New energy automobile battery module laser welding quality intelligent detection system

Abstract

The invention belongs to the technical field of welding quality detection, and discloses an intelligent detection system for laser welding quality of a battery module of a new energy automobile. Comprising a data acquisition module for acquiring laser welding data in real time and cleaning the data; the welding compensation module is used for judging a welding stage and performing phase compensation based on a stage judgment result; the reflection elimination module is used for carrying out spectral response analysis and carrying out reflection interference elimination based on a spectral analysis result; the virtual joint area analysis module is used for carrying out three-dimensional section analysis, extracting fusion welding error geometric parameters, identifying a welding virtual joint area and marking the welding virtual joint area; the false melting early warning module is used for shielding a false melting area to obtain false melting early warning information; the overall evaluation module is used for carrying out overall quality evaluation to obtain a welding quality evaluation record and sending the welding quality evaluation record to a preset control end; and the data credibility and the engineering adaptability of welding quality detection are improved.

Description

Technical Field

[0001] This invention relates to the field of welding quality inspection technology, and more specifically, to an intelligent inspection system for laser welding quality of new energy vehicle battery modules. Background Technology

[0002] As an important vehicle for global energy transition and carbon neutrality strategy, the manufacturing quality of the power battery module, a core component of new energy vehicles, directly determines the safety performance and service life of the entire vehicle. Among these, laser welding has been widely used as the mainstream technology in the connection process. Welding quality inspection, as a core link to ensure the reliability of battery modules, directly affects the safety and stability of battery modules. Currently, the traditional welding quality inspection methods commonly used in engineering practice can achieve conventional defect identification functions, but they are often limited in more complex welding scenarios and still face many technical bottlenecks.

[0003] In practical welding scenarios, oscillating laser welding is frequently used. During this process, the laser welding head decelerates to zero and moves in the opposite direction at the reversal point, causing a short-term collapse of the weld pool at that location. The depth data acquired by the OCT probe at this moment exhibits periodic instantaneous fluctuations. Traditional welding quality inspection methods lack the ability to perceive the relationship between weld depth and the motion path, easily misjudging this natural shallow weld phenomenon as insufficient edge weld depth, affecting the accuracy of the assessment. In addition, when scanning the weld contour, aluminum alloy or copper alloy parts often have colored oxide films forming on their surfaces at high temperatures. Certain colored bands diffusely reflect and suppress the main laser wavelength, causing traditional welding quality inspection methods to mistakenly identify this area as insufficient edge weld depth. The region is judged as a concave boundary, which leads to an overestimation of the weld width. At the same time, during the welding process, local gaps at the lap joint may form void weld holes. The weld depth reflected by the OCT signal may seem to meet the weld depth requirements, but in fact, there is an error. Traditional welding quality inspection methods lack the ability to identify the geometric features related to the hole shape structure and the sidewall taper. On the other hand, the metallurgical structure of the original welding position in the lap joint area of ​​the closed-loop weld is often destroyed due to secondary heating, forming a remelted cladding weld spot with a smooth surface but extremely poor internal connectivity. Traditional welding quality inspection methods often easily overlook the fact that this position has been melted twice, causing the quality judgment of the lap joint area to be unstable.

[0004] In view of this, the present invention proposes an intelligent inspection system for laser welding quality of new energy vehicle battery modules to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an intelligent inspection system for laser welding quality of new energy vehicle battery modules, comprising: The data acquisition module collects laser welding data in real time and performs data cleaning to obtain a multi-source monitoring dataset for welding. The welding compensation module determines the welding stage based on the welding multi-source monitoring dataset, and performs phase compensation on the welding multi-source monitoring dataset based on the stage determination results to obtain the commutation correction monitoring dataset. The reflection elimination module performs spectral response analysis on the commutation correction monitoring dataset, removes reflection interference based on the spectral analysis results, and outputs an accurate melt width dataset. The virtual weld area analysis module performs three-dimensional cross-sectional analysis based on the accurate weld width dataset to extract geometric parameters of weld error; it identifies and marks the virtual weld areas based on the geometric parameters of weld error, and generates a weld marking dataset. The false melt warning module masks the false melt area in the weld mark dataset to obtain false melt warning information; The overall assessment module combines the fusion welding mark dataset and false melting early warning information to conduct an overall quality assessment, obtain welding quality assessment records, and send them to the preset control terminal; the various modules are connected to each other via wired and / or wireless means.

[0006] Furthermore, the method for determining the welding stage includes: The welding multi-source monitoring dataset is grouped based on a preset sampling period to obtain a periodic welding data subset; the welding path coordinate sequence is extracted from the periodic welding data subset. Calculate the motion direction angle of adjacent welding path coordinates in the welding path coordinate sequence, calculate the first derivative of the motion direction angle to obtain the angle change rate, and draw the angle change rate curve based on the angle change rate; identify the minimum point of the angle change rate curve, and calculate the slope difference between adjacent points before and after the point; if the minimum value is less than the preset change threshold and the slope difference is higher than the preset slope threshold, then the minimum point is regarded as a candidate reversal node. Extract the melt depth data within the preset time window before and after the candidate reversal node, and calculate the average rate of change of the melt depth data; at the same time, identify the welding direction of the corresponding melt depth data segment. If the average rate of change is higher than the fluctuation threshold and the welding direction is reversed in the corresponding time segment, then determine that the time segment corresponding to the current candidate reversal node is the swing reversal stage.

[0007] Furthermore, the method for performing phase compensation includes: Identify the welding speed of each sampling point during the swing reversal phase; if the welding speed of a sampling point is less than the set reversal threshold and the welding direction before and after the sampling point is reversed, then the sampling point is determined to be a reversal point, and a local response window is constructed based on the reversal point as the center; Calculate the maximum difference in the intensity of the reflected signal within the local response window. If the maximum difference is greater than a preset ratio, then the reversing point corresponding to the window is determined to be a collapse interference point. Calculate the mean melt depth data within the local response window in adjacent time intervals, and obtain the mean before and mean after respectively; Calculate the relative position ratio of the window corresponding to the collapse interference point during the swing reversal stage, and obtain the melt depth phase compensation value by segmenting and weighting the front mean and the back mean based on the relative position ratio. The melt depth data corresponding to the collapse interference point is replaced by the melt depth phase compensation value, and the corrected melt depth data and other data are integrated to obtain the reversal correction monitoring dataset.

[0008] Furthermore, the method for performing spectral response analysis includes: Extract structured light welding image data located in the weld area from the reversal correction monitoring dataset, and filter the structured grayscale image data of multiple wavelength channels; Statistically analyze the grayscale values ​​of different wavelength channels in each structural grayscale image, and calculate the pixel grayscale ratio of adjacent wavelength channels in the same structural grayscale image. Calculate the slope of change of grayscale ratio of consecutive pixels and identify pixel regions in the structured grayscale image where the slope of change shows a continuous downward trend; if the average grayscale ratio of pixels in the pixel region is lower than the preset reflection imbalance threshold, the region is determined to be a spectral response attenuation region. Extract post-weld temperature data, and determine whether the post-weld temperature of the spectral response attenuation region is higher than the preset temperature standard based on the post-weld temperature data. If it is higher than the preset temperature standard, the corresponding spectral response attenuation region is identified as the oxidation interference region, which is the spectral analysis result.

[0009] Furthermore, the method for eliminating reflection interference includes: Identify continuous pixel segments where the oxidation interference area overlaps with the weld area, count the gray values ​​of these continuous pixel segments, and plot a horizontal gray-scale gradient curve based on the gray values. Calculate the gradient change rate of the horizontal gray-scale gradient change curve. If the gradient change rate is lower than the preset abrupt change threshold, the area where the corresponding continuous pixel segment is located is determined to be an edge error area. Extract the coordinate sequence of the weld centerline in the normal area, match and fit the edge error area based on the coordinate sequence of the weld centerline, and output a reasonable offset interval; based on the reasonable offset interval, fit and compensate the horizontal gray-level gradient change curve, output the compensated curve, and use the gray-level value corresponding to the compensated curve to cover the gray-level value of the original continuous pixel segment to obtain the compensated pixel segment. Edge recognition is performed again on the compensated pixel segment to obtain the edge contour of the compensated weld; the lateral weld width value is calculated based on the edge contour of the compensated weld; the weld width value is fitted to the corresponding oxidation interference area based on the lateral weld width value, and the weld width value of the adjusted area is output. The remaining data are integrated to obtain the accurate weld width dataset.

[0010] Furthermore, the method for performing three-dimensional cross-sectional analysis includes: Based on the welding path coordinate sequence, the accurate weld width dataset is divided into equidistant sampling segments, and the weld depth data, weld width data, and structured light welding image data are extracted from the equidistant sampling segments. Based on the weld depth and weld width data, a cross-sectional coordinate system is constructed for each equidistant sampling segment. The contour map of the weld cross-section is generated by combining the grayscale information of the weld area in the structured light welding image data. Geometric analysis of the contour diagram yields a set of welding error parameters, including weld width, weld depth, sidewall taper, bottom closure, and weld symmetry deviation rate.

[0011] Furthermore, the method for identifying the weld defect area includes: Read the welding error set parameters corresponding to each equidistant sampling segment in sequence, calculate the ratio of maximum weld depth to maximum weld width, and obtain the welding aspect ratio; The average taper is obtained by calculating the average taper of the sidewalls on both sides of the weld section; at the same time, the minimum radius of curvature is fitted based on the contour diagram of the weld section. Construct judgment conditions and set corresponding thresholds. Judgment conditions include: the weld aspect ratio is less than the corresponding ratio threshold, the average taper is greater than the corresponding threshold, the bottom closure is less than the corresponding closure threshold, the weld symmetry deviation rate is higher than the corresponding ratio threshold, and the minimum radius of curvature is less than the corresponding curvature threshold. If the equidistant sampling segment meets at least three judgment conditions, the corresponding equidistant sampling segment is judged as a virtual region; if there are consecutive adjacent virtual regions or consecutive adjacent equidistant sampling segments meet any two judgment conditions, and the melting depth change rate is lower than the preset melting depth change threshold, the region formed by the corresponding equidistant sampling segment is judged as a continuous virtual segment. The coordinates of individual virtual bonding regions and continuous virtual bonding segments are marked, and all data are integrated to obtain the welding mark dataset.

[0012] Furthermore, the method for shielding the virtual melting region includes: Extract the welding path coordinate sequence from the fusion welding mark dataset, identify the coordinates of the lap section in the weld area, and integrate the lap section coordinates into a process-specific segment coordinate set; Read the welding timestamps corresponding to the coordinate set of the special process segment, and extract the laser welding power and welding speed of the corresponding welding timestamps; calculate the thermal energy of the special process segment based on the laser welding power and welding speed, and output the thermal energy density; at the same time, obtain the average heat input value of the special process segment in the previous sampling period. If the thermal energy density is higher than the average heat input value of the preset heat ratio, then the special process segment is regarded as the residual heat segment. Extract the structured light welding image data corresponding to the residual heat section for curvature analysis and judgment, and output the potential remelting section; when the potential remelting section overlaps with the marked virtual area or the structural parameter fluctuation of the potential remelting section is greater than the preset parameter fluctuation threshold, the corresponding potential remelting section is marked as a virtual melting area and shielded, and virtual melting warning information is output.

[0013] Furthermore, the method for performing curvature analysis and judgment includes: Calculate the structural parameters of the structured light welding image data corresponding to the residual heat section. If the change angle of the boundary section unfolding angle is greater than the unfolding angle threshold, the average gray value of the lower boundary is less than the preset gray value ratio of the average gray value of the other areas, and the curvature change rate of the upper boundary is higher than the preset fluctuation threshold, then the corresponding residual heat section is determined to be a potential remelting section.

[0014] Furthermore, the methods for conducting the overall quality assessment include: Obtain the segment number and welding path coordinates of the centrally marked fusion welding mark dataset, compare the specific parameters of the corresponding segment with historical normal welding records to determine the risk segment; count the proportion of risk segments in the welding path, and output the welding quality level in order of welding path coordinates; Welding quality assessment records are obtained by integrating the section number of the risk section, the coordinates of the welding path, the welding quality level, and the corresponding specific parameters.

[0015] The technical effects and advantages of the intelligent inspection system for laser welding quality of new energy vehicle battery modules of this invention: By real-time acquisition and data cleaning of multi-source monitoring data during the welding process, an intelligent inspection system for laser welding quality of new energy vehicle battery modules was established. Based on the cleaned monitoring data, the system successively completed functions such as weld depth phase compensation, spectral reflection interference removal, weld width correction, weld malfunction identification, and malmelting warning. Compared with existing experience, by identifying the abrupt change nodes of laser motion during welding reversal, and combining the angle change rate with the welding speed to determine the reversal stage, a local response window was constructed and phase compensation was performed, avoiding misjudgment of shallow welds caused by reversal. Oxidation interference areas were identified by spectral response analysis of multi-band structured light images, and contour fitting and restoration were performed by combining edge grayscale curves, realizing the elimination of reflection errors of actual weld width. A parameter set was constructed using the weld cross-sectional contour to identify malmelting and continuous malmelting. Residual heat analysis and remelting structure identification were performed by combining the lap area, shielding the erroneously judged malmelting areas, improving the reliability of detection, and enhancing the data credibility and engineering adaptability of welding quality inspection. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the intelligent inspection system for laser welding quality of new energy vehicle battery modules according to the present invention; Figure 2This is a schematic diagram of the intelligent detection method for laser welding quality of new energy vehicle battery modules according to the present invention. Detailed Implementation

[0017] 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.

[0018] Example 1 Please see Figure 1 As shown in this embodiment, the intelligent inspection system for laser welding quality of new energy vehicle battery modules includes: The data acquisition module collects laser welding data in real time and performs data cleaning to obtain a multi-source monitoring dataset for welding. The welding compensation module determines the welding stage based on the welding multi-source monitoring dataset, and performs phase compensation on the welding multi-source monitoring dataset based on the stage determination results to obtain the commutation correction monitoring dataset. The reflection elimination module performs spectral response analysis on the commutation correction monitoring dataset, removes reflection interference based on the spectral analysis results, and outputs an accurate melt width dataset. The virtual weld area analysis module performs three-dimensional cross-sectional analysis based on the accurate weld width dataset to extract geometric parameters of weld error; it identifies and marks the virtual weld areas based on the geometric parameters of weld error, and generates a weld marking dataset. The false melt warning module masks the false melt area in the weld mark dataset to obtain false melt warning information; The overall assessment module combines the fusion welding mark dataset and false melting early warning information to conduct an overall quality assessment, obtain welding quality assessment records, and send them to the preset control terminal; the various modules are connected to each other via wired and / or wireless means.

[0019] In this embodiment, the laser welding data consists of multi-source data, including structured light welding image data collected by cameras at the industrial site, parameters reflecting the welding process such as laser welding speed and power collected by sensors, and parameters reflecting the degree of welding such as weld penetration and weld width. By filling missing values ​​in the parameter type data and filtering the image type data, a higher quality welding multi-source monitoring dataset is obtained.

[0020] Methods for determining the welding stage include: The welding multi-source monitoring dataset is grouped based on a preset sampling period to obtain a periodic welding data subset. The preset sampling period is a period length set based on known welding process specifications. Each sampling period corresponds to a complete laser welding process. The welding multi-source monitoring dataset is divided based on this sampling period to obtain a periodic welding data subset corresponding to each sampling period.

[0021] Extract the welding path coordinate sequence from the periodic welding data subset, where the welding path coordinate sequence represents the coordinate change sequence of the welding process in a unified spatial coordinate system during a certain sampling period, and each coordinate point is bound to the corresponding timestamp.

[0022] Calculate the motion direction angle of adjacent welding path coordinates in the welding path coordinate sequence, calculate the first derivative of the motion direction angle to obtain the angle change rate, and plot the angle change rate curve based on the angle change rate. Specifically, the motion direction angle is obtained by calculating the vector change of adjacent welding path coordinates, and the first derivative of the sequence composed of all motion direction angles is calculated to obtain the angle change rate of adjacent positions. The angle change rate curve is plotted based on the angle change rate of each welding path coordinate.

[0023] Identify the minimum point of the angle change rate curve and calculate the slope difference between adjacent points before and after that point. The minimum point indicates the location where a change of direction may occur. The slope difference between adjacent points before and after that point reflects the intensity of the change of direction.

[0024] If the minimum value is less than the preset change threshold and the slope difference is higher than the preset slope threshold, then the minimum value point is taken as a candidate reversal node. The preset change threshold and preset slope threshold are set based on historical judgment experience. If the minimum value is less than the preset change threshold, it means that a direction deflection has occurred. If the slope difference is higher than the preset slope threshold, it means that the direction deflection is severe. If the above two threshold conditions are met, the corresponding minimum value point is taken as a candidate reversal node.

[0025] Extract melt depth data within a preset time window before and after the candidate reversal node, and calculate the average rate of change of melt depth data. The length of the preset time window is set based on historical experience, and the average rate of change of melt depth parameters within the segment formed by the time window is calculated to reflect the magnitude of melt depth change before and after reversal.

[0026] Simultaneously, the welding direction of the corresponding penetration depth data segment is identified. If the average rate of change is higher than the fluctuation threshold and the welding direction reverses within the corresponding time segment, the time segment corresponding to the current candidate reversal node is determined to be the swing reversal stage. Specifically, by identifying the welding direction of laser welding within the penetration depth data segment corresponding to the time window, and comparing the average rate of change with the fluctuation threshold set based on historical experience, if the average rate of change is large and the welding direction reverses, it indicates that the penetration depth change is unstable while the laser welding head has moved in the opposite direction. Therefore, the time segment corresponding to the candidate reversal node is determined to be the swing reversal stage.

[0027] Methods for performing phase compensation include: The welding speed of each sampling point during the swing reversal phase is identified. If the welding speed of a sampling point is less than the set reversal threshold and the welding direction before and after the sampling point is reversed, the sampling point is determined to be a reversal point. At the same time, a local response window is constructed based on the reversal point as the center. The welding speed is introduced to further accurately locate the specific reversal point during the swing reversal phase. The reversal threshold is set to 0. The welding speed, position and direction are combined to determine whether the sampling point is decelerating to 0. If it decelerates to 0 and the direction is reversed, the specific reversal point can be determined. The window is constructed to both sides of the reversal point as the center. The window size is set based on historical experience to ensure that the subsequent compensation range is sufficiently covered.

[0028] The maximum difference in reflected signal intensity within a local response window is calculated. If this maximum difference is greater than a preset ratio, the commutation point corresponding to that window is determined as a collapse interference point. The reflected signal intensity refers to the intensity of the OCT reflected signal, which is easily affected by changes in the molten pool morphology. The difference between the maximum and minimum reflected signal intensity within the local response window is calculated, and a preset ratio is set based on historical normal signal fluctuations. If this difference is higher than the preset ratio, it indicates that the corresponding local response window is affected by laser reverse motion, and the corresponding commutation point is determined as a collapse interference point.

[0029] Calculate the mean melt depth data within the local response window in adjacent time intervals, obtaining the front mean and the back mean respectively. The front mean represents the mean melt depth before the switch, and the back mean represents the mean melt depth after the switch. These two means are used as input data for subsequent calculations.

[0030] Calculate the relative position ratio of the window corresponding to the collapse interference point during the swing reversal phase. Based on this relative position ratio, perform piecewise weighting on the previous and subsequent mean values ​​to obtain the melt depth phase compensation value. The calculation formula for piecewise weighting is as follows: ;in This represents the phase compensation value of the melt depth at the point of collapse interference; This indicates the relative position ratio of the window corresponding to the collapse interference point during the swing reversal phase; the value range is... ; This represents the previous mean; The mean value is represented by the value after the change of direction. The influence of the change of direction on different positions is quantified by the ratio of the relative positions, and the melt depth phase compensation value calculated by this method can better fit the data trend.

[0031] The melt depth data corresponding to the collapse interference point is replaced by the melt depth phase compensation value, and the corrected melt depth data and other data are integrated to obtain the reversal correction monitoring dataset.

[0032] Methods for performing spectral response analysis include: Structured light welding image data located in the weld area of ​​the reversal correction monitoring dataset is extracted, and structured grayscale image data of multiple wavelength channels are filtered. These multiple wavelength channels include the main wavelength channel and the auxiliary wavelength channel, which reflect the response of the weld area to illumination at different wavelengths. The location of the weld area is determined by the original welding path plan, and the image data of the corresponding area is extracted.

[0033] The grayscale values ​​of different wavelength channels in each structural grayscale image are statistically analyzed, and the pixel grayscale ratio of adjacent wavelength channels in the same structural grayscale image is calculated. The grayscale values ​​of different wavelength channels represent the reflectivity of a location to a specific wavelength of laser light. The comparison of grayscale values ​​of multiple wavelength channels can reflect whether there is excessive absorption behavior of a single wavelength in the region. The pixel grayscale ratio is obtained by summarizing the grayscale values ​​of adjacent channels in an image and performing a ratio calculation.

[0034] The slope of the change in the gray-scale ratio of consecutive pixels is calculated to identify pixel regions in the structured gray-scale image where the slope of change shows a continuous downward trend. The pixel gray-scale ratios are sorted according to the wavelength channel order, and the rate of change of the gray-scale ratio of consecutive pixels is calculated as the slope of change. By traversing the image, regions where the slope of change is downward, that is, regions where the gray-scale ratio is continuously decreasing, are identified.

[0035] If the average pixel grayscale ratio of a pixel region is lower than a preset reflection imbalance threshold, the region is determined to be a spectral response attenuation region. This is achieved by calculating the average pixel grayscale ratio of the pixel region and comparing it with a preset reflection imbalance threshold set based on historical experience. If the average pixel grayscale ratio is low, it indicates that the imaging brightness of the region is reduced, and therefore the pixel region is determined to be a spectral response attenuation region.

[0036] Post-weld temperature data is extracted, and the temperature of the spectral response attenuation region is determined to be higher than the preset temperature standard. If it is higher than the preset temperature standard, the corresponding spectral response attenuation region is identified as an oxidation interference region, which is the spectral analysis result. In this process, the temperature data of the spectral response attenuation region after welding is extracted simultaneously and compared with the preset temperature standard set based on the welding process specification. If the post-weld temperature is higher than the temperature standard, it indicates that the spectral response attenuation region is in a high-temperature state after welding. Therefore, it remains at a high temperature, and the spectral response attenuation indicates that the abnormality in this region is caused by the high-temperature oxide film. Therefore, the spectral response attenuation region is identified as an oxidation interference region and used as the spectral analysis result of the spectral response analysis.

[0037] Methods for eliminating reflection interference include: Identify continuous pixel segments that overlap with the oxidation interference area and the weld area, count the gray values ​​of these continuous pixel segments, and plot a horizontal gray-level gradient curve based on the gray values. Specifically, compare the coordinates of the oxidation interference area with the coordinates of the weld area, and sequentially select overlapping pixels along the welding direction to form continuous pixel segments. Count the gray values ​​of these continuous pixel segments and plot a horizontal gray-level gradient curve in the same direction as the welding direction, which serves as the data basis for subsequent operations.

[0038] Calculate the gradient change rate of the horizontal gray-level gradient curve. If the gradient change rate is lower than the preset abrupt change threshold, the area where the corresponding continuous pixel segment is located is determined to be an edge error area. The smaller the gradient change rate, the smoother the boundary change. Gray-level attenuation may come from reflection non-geometric changes. Based on historical interference removal experience, a preset abrupt change threshold is set. If the gradient change rate is lower than the preset abrupt change threshold, the area where the corresponding continuous pixel segment is located is determined to be an edge error area.

[0039] The coordinate sequence of the weld centerline in the normal region is extracted. Based on the coordinate sequence of the weld centerline, the edge error region is matched and fitted to output a reasonable offset interval. The weld centerline structure of the normal region in the historical normal welding record is extracted as the data basis. The matching and fitting refers to using the weld centerline as the reference to fit the left and right boundary regions respectively using a polynomial fitting algorithm, and extending the fitting function to the edge error region to restore the boundary shape of the region. The predicted result of the boundary offset is obtained, which is the reasonable offset interval.

[0040] Based on a reasonable offset range, the horizontal gray-level gradient change curve is fitted and compensated, and a compensated curve is output. The gray-level values ​​corresponding to the compensated curve are used to cover the gray-level values ​​of the original continuous pixel segments to obtain compensated pixel segments. The reasonable offset range is used as the gray-level error range between the edge error region and the normal weld contour. In this embodiment, a local weighted regression algorithm is used to process the original horizontal gray-level gradient change curve to construct a new compensated curve, which is used to reconstruct the gray-level changes that should exist within the real weld contour. The gray-level estimated values ​​of each pixel in the compensated curve are used to replace the gray-level values ​​of the corresponding pixels in the original continuous pixel segments to form compensated pixel segments.

[0041] The edge recognition of the compensated pixel segment is re-performed to obtain the edge contour of the compensated weld. Specifically, the edge recognition of the area where the compensated pixel segment is located in the compensated image is re-performed using an edge detection algorithm to obtain the edge contour of the compensated weld after eliminating visual interference.

[0042] The transverse weld width value is calculated based on the compensated weld edge profile. By using the projection distance of the edge point pairs in the compensated weld edge profile in the weld main axis direction, the transverse weld width is calculated, ensuring that the left and right boundary identification points come from real reflection and are not misled by oxidation, thus improving the weld width accuracy.

[0043] The melt width value is fitted to the corresponding oxidation interference region based on the lateral melt width value, and the melt width value of the adjusted region is output. The remaining data are integrated to obtain the melt width accurate dataset. The melt width value of the corresponding oxidation interference region is replaced by the lateral melt width value, and the replaced melt width data is integrated with other data such as image type and other parameters to form the melt width accurate dataset.

[0044] Methods for performing three-dimensional cross-sectional analysis include: Based on the welding path coordinate sequence, the accurate weld width dataset is divided into equidistant sampling segments. Weld depth data, weld width data, and structured light welding image data are extracted from the equidistant sampling segments. Specifically, based on the welding path coordinate sequence, the sampling points are matched with the accurate weld width dataset and divided according to the equidistant rule to obtain equidistant sampling segments. Weld depth data, weld width data, and structured light welding image data are extracted from the data of each equidistant sampling segment as the data basis for subsequent operations.

[0045] Based on the weld penetration and weld width data, a cross-sectional coordinate system is constructed for each equidistant sampling segment. A contour map of the weld cross-section is generated by combining the grayscale information of the weld area in the structured light welding image data. Within each sampling segment, a local two-dimensional cross-sectional coordinate system is constructed according to the welding path direction and its perpendicular direction. The weld penetration is used as the longitudinal axis of the cross-section, and the weld width as the lateral distance. Combining the grayscale spatial distribution of the weld area in the structured light welding image data, a polynomial fitting algorithm is used to construct the curve contour map of the cross-section. It should be noted that the unit length of the cross-sectional coordinate system is on the same order of magnitude as the unified spatial coordinate system; this coordinate system is only used to represent a two-dimensional cross-section in a local space.

[0046] Geometric analysis of the profile diagram yields a set of welding error parameters, including weld width, weld depth, sidewall taper, bottom closure, and weld symmetry deviation rate. Edge points, bottom concave points, and sidewall lines on both sides of the weld are extracted from the generated cross-sectional profile diagram. The relevant geometric parameters are calculated. Weld width represents the lateral spread range, weld depth represents the longitudinal fusion depth, sidewall taper represents the rate of change of the wall angle, bottom closure indicates whether the two sidewalls merge at the bottom, and weld symmetry deviation rate reflects the degree of left and right deformation of the cross-section.

[0047] Methods for identifying areas of poor weld bonding include: The welding error set parameters corresponding to each equidistant sampling segment are read sequentially, and the ratio of the maximum weld depth to the maximum weld width is calculated to obtain the welding aspect ratio. The welding error set parameters of each equidistant sampling segment are read separately, and the welding aspect ratio is obtained by calculating the ratio of the maximum weld depth to the maximum weld width. The larger the value, the thinner and longer the weld. If it is too small, it may indicate insufficient penetration. This is used as one of the subsequent judgment conditions.

[0048] The average taper is obtained by calculating the average value of the sidewall taper on both sides of the weld section. The sidewall taper reflects the angle change of the weld as it gradually shrinks from the upper surface to the inner side. A high average taper indicates that the weld bead expands outwards and shrinks inwards more steeply.

[0049] Simultaneously, the minimum radius of curvature is fitted based on the contour diagram of the weld section. The minimum radius of curvature is used to characterize the region of the greatest abrupt change in the curve contour of the weld. The curve is fitted downwards along the section contour to form a continuous curve, and the radius of curvature value at the sharpest bend in the curve is the minimum radius of curvature.

[0050] Judgment conditions are constructed and corresponding thresholds are set. The judgment conditions include: the weld aspect ratio is less than the corresponding ratio threshold, the average taper is greater than the corresponding threshold, the bottom closure is less than the corresponding closure threshold, the weld symmetry deviation rate is higher than the corresponding proportion threshold, and the minimum radius of curvature is less than the corresponding curvature threshold. The corresponding thresholds are all set based on historical welding recognition experience. Judgment conditions are set by combining the weld aspect ratio, average taper, bottom closure, weld symmetry deviation rate, and minimum radius of curvature. This reduces the misjudgment rate of judgment based on a single indicator and improves recognition accuracy.

[0051] If an equidistant sampling section meets at least three judgment conditions, the corresponding equidistant sampling section is determined to be a false weld area. In this embodiment, the equidistant sampling section needs to meet at least three judgment conditions to be determined as a false weld area. Among them, if the weld aspect ratio is significantly smaller, it indicates that the molten pool expands laterally but the longitudinal fusion is insufficient. In this case, it is easy to cause the upper and lower layers of false weld. Excessive taper means that the molten pool has not fully expanded laterally, and the energy distribution is concentrated and cannot be evenly diffused. Small bottom closure indicates that the top is well closed but the bottom is cracked, which is a typical feature of false weld. Large weld symmetry deviation rate indicates that the heat is mainly input on one side and the other side has not formed effective fusion, which is one of the factors that cause false weld. The minimum radius of curvature is too small, which shows sharp abrupt changes at the bottom or side wall, usually indicating that the local melting is insufficient or there is an abnormal forming situation.

[0052] If there are consecutive adjacent virtual merged regions or consecutive adjacent equidistant sampling segments that meet any two of the judgment conditions, and the melt depth change rate is lower than the preset melt depth change threshold, then the region formed by the corresponding equidistant sampling segments is judged as a consecutive virtual merged segment. If there are consecutive virtual merged regions, it can be judged as a consecutive virtual merged segment. Alternatively, if consecutive adjacent equidistant sampling segments meet any two of the judgment conditions, and the melt depth change rate is also lower than the melt depth change threshold set based on historical experience, it indicates that the melt depth remains stable but there is still a structural anomaly. Therefore, this situation can also be judged as a consecutive virtual merged segment.

[0053] The coordinates of individual virtual bonding regions and continuous virtual bonding segments are marked, and all data are integrated to obtain the welding mark dataset.

[0054] Methods for shielding virtual fusion regions include: Extract the welding path coordinate sequence from the fusion welding mark dataset, identify the coordinates of the lap section in the weld area, and integrate the coordinates of the lap section into a process-specific segment coordinate set. The coordinate position of the lap section in the weld area is determined based on the original welding path plan, and the coordinates corresponding to the lap section are integrated into the process-specific segment coordinate set.

[0055] Read the welding timestamps corresponding to the coordinate set of the special process segment, and extract the laser welding power and welding speed of the corresponding welding timestamp. In this process, by constructing an index relationship between coordinates and timestamps, the time segment corresponding to the special process segment is determined, and the welding power and welding speed of the time segment are extracted as the data basis for subsequent calculations.

[0056] The thermal energy of a special section of this process is calculated based on laser welding power and welding speed, and the heat energy density is output. The formula for the thermal energy calculation is as follows: ;in This represents the heat input per unit length in the output process-specific section, i.e., the thermal energy density. Indicates welding power; Indicates the welding speed.

[0057] Simultaneously, the average heat input value of the special process section in the previous sampling period is obtained. If the heat energy density is higher than the average heat input value of the preset heat ratio, the special process section is regarded as the waste heat section. The average heat input value of the special process section in the previous sampling period is calculated, which is the average heat energy density of the entire section. If the current heat energy density is higher than the average heat input value of the heat ratio set based on historical experience, it means that the current special process section has been irradiated by laser and is very likely in the secondary melting stage. Therefore, the special process section is regarded as the waste heat section.

[0058] The structured light welding image data corresponding to the residual heat section is extracted for curvature analysis and judgment, and potential remelting sections are output. In particular, by analyzing the geometric features of the structured light welding image data corresponding to the residual heat section, the geometric deformation phenomenon caused by high heat input is identified, and potential remelting sections are further screened after judgment.

[0059] When a potential remelting segment overlaps with a marked virtual merging area, or when the structural parameter fluctuation of the potential remelting segment exceeds a preset parameter fluctuation threshold, the corresponding potential remelting segment is marked as a virtual merging area and masked, and a virtual merging warning message is output. In this embodiment, two judgment conditions are given: first, the potential remelting segment overlaps with the previously marked virtual merging area, indicating that the segment may suffer metallurgical damage due to overlapping remelting; second, the change in various geometric feature structural parameters in the potential remelting segment exceeds the parameter fluctuation threshold set based on normal industrial specifications, indicating that the structure in the segment is quite unstable. If either or both judgment conditions are met, the corresponding potential remelting segment is determined to be a virtual merging area, and a masking tag is added to the area to construct the virtual merging warning message.

[0060] Methods for performing curvature analysis include: The structural parameters of the structured light welding image data corresponding to the residual heat section are calculated. If the change angle of the boundary section unfolding angle is greater than the unfolding angle threshold, the average gray value of the lower boundary is less than the preset gray value ratio of the average gray value of the other areas, and the curvature change rate of the upper boundary is higher than the preset fluctuation threshold, then the corresponding residual heat section is identified as a potential remelting section. Three geometric structural characteristic parameters are calculated: the boundary section unfolding angle, the average gray value of the lower boundary, and the curvature change rate of the upper boundary. The corresponding thresholds are set based on normal industrial specifications. A large boundary section unfolding angle indicates that the weld pool expands unevenly after heating. A small average gray value of the lower boundary indicates that the structured light reflection is weakened, and there is a metallurgical structure collapse or ablation phenomenon. A high curvature change rate of the upper boundary indicates that the boundary shape is abnormal due to high-heat remelting.

[0061] Methods for conducting overall quality assessments include: Obtain the segment number and welding path coordinates marked in the fusion welding mark dataset. Combine the specific parameters of the corresponding segment with historical normal welding records to determine risk segments. Specifically, obtain the segment number and corresponding welding path coordinates of the segment with added tags in the fusion welding mark dataset after previous processing to identify the location of the corresponding segment. Compare the relevant geometric parameters and parameters reflecting the welding process in the segment with historical normal welding records to screen risk segments based on historical experience.

[0062] The proportion of risky sections in the welding path is statistically analyzed, and the welding quality level is output in sequence according to the welding path coordinates. Specifically, the proportion of risky sections in the welding path for each sampling period is statistically analyzed. This value, along with the corresponding risky section label and related specific parameters, is used as training data for the risk assessment model. The risk assessment model is then used to output the welding quality level of each risky section in sequence according to the welding path coordinates. In this embodiment, the support vector machine model is used as the basic structure of the risk assessment model.

[0063] The welding quality assessment record is obtained by integrating the section number, welding path coordinates, welding quality grade and corresponding specific parameters of the risk section. The section number, welding path coordinates, welding quality grade and related geometric parameters and parameters reflecting the welding process of each risk section are integrated and transformed into a log format that can be recognized by a computer.

[0064] This embodiment establishes an intelligent laser welding quality inspection system for new energy vehicle battery modules by real-time acquisition and data cleaning of multi-source monitoring data during the welding process. Based on the cleaned monitoring data, it successively completes penetration depth phase compensation, spectral reflection interference removal, weld width correction, weld malfunction identification, and malmelting warning. Compared with existing experience, by identifying the laser motion abrupt change node during welding reversal, and combining the angle change rate with the welding speed to determine the reversal stage, a local response window is constructed and phase compensation is performed, avoiding misjudgment of shallow welds caused by reversal. Oxidation interference areas are identified by spectral response analysis of multi-band structured light images, and contour fitting restoration is performed by combining edge grayscale curves to achieve reflection error removal of actual weld width. A parameter set is constructed using the weld cross-sectional contour to identify malmelting and continuous malmelting. Residual heat analysis and remelting structure identification are performed by combining the lap area, shielding the erroneously judged malmelting areas, improving detection reliability, and enhancing the data credibility and engineering adaptability of welding quality inspection.

[0065] Example 2 Please see Figure 2 As shown, for parts not described in detail in this embodiment, please refer to the description in Embodiment 1. A smart inspection method for laser welding quality of new energy vehicle battery modules is provided, including: S1. Real-time acquisition of laser welding data and data cleaning to obtain a multi-source monitoring dataset for welding; S2. Based on the welding multi-source monitoring dataset, the welding stage is determined, and the phase compensation is performed on the welding multi-source monitoring dataset based on the stage determination results to obtain the commutation correction monitoring dataset; S3. Perform spectral response analysis on the commutation correction monitoring dataset, remove reflection interference based on the spectral analysis results, and output an accurate melting width dataset. S4. Perform three-dimensional cross-sectional analysis based on the accurate weld width dataset to extract the geometric parameters of weld error; identify and mark the weld misalignment area based on the geometric parameters of weld error to generate a weld marking dataset; S5. Mask the virtual weld area in the weld mark dataset to obtain virtual weld early warning information; S6. Combine the fusion welding mark dataset and false melting early warning information to conduct an overall quality assessment, obtain welding quality assessment records, and send them to the preset control terminal.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0067] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0068] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An intelligent inspection system for laser welding quality of new energy vehicle battery modules, characterized in that, include: The data acquisition module collects laser welding data in real time and performs data cleaning to obtain a multi-source monitoring dataset for welding. The welding compensation module determines the welding stage based on the welding multi-source monitoring dataset, and performs phase compensation on the welding multi-source monitoring dataset based on the stage determination results to obtain the commutation correction monitoring dataset. The reflection elimination module performs spectral response analysis on the commutation correction monitoring dataset, removes reflection interference based on the spectral analysis results, and outputs an accurate melt width dataset. The virtual weld area analysis module performs three-dimensional cross-sectional analysis based on the accurate weld width dataset to extract geometric parameters of weld error. Based on the geometric parameters of fusion welding error, the welded false joint area is identified and marked to generate a fusion welding mark dataset; The false melt warning module masks the false melt area in the weld mark dataset to obtain false melt warning information; The overall assessment module combines the fusion welding mark dataset and false melting early warning information to conduct an overall quality assessment, obtain welding quality assessment records, and send them to the preset control terminal; the various modules are connected to each other via wired and / or wireless means.

2. The intelligent inspection system for laser welding quality of new energy vehicle battery modules according to claim 1, characterized in that, The methods for determining the welding stage include: The welding multi-source monitoring dataset is grouped based on a preset sampling period to obtain a periodic welding data subset; the welding path coordinate sequence is extracted from the periodic welding data subset. Calculate the motion direction angle of adjacent welding path coordinates in the welding path coordinate sequence, calculate the first derivative of the motion direction angle to obtain the angle change rate, and draw the angle change rate curve based on the angle change rate; identify the minimum point of the angle change rate curve, and calculate the slope difference between adjacent points before and after the point; if the minimum value is less than the preset change threshold and the slope difference is higher than the preset slope threshold, then the minimum point is regarded as a candidate reversal node. Extract the melt depth data within the preset time window before and after the candidate reversal node, and calculate the average rate of change of the melt depth data; at the same time, identify the welding direction of the corresponding melt depth data segment. If the average rate of change is higher than the fluctuation threshold and the welding direction is reversed in the corresponding time segment, then determine that the time segment corresponding to the current candidate reversal node is the swing reversal stage.

3. The intelligent inspection system for laser welding quality of new energy vehicle battery modules according to claim 2, characterized in that, The methods for performing phase compensation include: Identify the welding speed of each sampling point during the swing reversal phase; if the welding speed of a sampling point is less than the set reversal threshold and the welding direction before and after the sampling point is reversed, then the sampling point is determined to be a reversal point, and a local response window is constructed based on the reversal point as the center; Calculate the maximum difference in the intensity of the reflected signal within the local response window. If the maximum difference is greater than a preset ratio, then the reversing point corresponding to the window is determined to be a collapse interference point. Calculate the mean melt depth data within the local response window in adjacent time intervals, and obtain the mean before and mean after respectively; Calculate the relative position ratio of the window corresponding to the collapse interference point during the swing reversal stage, and obtain the melt depth phase compensation value by segmenting and weighting the front mean and the back mean based on the relative position ratio. The melt depth data corresponding to the collapse interference point is replaced by the melt depth phase compensation value, and the corrected melt depth data and other data are integrated to obtain the reversal correction monitoring dataset.

4. The intelligent inspection system for laser welding quality of new energy vehicle battery modules according to claim 3, characterized in that, The methods for performing spectral response analysis include: Extract structured light welding image data located in the weld area from the reversal correction monitoring dataset, and filter the structured grayscale image data of multiple wavelength channels; Statistically analyze the grayscale values ​​of different wavelength channels in each structural grayscale image, and calculate the pixel grayscale ratio of adjacent wavelength channels in the same structural grayscale image. Calculate the slope of change of grayscale ratio of consecutive pixels and identify pixel regions in the structured grayscale image where the slope of change shows a continuous downward trend; if the average grayscale ratio of pixels in the pixel region is lower than the preset reflection imbalance threshold, the region is determined to be a spectral response attenuation region. Extract post-weld temperature data, and determine whether the post-weld temperature of the spectral response attenuation region is higher than the preset temperature standard based on the post-weld temperature data. If it is higher than the preset temperature standard, the corresponding spectral response attenuation region is identified as the oxidation interference region, which is the spectral analysis result.

5. The intelligent inspection system for laser welding quality of new energy vehicle battery modules according to claim 4, characterized in that, The methods for eliminating reflection interference include: Identify continuous pixel segments where the oxidation interference area overlaps with the weld area, count the gray values ​​of these continuous pixel segments, and plot a horizontal gray-scale gradient curve based on the gray values. Calculate the gradient change rate of the horizontal gray-scale gradient change curve. If the gradient change rate is lower than the preset abrupt change threshold, the area where the corresponding continuous pixel segment is located is determined to be an edge error area. Extract the coordinate sequence of the weld centerline in the normal area, match and fit the edge error area based on the coordinate sequence of the weld centerline, and output a reasonable offset interval; based on the reasonable offset interval, fit and compensate the horizontal gray-level gradient change curve, output the compensated curve, and use the gray-level value corresponding to the compensated curve to cover the gray-level value of the original continuous pixel segment to obtain the compensated pixel segment. Edge recognition is performed again on the compensated pixel segment to obtain the edge contour of the compensated weld; the lateral weld width value is calculated based on the edge contour of the compensated weld; the weld width value is fitted to the corresponding oxidation interference area based on the lateral weld width value, and the weld width value of the adjusted area is output. The remaining data are integrated to obtain the accurate weld width dataset.

6. The intelligent inspection system for laser welding quality of new energy vehicle battery modules according to claim 5, characterized in that, The methods for performing three-dimensional cross-sectional analysis include: Based on the welding path coordinate sequence, the accurate weld width dataset is divided into equidistant sampling segments, and the weld depth data, weld width data, and structured light welding image data are extracted from the equidistant sampling segments. Based on the weld depth and weld width data, a cross-sectional coordinate system is constructed for each equidistant sampling segment. The contour map of the weld cross-section is generated by combining the grayscale information of the weld area in the structured light welding image data. Geometric analysis of the contour diagram yields a set of welding error parameters, including weld width, weld depth, sidewall taper, bottom closure, and weld symmetry deviation rate.

7. The intelligent inspection system for laser welding quality of new energy vehicle battery modules according to claim 6, characterized in that, The methods for identifying weld defects include: Read the welding error set parameters corresponding to each equidistant sampling segment in sequence, calculate the ratio of maximum weld depth to maximum weld width, and obtain the welding aspect ratio; The average taper is obtained by calculating the average taper of the sidewalls on both sides of the weld section; at the same time, the minimum radius of curvature is fitted based on the contour diagram of the weld section. Construct judgment conditions and set corresponding thresholds. Judgment conditions include: the weld aspect ratio is less than the corresponding ratio threshold, the average taper is greater than the corresponding threshold, the bottom closure is less than the corresponding closure threshold, the weld symmetry deviation rate is higher than the corresponding ratio threshold, and the minimum radius of curvature is less than the corresponding curvature threshold. If the equidistant sampling segment meets at least three judgment conditions, the corresponding equidistant sampling segment is judged as a virtual region; if there are consecutive adjacent virtual regions or consecutive adjacent equidistant sampling segments meet any two judgment conditions, and the melting depth change rate is lower than the preset melting depth change threshold, the region formed by the corresponding equidistant sampling segment is judged as a continuous virtual segment. The coordinates of individual virtual bonding regions and continuous virtual bonding segments are marked, and all data are integrated to obtain the welding mark dataset.

8. The intelligent inspection system for laser welding quality of new energy vehicle battery modules according to claim 7, characterized in that, The methods for shielding the virtual melting region include: Extract the welding path coordinate sequence from the fusion welding mark dataset, identify the coordinates of the lap section in the weld area, and integrate the lap section coordinates into a process-specific segment coordinate set; Read the welding timestamps corresponding to the coordinate set of the special process segment, and extract the laser welding power and welding speed of the corresponding welding timestamps; calculate the thermal energy of the special process segment based on the laser welding power and welding speed, and output the thermal energy density; at the same time, obtain the average heat input value of the special process segment in the previous sampling period. If the thermal energy density is higher than the average heat input value of the preset heat ratio, then the special process segment is regarded as the residual heat segment. Extract the structured light welding image data corresponding to the residual heat section for curvature analysis and judgment, and output the potential remelting section; when the potential remelting section overlaps with the marked virtual area or the structural parameter fluctuation of the potential remelting section is greater than the preset parameter fluctuation threshold, the corresponding potential remelting section is marked as a virtual melting area and shielded, and virtual melting warning information is output.

9. The intelligent inspection system for laser welding quality of new energy vehicle battery modules according to claim 8, characterized in that, The methods for performing curvature analysis and judgment include: Calculate the structural parameters of the structured light welding image data corresponding to the residual heat section. If the change angle of the boundary section unfolding angle is greater than the unfolding angle threshold, the average gray value of the lower boundary is less than the preset gray value ratio of the average gray value of the other areas, and the curvature change rate of the upper boundary is higher than the preset fluctuation threshold, then the corresponding residual heat section is determined to be a potential remelting section.

10. The intelligent inspection system for laser welding quality of new energy vehicle battery modules according to claim 9, characterized in that, The methods for conducting overall quality assessment include: Obtain the segment number and welding path coordinates of the centrally marked fusion welding mark dataset, compare the specific parameters of the corresponding segment with historical normal welding records to determine the risk segment; count the proportion of risk segments in the welding path, and output the welding quality level in order of welding path coordinates; Welding quality assessment records are obtained by integrating the section number of the risk section, the coordinates of the welding path, the welding quality level, and the corresponding specific parameters.

Images