carcass. For defects at shallower depths, their phase response is more significant in the higher-frequency deep phase diagram; for defects at deeper depths, their phase response is more significant in the lower-frequency deep phase diagram. If the phase diagrams after stripping all frequencies from the dominant deep frequency set are directly averaged, the high-frequency phase response corresponding to shallow-depth defects and the low-frequency phase response corresponding to deep-depth defects will be diluted, resulting in a decrease in the detection sensitivity for defects at different depths.

[0045] For example: Suppose that a shallow defect with a depth of 2 mm and a deep defect with a depth of 6 mm exist simultaneously inside the tire. The shallow defect has a higher frequency of occurrence than the deep defect. The corresponding phase anomaly amplitude in the stripped phase diagram is 0.20 radians, occurring at lower deep frequencies. The amplitude in the corresponding stripped phase diagram is only 0.06 radians; deep defects occur at lower deep frequencies. The corresponding phase anomaly amplitude in the stripped phase diagram is 0.18 radians, occurring at higher deep frequencies. The amplitude in the corresponding stripped phase map is only 0.05 radians. If the average is taken directly, the phase anomaly amplitudes of both defects are suppressed to about 0.13 radians, resulting in decreased defect contrast. Therefore, it is necessary to construct an adaptive weighted fusion index based on the defect response intensity in the phase map of each deep frequency, so that the phase map of each frequency is weighted according to its contribution to the defect response at the current spatial location, thereby preserving the phase anomalies of defects at different depths in the fusion result.

[0046] For each frequency in the deep dominant frequency set Shallow phase shading intensity obtained based on S3 Phase diagram corresponding to this frequency Perform pixel-by-pixel stripping to obtain the frequency. Corresponding stripped phase diagram : ;in, For frequency The corresponding original phase map at the pixel position Phase value at that point, pixel position The shallow phase masking intensity, The stripping coefficient is adaptively determined by minimizing the spatial standard deviation of the stripped phase map within a known defect-free reference region. The stripping operation is performed on all frequencies in the deep dominant frequency set to obtain the stripped phase map set.

[0047] For each frequency in the stripped phase map set , per pixel Centered on, with side length as The difference between the maximum and mean phase values ​​after stripping within the window is used as the frequency. In this pixel Defect response intensity at the location Among them, the window side length The value is 7. Based on the defect response intensity, all frequencies in the deep dominant frequency set are analyzed at the pixel level. The frequency is obtained by summing and normalizing the defect response intensities at the specified locations. In pixels Fusion weights at the point Based on fusion weights By performing a weighted summation on the phase maps of all frequencies stripped from the deep dominant frequency set, the deep frequency adaptive fusion index is obtained: ;in, For pixels Deep frequency adaptive fusion index at the location For frequency In pixels The fusion weight at the point, For frequency The corresponding stripped phase map at the pixel Phase value at that point, It is a set of deep dominant frequencies.

[0048] At locations with shallow-depth defects, the defect response intensity of the post-peel phase diagram corresponding to higher depth frequencies at that location is... Larger, corresponding to fusion weight At higher frequencies, the phase value after stripping dominates the fusion result. In locations with deep defects, lower deep frequencies correspond to stronger defect response intensities and thus higher fusion weights; the phase value after stripping at these frequencies dominates the fusion result. In defect-free normal regions, the phase values ​​after stripping at all frequencies approach zero after shallow masking stripping, indicating relatively low defect response intensity. Both are relatively small, and the deep frequency adaptive fusion index is relatively small. Maintain at a low level. Deep frequency adaptive fusion index. An adaptive weighted integration of the responses to defects at different depths was achieved, enabling both shallow and deep defects to obtain significant phase anomaly contrast in the same fused image, providing high-quality input covering the entire depth range for defect identification in step S5.

[0049] S5: Defect detection and result output.

[0050] The standard pulse phase thermal imaging algorithm performs a pixel-by-pixel discrete Fourier transform on the thermal imaging sequence and directly selects the phase map corresponding to a fixed frequency as the defect criterion. This approach has significant limitations in the multi-layer composite structure of new energy tires: the shallow background phase components generated at the interfaces of each material layer and the phase response of deep defects are superimposed in the frequency domain, and the fixed-frequency phase map cannot separate the two, resulting in contrast distortion of deep defects.

[0051] The core improvement of the standard pulse phase thermal imaging algorithm in this embodiment lies in replacing the static output of the fixed-frequency phase map with a frequency-adaptive layering stripping and multi-frequency fusion mechanism based on spatial consistency. The specific steps are as follows: Step 1: After obtaining the phase maps of each frequency through Fourier transform, instead of directly selecting a fixed frequency, the spatial consistency of each frequency phase map is calculated to adaptively divide all frequencies into a shallow dominant frequency set and a deep dominant frequency set, replacing manual frequency selection with the spatial distribution pattern of the data itself. Step 2: In the phase map corresponding to the shallow dominant frequency set, the shallow background phase reference for each pixel is extracted and corrected with local spatial stability weights to obtain the pixel-by-pixel shallow phase occlusion intensity. Then, using the shallow phase occlusion intensity as a weight, pixel-by-pixel stripping is performed on the phase map of each frequency in the deep dominant frequency set, removing the residual penetration component of the shallow interface background phase from the deep phase map pixel by pixel. This step is not present in the standard algorithm, fundamentally solving the problem of shallow background occlusion of the phase response of deep defects. Step 3: For the multiple deep frequency phase maps after stripping, adaptive weighted fusion is performed with the defect response intensity of each frequency at each pixel position as the fusion weight to obtain the deep frequency adaptive fusion index. This ensures that the phase response of defects at different depths is fully preserved in the same fused map, overcoming the limitation of the standard algorithm that a single frequency phase map can only reflect defects at a specific depth.

[0052] Based on the deep frequency adaptive fusion index, an adaptive threshold segmentation method is used to identify defect regions in the fusion result image. The mean of all pixels within the entire image range of the deep frequency adaptive fusion index, plus three times the standard deviation, is used as the segmentation threshold. Pixels exceeding the segmentation threshold are marked as candidate defect pixels. Connectivity analysis is performed on the candidate defect pixels. Connectivity regions with an area less than 20 pixels are identified as noise points and removed. Connectivity regions with an area of ​​20 pixels or more are retained as valid defect regions. For each valid defect region, the spatial location and extent of the defect are marked with its bounding rectangle, and the maximum value of the deep frequency adaptive fusion index within the region is recorded as a quantitative representation of the defect confidence. A higher defect confidence indicates a more significant phase anomaly caused by the defect at that location. The location, extent, and defect confidence of all valid defect regions are superimposed on the original thermal imaging frame, outputting the final defect detection result image. This enables spatial localization and quantitative characterization of internal defects in the molding quality of new energy tires, providing a reliable basis for tire factory quality assessment.

Claims

1. A method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging, characterized in that, include: S1. Obtain the thermal imaging sequence of the new energy tire under inspection after applying pulsed thermal excitation; S2. Perform discrete Fourier transform on each pixel of the thermal imaging sequence to obtain the phase map of each frequency. Calculate the local phase standard deviation of each phase map using a sliding window, and construct a phase spatial stability index that reflects the spatial consistency of the phase maps at each frequency. S3. Based on the phase spatial stability index, divide all frequencies into a shallow dominant frequency set and a deep dominant frequency set. Extract the shallow background phase reference value of the phase values ​​within the shallow dominant frequency set pixel by pixel, and correct it with local spatial stability weights to obtain the shallow phase occlusion intensity of each pixel. Local spatial stability reflects the uniformity of the shallow background phase reference value in the neighborhood of that pixel. S4. Perform pixel-by-pixel stripping on the phase map of each frequency in the deep dominant frequency set using the shallow phase occlusion intensity as the weight to obtain the stripped phase map. Then, use the defect response intensity of each frequency at each pixel position as the fusion weight to perform weighted fusion on the stripped phase map to obtain the deep frequency adaptive fusion index. S5. Perform adaptive threshold segmentation and connected component analysis on the deep frequency adaptive fusion index to output the spatial location of the defect region and the defect confidence level.

2. The method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging according to claim 1, characterized in that, The formula for calculating the phase space stationarity index includes: ； in, For frequency The corresponding phase space stationarity index, For frequency The mean of the corresponding local phase standard deviation map across all pixels in the entire image. This represents the maximum value in the local phase standard deviation plot corresponding to all frequencies.

3. The method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging according to claim 2, characterized in that, frequency The methods for obtaining the corresponding local phase standard deviation plot include: For frequency The corresponding phase diagram, with side length as A sliding window of pixels is used to traverse the entire phase map. Within each window, the standard deviation of the phase values ​​of all pixels within the window is calculated, and the local phase standard deviation corresponding to the center pixel of that window is obtained. After traversing all windows, the frequency is obtained. The corresponding local phase standard deviation plot.

4. The method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging according to claim 1, characterized in that, Methods for obtaining shallow background phase reference values ​​include: Frequency groups with a phase space stability index greater than or equal to the consistency threshold are classified into the shallow dominant frequency set; and at each pixel location, the average phase value of all frequencies in the shallow dominant frequency set is calculated along the frequency axis to obtain the shallow background phase reference value of that pixel.

5. The method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging according to claim 1, characterized in that, The shallow phase masking intensity needs to satisfy the following relationship: ； in, For pixels The shallow phase masking intensity at that location, For pixels The shallow background phase reference value at that location, In pixels The normalized standard deviation of the shallow background phase reference values ​​of all pixels within the centered window, the normalized standard deviation corresponding to the pixel Local spatial stability.

6. The method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging according to claim 1, characterized in that, The stripped phase map is obtained, including: For each frequency in the deep dominant frequency set, the phase value of its original phase map at each pixel is subtracted from the product of the stripping coefficient and the shallow phase occlusion intensity to obtain the stripped phase map corresponding to that frequency; wherein, the stripping coefficient is adaptively determined by minimizing the spatial standard deviation of the stripped phase map within a known defect-free reference region.

7. The method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging according to claim 1, characterized in that, Methods for obtaining defect response intensity include: For each frequency in the stripped phase map set, with each pixel as the center and a side length of... The difference between the maximum and mean values ​​of the phase values ​​after stripping within the window is taken as the defect response intensity at that pixel location at that frequency.

8. The method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging according to claim 1, characterized in that, The fusion weights are based on the defect response intensity at each pixel location for each frequency, including: Based on the defect response intensity, the defect response intensity of all frequencies in the deep dominant frequency set at each pixel is summed and normalized to obtain the fusion weight of each frequency at each pixel position.

9. The method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging according to claim 1, characterized in that, This also includes preprocessing the thermal imaging sequences: The preprocessing specifically includes: performing bad pixel repair on each frame of the thermal imaging sequence, performing mean filtering on the thermal imaging sequence along the time axis, and performing spatial alignment correction on each frame of the thermal imaging sequence.

10. The method for detecting defects in the molding quality of new energy tires based on active infrared thermal imaging according to claim 1, characterized in that, The method of applying pulsed thermal excitation includes: applying it by a symmetrically arranged array of halogen lamps; dividing the tire under test into several detection sections along the circumference, with each detection section independently completing pulsed thermal excitation and thermal imaging sequence acquisition, and then stitching the acquisition results of each section together to obtain full-circumference thermal imaging data covering the entire circumference of the tire.

Images