Image big data driven process optimization method for crankshaft machining

Abstract

This invention relates to the field of manufacturing management technology, specifically to an image big data-driven process optimization method for crankshaft machining. The method includes the following steps: acquiring crankshaft reflection images using a bidirectional polarized light source and normalizing them to generate a dataset; calculating the phase delay difference using Fourier transform and compensating for angle to generate a defect distribution map; calculating defect density and gradient analysis to determine severity, clustering to identify types, and generating quality levels; combining cutting parameters with support vector machine analysis to generate adjustment schemes for parameters exceeding limits; solving for optimal parameters to select those satisfying roughness and tolerance thresholds, and generating optimization instructions. In this invention, crankshaft surface reflections are acquired using multiple polarized light sources and combined with synchronous acquisition of rotational position and time; frequency domain and phase delay difference are used to locate and compensate for minor defects; abnormal features are quantified into severity vectors; cluster analysis identifies patterns to generate quality levels; machining parameters are correlated to select key variables; and adjustment schemes are provided in real time, achieving closed-loop control of defect monitoring and process optimization.

Description

Technical Field

[0001] This invention relates to the field of manufacturing management technology, and in particular to a process optimization method driven by image big data for crankshaft machining. Background Technology

[0002] The field of manufacturing management technology involves the systematic management and optimization of resources, processes, procedures, and quality in the manufacturing process. It covers core aspects such as production planning, processing technology design, equipment scheduling management, process data collection and analysis, quality monitoring, and production efficiency improvement. This field achieves controllability and consistency in the processing process through the informatization, digitalization, and intelligent management of the entire production process. Especially in the processing of mechanical parts such as crankshaft manufacturing, it is necessary to comprehensively consider multi-dimensional information such as material properties, process parameters, processing paths, equipment status, and test data, and to improve processes through data-driven management methods.

[0003] Among them, the traditional crankshaft machining process optimization method refers to the process of adjusting the machining process to meet the design requirements in order to address the machining accuracy and stability issues of the crankshaft, a key component, during the production process. This is achieved through methods such as analyzing machining process records based on human experience, using geometric dimension measurement data obtained from testing equipment, adjusting tool compensation parameters using statistical process control methods, and periodically optimizing feed rate and depth of cut based on machine tool load monitoring results.

[0004] Existing technologies in crankshaft machining and inspection mainly rely on indirect parameters such as dimensions and load monitoring, lacking real-time quantitative data acquisition of surface conditions. When defects are small or have complex spatial distribution characteristics, they are easily overlooked. Parameter adjustments rely heavily on periodic statistical trend analysis, making it difficult to identify and correct anomalies in a timely manner when they first appear. This leads to the continuous accumulation of problems in the machining of multiple parts, resulting in a concentrated increase in scrap. Furthermore, the location of anomalies is extensive and time-consuming, increasing the cost of investigation and adjustment, and significantly impacting production stability. Summary of the Invention

[0005] To address the shortcomings of existing crankshaft machining inspection technologies, which primarily rely on indirect parameters such as dimensions and load monitoring, lacking real-time quantitative data on surface conditions, and prone to overlooking minor defects or those with complex spatial distributions, and whose parameter adjustments depend on periodic statistical trend analysis, making it difficult to identify and correct anomalies as they first appear, leading to the accumulation of problems across multiple parts and a concentrated increase in scrap, as well as the large scope and time-consuming process of locating anomalies, thus increasing troubleshooting and adjustment costs and significantly impacting production stability, this invention provides an image-based big data-driven process optimization method for crankshaft machining, comprising the following steps:

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a process optimization method driven by image big data for crankshaft machining, comprising the following steps:

[0007] S1: Obtain crankshaft surface reflection image data through bidirectional polarized light source, collect light intensity distribution in different polarization directions, monitor pixel light intensity and spatial coordinates, detect speed encoder angular position signal, obtain timestamp and crankshaft rotation angle parameters, and perform light intensity normalization to generate crankshaft surface image dataset.

[0008] S2: Perform Fourier transform frequency domain analysis based on the crankshaft surface image dataset, calculate the phase information of the sampling points, calculate the phase delay difference dataset through phase difference calculation, call the crankshaft rotation angle parameter to perform angle compensation correction, and generate a surface defect distribution map.

[0009] S3: Calculate the spatial distribution density and gradient change rate of the defect area by calling the surface defect distribution map, calculate the phase delay difference gradient between adjacent points, analyze the defect severity vector by gradient magnitude and direction angle, identify the defect pattern type by cluster analysis, and generate the processing quality prediction level.

[0010] S4: Call the machining quality prediction level, combine it with the current cutting speed, feed rate, and depth of cut, input the machining quality prediction level into the support vector machine algorithm to perform parameter correlation analysis, identify parameters whose correlation coefficient exceeds the correlation threshold, and generate a process parameter adjustment plan.

[0011] As a further aspect of the present invention, the crankshaft surface image dataset includes a pixel grayscale matrix, spatial geometric features, and reflected light intensity characteristics; the surface defect distribution map includes defect spatial coordinates, defect area data, and defect shape features; the processing quality prediction level includes a level number, a quality score value, and a corresponding defect category; and the process parameter adjustment scheme includes target optimization parameters, constraint parameters, and allowable variation range.

[0012] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0013] S101: Obtain crankshaft surface reflection image data through bidirectional polarization light source, monitor pixel light intensity values ​​and spatial coordinates in different polarization directions, vectorize and combine the light intensity at the same pixel position, and call the combined light intensity vector to perform position mapping in the two-dimensional data domain to generate a pixel light intensity matrix.

[0014] S102: Call the pixel light intensity matrix and the angular position signal of the speed encoder, match the pixel light intensity of each row in the matrix with the corresponding crankshaft rotation angle parameter, integrate the timestamps of the matching records and serialize them into a multi-dimensional data table to obtain the light intensity angle time association table.

[0015] S103: Based on the light intensity values ​​in the light intensity angle time correlation table, calculate the light intensity range ratio for the set of pixels at the same angle, calculate the normalization coefficient based on the ratio, transform and aggregate the set of light intensity values ​​into two-dimensional image frames, and generate a crankshaft surface image dataset.

[0016] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0017] S201: Based on the crankshaft surface image dataset, perform a two-dimensional discrete Fourier transform operation on the pixel grayscale matrix of the sampling points, transform the two-dimensional grayscale distribution in the spatial domain into a frequency domain amplitude matrix and a frequency domain phase matrix, extract the phase values ​​and arrange them by index to generate the sampling point phase matrix;

[0018] S202: Call the sampling point phase matrix, perform differential operation on the phase values ​​of adjacent sampling points in the matrix, and form a phase difference matrix according to the index order of the phase difference results, keeping the correspondence between positive and negative difference values ​​and sampling point indices consistent, and generate a phase delay difference dataset.

[0019] S203: Based on the phase delay difference dataset, call the crankshaft rotation angle parameter, compensate and adjust the phase delay difference according to the rotation angle parameter, map the compensated and corrected phase delay difference to the corresponding sampling point spatial position, and generate a surface defect distribution map.

[0020] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0021] S301: Based on the spatial coordinates of pixels in the surface defect distribution map and the defect region identifier encoding, the phase delay parameters of adjacent pixels in the two-dimensional spatial axis are called, and the phase delay difference of adjacent pixel pairs is vector decomposed according to the horizontal and vertical directions to generate a phase gradient distribution sequence.

[0022] S302: Based on the gradient magnitude and direction angle information of the phase gradient distribution sequence, retrieve the changing trend of the gradient at adjacent positions and filter the direction vector set of the continuous path. Weighted aggregate the magnitude of each vector in the path and the distribution weight of its direction to obtain the defect severity vector set.

[0023] S303: Call the amplitude features and orientation distribution parameters of the defect severity vector set, use a similarity metric based on feature space distance to divide the vector set into categories, identify the defect pattern number corresponding to each category and map it to the matching quality level index, and generate the processing quality prediction level.

[0024] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0025] S401: Based on the machining quality prediction level, multi-dimensional process parameters are constructed by combining cutting speed, feed rate, and depth of cut. The original parameter values ​​are assembled into column vectors according to their corresponding dimensional positions. Vector inner product and variance operations are performed, and the covariance is merged into coupling strength data to generate a set of parameter coupling coefficients.

[0026] S402: Call the parameter coupling coefficient set, extract the coupling coefficients and compare them with the stored associated thresholds item by item, record the indexes with values ​​greater than the thresholds as parameter identifiers, and aggregate the identifiers to form an over-limit parameter identifier group;

[0027] S403: Locate the corresponding value in the multi-dimensional process parameter set according to the over-limit parameter identification group, calculate the adjustment range in combination with the processing quality prediction level reference range, serialize the difference between the adjustment range and the original value, and generate a process parameter adjustment scheme.

[0028] As a further aspect of the present invention, the parameter coupling coefficient refers to a dimensionless value obtained by normalizing the covariance between multidimensional process parameters.

[0029] The association threshold refers to the numerical limit of whether the coupling coefficient of the determination parameter reaches the preset coupling determination condition.

[0030] As a further aspect of the present invention, the method further includes step S5:

[0031] S5: Calculate a new combination of cutting speed, feed rate, and depth of cut based on the process parameter adjustment scheme, solve the optimized values ​​of process parameters through a multi-objective function, verify the processing equipment capability constraints, screen parameters that simultaneously meet the surface roughness threshold and dimensional tolerance threshold, and generate crankshaft processing optimization instructions;

[0032] The crankshaft machining process optimization instructions include optimizing cutting speed, optimizing feed rate, and optimizing depth of cut.

[0033] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0034] S501: Based on the process parameter adjustment scheme, the cutting speed, feed rate and depth of cut are formed into a multi-dimensional parameter sequence, and the process coefficients are calculated and combined one by one. The multi-objective function constraint calculation is called to eliminate non-convergent combinations and generate an initial parameter combination set.

[0035] S502: Call each set of cutting speed, feed rate, and depth of cut data in the initial parameter combination set, perform difference calculation and sign judgment on them with the power limit value and stroke boundary value in the processing equipment capacity constraint conditions, eliminate parameter combinations that exceed the limits, and obtain the feasible parameter set of the equipment;

[0036] S503: Based on the combination data of the feasible parameter set of the equipment, perform absolute difference calculation on the measured surface roughness value and the surface roughness threshold, and perform difference judgment on the measured size value and the size tolerance threshold, eliminate any combination whose difference exceeds the threshold, and generate crankshaft machining process optimization instructions.

[0037] As a further aspect of the present invention, the surface roughness threshold refers to a surface roughness limit value set according to product design documents and processing technology standards;

[0038] The dimensional tolerance threshold refers to the dimensional deviation limit set according to the product design documents and tolerance standards.

[0039] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0040] In this invention, the surface reflection distribution of the crankshaft is acquired by a multi-polarization direction light source and synchronously collected in combination with rotation position and time parameters. This reduces surface illumination interference and accurately matches the processing parts during the imaging stage. Spatial positioning and compensation correction of minute defects are achieved by calculating the frequency domain and phase delay difference. Abnormal features are quantified by defect distribution density and gradient changes to form a severity vector. Defect patterns are identified based on cluster analysis and a quality prediction level is generated. Multi-dimensional correlation is established by combining real-time processing parameters to quickly screen key variables affecting quality. This supports providing targeted adjustment solutions in real time during the production process and realizes real-time closed-loop control of defect monitoring and process optimization. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of the steps of the present invention;

[0043] Figure 2 This is a detailed schematic diagram of S1 of the present invention;

[0044] Figure 3 This is a detailed schematic diagram of S2 of the present invention;

[0045] Figure 4 This is a detailed schematic diagram of S3 of the present invention;

[0046] Figure 5 This is a detailed schematic diagram of S4 of the present invention;

[0047] Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0048] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0049] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0050] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0051] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0052] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0053] Please see Figure 1 This invention provides a process optimization method driven by image big data for crankshaft machining, comprising the following steps:

[0054] S1: Obtain crankshaft surface reflection image data through bidirectional polarized light source, collect light intensity distribution in different polarization directions, monitor pixel light intensity and spatial coordinates, detect speed encoder angular position signal, obtain timestamp and crankshaft rotation angle parameters, and perform light intensity normalization to generate crankshaft surface image dataset.

[0055] S2: Perform Fourier transform frequency domain analysis based on crankshaft surface image dataset, calculate phase information of sampling points, calculate phase delay difference dataset through phase difference operation, call crankshaft rotation angle parameter to perform angle compensation correction, and generate surface defect distribution map;

[0056] S3: Calculate the spatial distribution density and gradient change rate of the defect area by calling the surface defect distribution map, calculate the phase delay difference gradient between adjacent points, analyze the defect severity vector by gradient magnitude and direction angle, identify the defect pattern type by cluster analysis, and generate the processing quality prediction level.

[0057] S4: Call the machining quality prediction level, combine it with the current cutting speed, feed rate, and depth of cut, input the machining quality prediction level into the support vector machine algorithm to perform parameter correlation analysis, identify parameters whose correlation coefficient exceeds the correlation threshold, and generate a process parameter adjustment plan;

[0058] S5: Calculate a new combination of cutting speed, feed rate, and depth of cut based on the process parameter adjustment scheme, solve the optimized values ​​of process parameters through multi-objective functions, verify the processing equipment capability constraints, screen parameters that simultaneously meet the surface roughness threshold and dimensional tolerance threshold, and generate crankshaft processing optimization instructions;

[0059] The crankshaft surface image dataset includes a pixel grayscale matrix, spatial geometric features, and reflected light intensity characteristics. The surface defect distribution map includes defect spatial coordinates, defect area data, and defect shape features. The machining quality prediction level includes a level number, quality score value, and corresponding defect category. The process parameter adjustment scheme includes target optimization parameters, constraint parameters, and allowable variation range. The crankshaft machining process optimization instructions include optimizing cutting speed, optimizing feed rate, and optimizing depth of cut.

[0060] Please see Figure 2 The specific steps of S1 are as follows:

[0061] S101: Obtain crankshaft surface reflection image data through bidirectional polarization light source, monitor pixel light intensity values ​​and spatial coordinates in different polarization directions, vectorize and combine the light intensity at the same pixel position, and call the combined light intensity vector to perform position mapping in the two-dimensional data domain to generate a pixel light intensity matrix.

[0062] When acquiring crankshaft surface reflection image data using a bidirectional polarized light source, the light source is first set to mutually perpendicular polarization directions: one horizontal (0°) and the other vertical (90°). The light source is then fixed to the crankshaft machining area at a distance of approximately 300mm from the crankshaft surface on both sides using a mechanical bracket. The reflected light intensity values ​​of the same spatial pixel are then collected under both polarization directions. For example, at pixel position (x=120, y=250), the light intensity measured in the horizontal polarization direction is... The light intensity measured in the perpendicular polarization direction is Combine the two light intensity values ​​to form a light intensity vector. The same acquisition and combination operation is performed to traverse all pixels of the crankshaft reflection image. For example, 1920 pixels are monitored on the crankshaft surface. The light intensity values ​​of each pixel in both horizontal and vertical polarization directions need to be recorded and combined to form a two-dimensional data pair. After vectorization and combination, the pixel light intensity vectors are mapped according to spatial coordinates (x, y), that is, each pixel vector is placed into a matrix cell corresponding to its position in the image plane, constructing a preliminary two-dimensional pixel light intensity matrix. For example, the vector stored in the (120, 250)th cell of the matrix is ​​(145, 132), and the vector stored in the (500, 800)th cell of the matrix is ​​(138, 141). During this mapping process, it is necessary to cyclically scan each row of pixels in the image and fill the corresponding cells of the matrix with the measured light intensity values ​​in sequence. If the light intensity of a certain pixel exceeds the normal range (for example, the light intensity value should be within a certain range), the system will not perform a mapping. to If the value is between [a certain value] and the original measurement is recorded in the log table, it will be marked as abnormal. Abnormal data will be stored separately for re-detection. The final pixel intensity matrix is ​​the same size as the imaging pixel resolution, and each unit in the matrix corresponds to a dual-polarized light intensity combination value at a physical location on the crankshaft surface, forming the basic optical reflection data foundation for the crankshaft machining process. For example, in a crankshaft main journal machining monitoring, if the surface roughness increases due to tool wear, the difference between the horizontal and vertical polarized light intensities at that location will be greater than the set reference value. This difference can be visually reflected in a matrix by showing the range of such differences (e.g., greater than). This is an abnormal range, within... to This is a transitional interval, less than By marking the normal range, a clear data input can be provided for the data integration of subsequent processes, and finally the pixel light intensity matrix with completed spatial mapping can be obtained.

[0063] Table 1: Intensity Acquisition Table of Dual-Polarized Light Pixels on Crankshaft Surface

[0064]

[0065] As shown in Table 1, the table presents the light intensity measurements at different pixel positions on the crankshaft surface under the action of a bidirectional polarized light source. The table also includes status markers based on the light intensity difference and a set range, with "abnormal" indicating that the light intensity difference exceeds [a certain threshold]. The threshold setting range will be directly reflected in the pixel light intensity matrix and affect the subsequent data processing results.

[0066] S102: Call the pixel light intensity matrix and the angular position signal of the speed encoder, match the pixel light intensity of each row in the matrix with the corresponding crankshaft rotation angle parameter, integrate the timestamps of the matching records and serialize them into a multi-dimensional data table to obtain the light intensity angle time association table.

[0067] When calling the pixel intensity matrix and the angular position signal of the speed encoder, the intensity vector data of each pixel in the constructed pixel intensity matrix is ​​read first. For example, the intensity vector data of the pixel in the first row of the matrix is ​​read first. The row contains an array of light intensity in the horizontal direction. and vertical light intensity array Subsequently, the real-time rotation angle during crankshaft machining was acquired using a speed encoder. and the corresponding timestamp Each row of pixel data is matched one-to-one with the angle parameters output by the encoder, for example, in time... The angle recorded by the time encoder is At this point, the pixel light intensity vector in row 250 of the matrix is ​​taken and corresponds to the angle parameter, and time is added to this set of records. The encoder identifies the rows and then sequentially associates them with the acquired angle parameters according to the scan order. During execution, when the encoder outputs a set of continuous angle signals, for example... The matrix rows need to be categorized into corresponding intervals to ensure that data at each angle position can be mapped to one or more matrix rows. During the matching process, data validity checks also need to be performed. For example, if there is a sudden change in light intensity at the same angle position within a short period of time (such as the difference in light intensity of the same pixel at adjacent acquisition times),... > If the pixel coordinates of each matching record are not found in the data, then that time period is marked as an abnormal rotation segment. During data integration, the pixel coordinates of each matching record are... Intensity value of dual polarized light ,angle and timestamp The data is merged into a single record and serialized chronologically to form a multidimensional data table. During this process, if a data record for a certain angle segment is lost in the time series (e.g., a row of data is not collected), interpolation is used to generate a fitted value within that angle segment that matches the light intensity trend before and after it, ensuring the data table's continuity and integrity. Finally, once the pixel row is associated with its corresponding angle and time, a three-dimensional data structure of light intensity-angle-time is obtained. This structure can be used in subsequent steps for further pixel set feature analysis and processing traceability, such as in monitoring a processing operation at an angle. corresponding time The light intensity vector of 1920 pixels was recorded at that point. Comparing it with the reference light intensity distribution can verify the integrity and stability of the crankshaft surface imaging at that moment.

[0068] S103: Based on the light intensity values ​​in the light intensity angle time correlation table, calculate the light intensity range ratio for the set of pixels with the same angle, calculate the normalization coefficient based on the ratio, transform and aggregate the set of light intensity values ​​into two-dimensional image frames, and generate a crankshaft surface image dataset.

[0069] Based on the light intensity values ​​recorded in the light intensity angle-time correlation table, firstly, for the same angle... Extract the array of horizontal polarization light intensity values ​​of pixels at this angle from the pixel set. and array of vertically polarized light intensity values Then, calculate the maximum and minimum light intensity values ​​for that pixel set, for example, at an angle. In the corresponding pixel set, the maximum horizontal polarization light intensity is:

[0070] The minimum value is ;

[0071] The maximum intensity of vertically polarized light is:

[0072] The minimum value is ;

[0073] When calculating the intensity range ratio of this set of light, first calculate the range of the horizontal components separately. and vertical component range Then, the two values ​​are averaged to obtain the reference value for the light intensity range:

[0074] ;

[0075] Then, based on the set normalization interval Calculate the normalization coefficient for each pixel. This involves subtracting the minimum light intensity value of the set from the current pixel's light intensity value, and then dividing by the reference value. For example, the intensity of horizontally polarized light in a certain pixel is The intensity of vertically polarized light is Then the corresponding , Applying a normalization coefficient to the original light intensity value of the pixel yields the normalized light intensity. and Then, a final normalized single light intensity is formed through a two-component fusion method. This process iteratively executes on pixels at each angle, and reconstructs a two-dimensional image frame by sequentially arranging the normalized light intensity sequence at the same angle according to pixel coordinates, ensuring that the pixel positions accurately correspond to the coordinates of the original imaging matrix when forming the image frame. During processing, for pixel sets with excessively low light intensity range ratios (e.g., ... The image is directly identified as a low-contrast set, and the original light intensity value is not included in the normalization calculation to avoid amplifying noise signals. Finally, after all the two-dimensional image frames of the angle are generated, the image frames arranged in the order of processing time are saved sequentially to form a crankshaft surface time-series image dataset. Each frame in this dataset corresponds to the crankshaft surface state at a specific angle, which can provide intuitive image input for subsequent data analysis, defect detection and other tasks.

[0076] Please see Figure 3 The specific steps of S2 are as follows:

[0077] S201: Based on the crankshaft surface image dataset, perform a two-dimensional discrete Fourier transform operation on the pixel gray-level matrix of the sampling points, transform the two-dimensional gray-level distribution in the spatial domain into a frequency domain amplitude matrix and a frequency domain phase matrix, extract the phase values ​​and arrange them by index to generate the sampling point phase matrix;

[0078] Based on the crankshaft surface image dataset, when performing a two-dimensional discrete Fourier transform on the pixel grayscale matrix of the sampling points, a single-frame two-dimensional grayscale matrix within a specific processing time period is first selected from the sampling frames of the crankshaft machining. The dimension of this matrix is ​​consistent with the image resolution; for example, selecting... A pixel region, where the grayscale value of each pixel is an integer range. to ,by To represent light reflection intensity in units, the grayscale matrix needs to be mean-shifted during sampling. This means subtracting the mean grayscale value of the entire matrix from the grayscale value of each pixel. For example, if the mean is... Then the grayscale value Pixel adjustment to Subsequently, a fast iterative two-dimensional discrete Fourier decomposition process is performed in both the horizontal and vertical directions to transform the gray-level distribution in the spatial domain row by row and column by column into a complex frequency domain representation. For the matrix located in the row... ,List The frequency domain position of , whose complex spectral value is determined by the real part With the imaginary part Composition, such as sampling calculation , Then, the amplitude is obtained by performing square root and square root operations. Simultaneously, the arctangent function is called to calculate the phase. This operation is repeated at the frequency domain coordinates of the grayscale matrix, thus forming two two-dimensional structures: the amplitude matrix and the phase matrix, ensuring an accurate and unbiased correspondence between the matrix coordinates and the frequency indices in the frequency domain. To ensure that subsequent operations on the phase matrix between sampling points do not result in index misalignment, the phase matrix... According to the two-dimensional index Phase values ​​are extracted sequentially from left to right and from top to bottom and then linearly rearranged into a one-dimensional phase vector. Then, the linear structure is remapped according to the sampling point number into a matrix structure corresponding to the original sampling point position. This matrix is ​​the sampling point phase matrix. In actual processing and inspection, when the reflection mode of a sampling point changes due to a minor surface defect, its phase value in the frequency domain changes from the normal value. Offset to These changes are accurately recorded in the sampling point phase matrix after Fourier transformation, providing clear input data for subsequent phase difference calculations, and finally generating the sampling point phase matrix.

[0079] S202: Call the sampling point phase matrix, perform differential operations on the phase values ​​of adjacent sampling points in the matrix, and form a phase difference matrix according to the index order of the phase difference results, keeping the correspondence between positive and negative difference values ​​and sampling point indices consistent, and generate a phase delay difference dataset.

[0080] To access the sampling point phase matrix, first read the set of two-dimensional phase values ​​arranged according to the spatial position of the sampling points from the matrix. And determine the index pairs of adjacent sampling points in row priority order, for example, in the row index Column index The sampling point and its horizontally adjacent points A pair of differential data is formed between them, and the phase values ​​at the two points are recorded as follows: and Perform phase difference calculation to obtain A negative difference indicates that the second sampling point lags behind the first sampling point in phase. For adjacent sampling points in the vertical direction, such as coordinates... and Perform the difference calculation as in the previous step, such as... and Obtain the phase difference A positive difference indicates that the phase of the next sampling point is advanced. This difference calculation is performed bidirectionally and cyclically throughout the entire phase matrix, meaning that adjacent differences in both the horizontal and vertical directions are calculated. Simultaneously, the sign and positional relationship of the difference corresponding to each pair of indices are recorded. If the absolute value of the difference exceeds a preset baseline threshold (for example, a threshold set to...),... (Based on empirical settings of typical phase fluctuation amplitudes during crankshaft surface machining), the index pair is marked as a phase abrupt change point, and the marking information is appended to the difference record. When generating the phase difference matrix, the differences between adjacent sampling points are stored in the original matrix index order. If a sampling point has no complete adjacent points at the boundary position, it is recorded as a zero difference at the corresponding matrix position and marked as a "boundary sampling point" in the internal record table. After recording, the calculated phase difference values ​​are recombined into a two-dimensional matrix in index order to form the phase difference matrix, and a mapping table of corresponding sampling point indices and positive and negative differences is stored separately outside the matrix. For example, in a crankshaft main journal phase analysis, the angle... The corresponding set of phase difference matrix data shows that the phase difference values ​​are concentrated in to Within the range, only a few locations exceeded [the limit]. The threshold, the point index, and the difference, along with the spatial coordinates, are entered into the phase delay difference dataset as the raw input for subsequent angle compensation operations, thereby generating the phase delay difference dataset.

[0081] S203: Based on the phase delay difference dataset, call the crankshaft rotation angle parameter, compensate and adjust the phase delay difference according to the rotation angle parameter, map the compensated and corrected phase delay difference to the corresponding sampling point spatial position, and generate a surface defect distribution map.

[0082] Based on the phase delay difference dataset, first read the phase delay difference value contained in each record. Spatial coordinates of sampling points and the corresponding sampling timestamp Then call the crankshaft rotation angle parameter Instead of establishing a one-to-one mapping relationship, for example, a sampling point in coordinates At that point, its phase delay difference The corresponding angle is Then, an angle-based compensation adjustment is performed on the difference, that is, the abnormal phase delay difference is corrected according to the machining characteristics corresponding to the rotation angle, for example, by setting an angle compensation coefficient. The coefficients for different angle ranges are distributed with reference to empirical data on the rotational inertia and structural deformation of crankshafts, for example, in... interval selection ,exist interval selection Then the phase delay difference after correction of the above sampling points is This operation is performed sequentially on each sampled point in the dataset. During the correction process, it is also necessary to determine whether the corrected value exceeds the normal processing fluctuation range (for example, the normal fluctuation range is set to...). If the absolute value still exceeds a certain point after correction, These points are then marked as defect points. Next, the compensated phase delay difference is mapped back to the corresponding two-dimensional imaging plane coordinate system according to the spatial location of the sampling points. The final values ​​corresponding to the same coordinates are directly stored in the two-dimensional matrix units, thus forming a phase delay difference distribution matrix that corresponds one-to-one with the crankshaft surface position. Finally, this matrix is ​​visualized and rendered into a surface defect distribution map, with different phase delay difference intervals marked with different colors. The corresponding color is green. The corresponding color is yellow. The corresponding color is red, which visually represents the difference status of different regions and preserves the precise difference values ​​as an additional layer of the image for use in subsequent numerical reading operations.

[0083] Table 2: Compensation Results for Phase Delay Difference at Some Sampling Points on the Crankshaft Surface

[0084]

[0085] As shown in Table 2, the table lists the phase delay difference data of some sampling points before and after angle compensation correction, as well as the defect marking situation based on the normal fluctuation range. Among them, the phase delay difference after correction still exceeds The sampling points are marked as "yes", and their spatial locations will be directly reflected in the final generated crankshaft surface defect distribution map.

[0086] Please see Figure 4 The specific steps of S3 are as follows:

[0087] S301: Based on the spatial coordinates of pixels in the surface defect distribution map and the defect region identifier encoding, the phase delay parameters of adjacent pixels in the two-dimensional spatial axis are called, and the phase delay difference of adjacent pixel pairs is vector decomposed according to the horizontal and vertical directions to generate a phase gradient distribution sequence.

[0088] Based on the spatial coordinates of pixels in the surface defect distribution map and the defect region identifier encoding, the generated defect distribution map is first read pixel by pixel, and each pixel is extracted. spatial coordinates and defect area identification code The method for determining whether a pixel belongs to a defect area is to directly compare the encoded values. For example, an encoded value of 1 indicates a defective pixel, and 0 indicates a non-defective pixel. After confirming the pixels in the defective area, the phase delay parameters of the pixels adjacent to that pixel in the two-dimensional plane are read. And calculate the phase delay difference for adjacent pixel pairs along both the horizontal (x-direction) and vertical (y-direction) axes, for example, at pixel and Between, the delay parameter is and Then the horizontal phase delay difference is In the vertical direction of pixels and The delay parameter is The vertical phase delay difference is Next, the horizontal and vertical differences are recorded as the x and y components of a vector, respectively, in the phase gradient sequence, using the decomposition method of a two-dimensional vector. For example, the gradient vector of this pixel is... This process is repeated for pixels within the defect area. During vector decomposition, the phase delay variation magnitude needs to be classified according to the difference interval, for example, defining... For slight changes, The change is moderate. To prevent drastic changes, the classification information is retained in the phase gradient sequence for subsequent gradient magnitude and direction statistics. After the calculation of each pixel is completed, the spatial coordinates of the pixel, gradient vector components and change category are added to the gradient distribution sequence in sequence to ensure that the phase delay of the defect area can be fully reflected in the two-dimensional space as the position changes. The generated sequence finally constitutes the phase gradient distribution sequence.

[0089] S302: Based on the gradient magnitude and direction angle information of the phase gradient distribution sequence, retrieve the changing trend of the gradient at adjacent positions and filter the direction vector set of the continuous path. Weight the magnitude of each vector in the path and the distribution weight of its direction to obtain the defect severity vector set.

[0090] Based on the gradient magnitude and direction angle information recorded in the phase gradient distribution sequence, the coordinates of adjacent pixels in the sequence are first compared to determine whether they are spatially adjacent points. For example, when the coordinates of two pixels are respectively... and When the two are adjacent along the horizontal axis, the gradient magnitude is calculated. and They are respectively and and direction angle , Then, it checks whether the gradient directions of adjacent positions show a changing trend, i.e., it calculates... and In this example and Less than the direction consistency threshold A continuous gradient path segment that can be identified as exhibiting a changing trend is recorded as part of the path. This process is performed sequentially across the entire gradient distribution sequence, comparing pixel by pixel indices. Adjacent pixel gradients that conform to the changing trend are merged into continuous paths with the same direction, and a set of direction vectors is extracted from each path. ,in For amplitude, This refers to the direction angle. When processing each direction vector within the path, it is weighted and aggregated according to the distribution weight of its direction. For example, the direction distribution weight is set to be calculated based on the concentration of actual defect directions. If there are defects within a certain direction segment... All vector directions are in If the concentration is within the range, then the directional concentration weight coefficient is taken as... Then, multiply each amplitude within the segment by its weight, sum them, and take the average to obtain the weighted amplitude. For example, the amplitude set is:

[0091] ;

[0092] The weighted result is:

[0093] ;

[0094] This value is linked to the direction angle to form a path severity vector. The severity vectors of the paths are integrated to form a defect severity vector set. Each element in this vector set represents the intensity of the phase delay change of the defect on the processed surface along a certain stable direction and its direction parameter.

[0095] S303: Call the amplitude features and orientation distribution parameters of the defect severity vector set, use a similarity measure based on feature space distance to divide the vector set into categories, identify the defect pattern number corresponding to each category and map it to the matching quality level index, and generate the processing quality prediction level.

[0096] To retrieve the amplitude characteristics and orientation distribution parameters of the defect severity vector set, the vector set is first read in record order, and the amplitude of each vector is extracted separately. with direction and angle Then, the distance between any two vectors in the feature space is calculated to measure similarity. This distance is calculated by first normalizing the amplitude and direction angle to a uniform dimension, for example, setting the amplitude interval to... Mapping to [0, 1], and mapping the direction angle interval [0°, 180°] to [0, 1], the two-dimensional Euclidean distance is calculated based on this. For example, for amplitudes of 0.0174 rad and 0.0186 rad, their normalized values ​​are respectively , The directions and angles are respectively , The normalized value is , The distance between the two was calculated. Vectors with a distance less than a preset clustering threshold of 0.1 are grouped into the same class. The threshold of 0.1 is statistically obtained based on the minimum separation distance of different types of defects in the feature space in the original crankshaft machining case. After completing the distance calculation and clustering of the entire vector set, a defect pattern number is assigned to each class sequentially, such as pattern 1, pattern 2, etc. The pattern number is then matched with a pre-established quality level index table. For example, pattern 1 corresponds to quality level A, and pattern 2 corresponds to quality level B. When the amplitude of the severity vector set corresponding to a certain pattern number is greater than 0.03 rad (i.e., greater than the severe defect benchmark value), it is matched to the lowest quality level C. If the amplitude is less than 0.015 rad and the directional concentration is high, it is matched to the highest quality level A. Finally, according to the spatial distribution of the samples, the pattern number of each defect location is mapped to the corresponding quality level index to generate a machining quality prediction level result for machining process evaluation.

[0097] Table 3: Defect Pattern Clustering and Quality Grade Mapping Table

[0098]

[0099] As shown in Table 3, the correspondence between defect pattern number and quality level is determined based on the similarity measurement of amplitude and direction features and the results of clustering operations. Pattern 3 is listed as quality level C because its average amplitude exceeds the severe defect benchmark value of 0.03 rad, and is marked as an area with high processing risk in the final quality prediction level output.

[0100] Please see Figure 5 The specific steps of S4 are as follows:

[0101] S401: Based on the machining quality prediction level, multi-dimensional process parameters are constructed by combining cutting speed, feed rate, and depth of cut. The original parameter values ​​are assembled into column vectors according to their corresponding dimensional positions. Vector inner product and variance operations are performed, and the covariance is merged into coupling strength data to generate a set of parameter coupling coefficients.

[0102] When constructing multi-dimensional process parameters based on machining quality prediction levels and cutting speed, feed rate, and depth of cut, the first step is to extract the original set of process parameters corresponding to each quality prediction level label from the production database of crankshaft machining batches. For example, if a workpiece has a quality level of B, its collected data would be the cutting speed. Feed rate Depth of cut The three parameters of each batch are assembled into a column vector in a fixed order according to their physical dimensions. For example, for three samples , , Form three column vectors, then perform a vector dot product operation on each pair of column vectors. The calculation method is to multiply the corresponding three terms and then add them together, for example... This process is repeated for each sample pair to obtain the inner product matrix. Then, the variance is calculated separately for each parameter dimension. For example, in the variance calculation of cutting speed, the mean is calculated first. Then calculate the sum of the differences of squares. Divide by the sample size to get the variance. Similarly, the dimensional variance can be obtained from the feed rate and depth of cut. , Then, the covariance value is constructed based on the corresponding positions of the inner product matrix and the variance of each dimension. When calculating the covariance, the average value of the product of the sample deviations corresponding to the two-dimensional parameters is directly taken, for example, the covariance between cutting speed and feed rate:

[0103] ;

[0104] After generating the complete covariance matrix, the coupling strength data between each parameter pair is calculated. During the calculation, the absolute value of the covariance is normalized to the [0, 1] interval to reflect the strength ratio. For example, let the coupling strength between cutting speed and feed rate be... The coupling strength between feed rate and depth of cut is 0.3, and the coupling strength between cutting speed and depth of cut is 0.8. Finally, the coupling strength of the parameter pairs is recorded in the parameter coupling coefficient set. The coupling coefficient of each pair of parameters is directly assigned by the normalized value of the corresponding covariance position, forming a parameter coupling coefficient set suitable for subsequent over-limit judgment.

[0105] S402: Call the parameter coupling coefficient set, extract the coupling coefficients and compare them with the stored associated thresholds item by item, record the indexes with values ​​greater than the threshold as parameter identifiers, and aggregate the identifiers to form an over-limit parameter identifier group;

[0106] To access the parameter coupling coefficient set, first read the coupling coefficient values ​​of each group in the predetermined parameter pair order. Simultaneously, it reads the associated threshold corresponding to this parameter pair from the threshold storage table. Then, the difference is calculated item by item, that is, the difference is obtained by subtracting the threshold from the coupling coefficient. For example, the coupling coefficient between cutting speed and feed rate is... Association threshold The difference is The coupling coefficient between cutting speed and depth of cut is threshold The difference is The coupling coefficient between feed rate and depth of cut is threshold The difference is When determining whether a limit has been exceeded, the difference is directly compared to zero. This indicates that the coupling coefficient exceeds the threshold. The corresponding parameter pair index is recorded. For example, in the above example, the parameter pairs that exceed the limit are (cutting speed, feed rate) and (cutting speed, depth of cut). When recording the index, parameter identifier encoding is used: 1 represents cutting speed, 2 represents feed rate, and 3 represents depth of cut. Therefore, (cutting speed, feed rate) is the encoding group (1, 2), and (cutting speed, depth of cut) is the encoding group (1, 3). After completing the item-by-item judgment of the entire coefficient set, the exceeding index pairs are aggregated in ascending order to form the exceeding parameter identifier group. For example, the aggregation result in this example is {(1, 2), (1, 3)}, which is used as the input set for subsequent process parameter adjustment calculations.

[0107] S403: Locate the corresponding value in the multi-dimensional process parameter set based on the over-limit parameter identification group, calculate the adjustment range in combination with the processing quality prediction level reference range, serialize the difference between the adjustment range and the original value, and generate a process parameter adjustment plan;

[0108] When locating the corresponding value in the multidimensional process parameter set based on the out-of-limit parameter identifier group, first find the current set value corresponding to each pair of parameters in the identifier group in the original machining parameter dataset. For example, identifier (1, 2) corresponds to cutting speed and feed rate, and the current value is... , The identifier (1, 3) corresponds to the cutting speed and depth of cut, and the current value is... , Then call the machining quality prediction level reference range table. For example, for quality level B, the recommended cutting speed range is... The feed rate range is The cutting depth range is The reference range and the current actual value are used to calculate the interval difference, such as the difference in cutting speed. (Above the upper limit), feed rate difference (Above the upper limit), difference in cutting depth (Above the upper limit). Next, determine the adjustment direction based on the sign of the difference: if it's above the upper limit, adjust downwards; if it's below the lower limit, adjust upwards, and adjust according to the allowable adjustment range coefficient. Calculate the adjustment interval, for example, let... The cutting speed adjustment amount is The feed rate adjustment is The depth of cut adjustment amount is The final adjustment range is obtained by combining the adjustment direction and the adjustment amount. For example, if the cutting speed is adjusted to... Downward control, feed rate adjusted to Downward control, cutting depth adjusted to Downward control. Finally, the parameters are arranged and serialized in order to form a structured process parameter adjustment scheme table.

[0109] Table 4: Process Parameter Adjustment Scheme

[0110]

[0111] As shown in Table 4, the adjustment scheme provides the current value of each out-of-limit parameter, the corresponding reference range, the difference between the reference range and the reference range, the calculated adjustment direction and adjustment amount, and the final suggested range after adjustment. The generated process parameter adjustment scheme can be directly used to guide the parameter setting of subsequent crankshaft machining batches.

[0112] Please see Figure 6 The specific steps of S5 are as follows:

[0113] S501: Based on the process parameter adjustment scheme, the cutting speed, feed rate and depth of cut are formed into a multi-dimensional parameter sequence. The process coefficients are calculated and combined one by one. The multi-objective function constraint calculation is called to eliminate non-convergent combinations and generate an initial parameter combination set.

[0114] Based on the process parameter adjustment scheme, the cutting speed, feed rate, and depth of cut are read sequentially according to the process sequence. The data from each combination are then arranged in a fixed order to form a three-dimensional parameter sequence. For example, the median value within the suggested range in the adjustment scheme is used for the cutting speed. Feed rate Depth of cut , in sequence, form column vectors Then, the coefficient values ​​corresponding to each parameter dimension in the process coefficient table are called, such as the cutting speed coefficient. Feed rate coefficient Cutting depth coefficient Calculate each dimension item by item, that is, calculate them separately. , , The above processing is performed sequentially on the combinations of values ​​within the suggested range, forming a multi-dimensional parameter combination set after process coefficient processing. Next, a merging operation is performed, combining the dimensional adjustment results into a unified matrix structure as column vectors. Each column represents a parameter dimension, and each row corresponds to a complete three-dimensional parameter combination, for example... After generating the matrix, the combined data is fed into a multi-objective function constraint operation. This multi-objective function simultaneously considers machining efficiency constraints, tool wear constraints, and vibration amplitude constraints. Each combination needs to be verified under each of the three objective constraints. For example, for the combination [190.05, 0.21854, 1.782], if the output time cost of 2.84 min in the machining efficiency calculation is less than the target upper limit of 3.00 min, this sub-condition is met; if the expected wear amount in the tool wear calculation is 0.019 mm, which is lower than the maximum allowable wear value of 0.025 mm, this sub-condition is met; if the estimated value in the vibration amplitude model calculation is 0.031 mm, which is higher than the maximum allowable amplitude of 0.030 mm, then this combination does not meet the condition under the vibration amplitude objective and is considered a non-convergent combination, requiring elimination. This operation is performed sequentially, eliminating combinations that do not converge under at least one objective function condition, and finally retaining the three-dimensional parameter combinations that satisfy all objective constraints, forming an initial parameter combination set for subsequent feasibility screening of machining equipment.

[0115] S502: Call each set of cutting speed, feed rate, and depth of cut data in the initial parameter combination set, perform difference calculation and sign judgment on them with the power limit value and stroke boundary value in the processing equipment capacity constraint conditions, eliminate parameter combinations that exceed the limits, and obtain the feasible parameter set of the equipment;

[0116] When retrieving each set of cutting speed, feed rate, and depth of cut data from the initial parameter combination set, the parameter values ​​of each row in the combination set are read sequentially first. For example, if a certain combination is for cutting speed... Feed rate Depth of cut Then, the difference between the data and the power limit value and travel boundary value in the processing equipment capacity constraints is calculated. The power limit value is given by the equipment's rated parameters, such as the rated power. The power consumption is estimated based on the empirical formula for calculating cutting power. The input terms are taken from the three parameters of the combination and the material cutting resistance constant. The cutting power corresponding to the combination is then calculated. Then perform the difference calculation. The sign is determined; a negative value indicates that the power limit has not been exceeded. The travel boundary value is jointly determined by the travel constraints of the equipment's worktable and tool holder. For example, the upper limit of the x-axis travel is... The maximum travel distance in the y-direction is When the cutting depth in the combination The corresponding estimated value of the tool axial displacement (including safety margin) is At that time, the difference A negative sign indicates that the longitudinal travel limit has not been exceeded. Similarly, the estimated x-direction displacement under the combined effects of cutting speed, feed rate, etc. Difference The difference and sign are both negative, indicating that the combination meets the stroke requirements. In this step, the above difference and sign judgment is performed on each group of the initial parameter combination set. If a combination has a positive difference in any item, it is directly eliminated. All combinations with negative or zero differences under both power and stroke constraints are retained. The summaries are used to obtain the feasible parameter set of the equipment as the input for the next step of roughness and dimensional accuracy dual constraint calculation.

[0117] S503: Based on the combination data of the feasible parameter set of the equipment, perform absolute difference calculation on the measured surface roughness value and the surface roughness threshold, and perform difference judgment on the measured size value and the size tolerance threshold, eliminate any combination whose difference exceeds the threshold, and generate crankshaft machining process optimization instructions;

[0118] Based on the combined data of the feasible parameter set of the equipment, the actual measured values ​​of each group of cutting speed, feed rate, and depth of cut are read sequentially after machining. First, the measured values ​​of surface roughness are checked. With surface roughness threshold Perform absolute difference calculations, for example, to measure the surface roughness of a certain combination during trial cutting. Reference threshold Its absolute difference is If the difference is greater than the roughness control upper limit of 0.05, the combination is considered passed; otherwise, it is rejected. Next, the dimensional accuracy is checked by taking the measured dimensional values. With respect to dimensional tolerance threshold range Comparison, such as measured values The tolerance threshold range is If the roughness difference exceeds the upper or lower limit, the combination is discarded. This step requires two conditions to be met simultaneously: the roughness difference must be within the allowable range and the measured dimensional value must be within the tolerance zone. Only combinations that meet both conditions are retained. To improve screening efficiency, the absolute roughness difference and dimensional judgment results are stored in Boolean form, for example, "1" indicates qualified and "0" indicates unqualified. Finally, the combinations that meet the conditions are output and presented in a table to form crankshaft machining process optimization instructions to guide the next batch of machining.

[0119] Table 5: Crankshaft Machining Process Optimization Instruction Table

[0120]

[0121] As shown in Table 5, only combinations that meet the processing requirements in terms of surface roughness difference and dimensional accuracy will be retained and recorded in the process optimization instruction for direct use in subsequent processing.

[0122] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A process optimization method driven by image big data for crankshaft machining, characterized in that, Includes the following steps: S1: Obtain crankshaft surface reflection image data through bidirectional polarized light source, collect light intensity distribution in different polarization directions, monitor pixel light intensity and spatial coordinates, detect speed encoder angular position signal, obtain timestamp and crankshaft rotation angle parameters, and perform light intensity normalization to generate crankshaft surface image dataset. S2: Perform Fourier transform frequency domain analysis based on the crankshaft surface image dataset, calculate the phase information of the sampling points, calculate the phase delay difference dataset through phase difference calculation, call the crankshaft rotation angle parameter to perform angle compensation correction, and generate a surface defect distribution map. S3: Calculate the spatial distribution density and gradient change rate of the defect area by calling the surface defect distribution map, calculate the phase delay difference gradient between adjacent points, analyze the defect severity vector by gradient magnitude and direction angle, identify the defect pattern type by cluster analysis, and generate the processing quality prediction level. S4: Call the machining quality prediction level, combine it with the current cutting speed, feed rate, and depth of cut, input the machining quality prediction level into the support vector machine algorithm to perform parameter correlation analysis, identify parameters whose correlation coefficient exceeds the correlation threshold, and generate a process parameter adjustment plan.

2. The image big data-driven process optimization method for crankshaft machining according to claim 1, characterized in that, The crankshaft surface image dataset includes a pixel grayscale matrix, spatial geometric features, and reflected light intensity characteristics. The surface defect distribution map includes defect spatial coordinates, defect area data, and defect shape features. The processing quality prediction level includes a level number, a quality score value, and a corresponding defect category. The process parameter adjustment scheme includes target optimization parameters, constraint parameters, and allowable variation range.

3. The image big data-driven process optimization method for crankshaft machining according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Obtain crankshaft surface reflection image data through bidirectional polarization light source, monitor pixel light intensity values ​​and spatial coordinates in different polarization directions, vectorize and combine the light intensity at the same pixel position, and call the combined light intensity vector to perform position mapping in the two-dimensional data domain to generate a pixel light intensity matrix. S102: Call the pixel light intensity matrix and the angular position signal of the speed encoder, match the pixel light intensity of each row in the matrix with the corresponding crankshaft rotation angle parameter, integrate the timestamps of the matching records and serialize them into a multi-dimensional data table to obtain the light intensity angle time association table. S103: Based on the light intensity values ​​in the light intensity angle time correlation table, calculate the light intensity range ratio for the set of pixels at the same angle, calculate the normalization coefficient based on the ratio, transform and aggregate the set of light intensity values ​​into two-dimensional image frames, and generate a crankshaft surface image dataset.

4. The image big data-driven process optimization method for crankshaft machining according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Based on the crankshaft surface image dataset, perform a two-dimensional discrete Fourier transform operation on the pixel grayscale matrix of the sampling points, transform the two-dimensional grayscale distribution in the spatial domain into a frequency domain amplitude matrix and a frequency domain phase matrix, extract the phase values ​​and arrange them by index to generate the sampling point phase matrix; S202: Call the sampling point phase matrix, perform differential operation on the phase values ​​of adjacent sampling points in the matrix, and form a phase difference matrix according to the index order of the phase difference results, keeping the correspondence between positive and negative difference values ​​and sampling point indices consistent, and generate a phase delay difference dataset. S203: Based on the phase delay difference dataset, call the crankshaft rotation angle parameter, compensate and adjust the phase delay difference according to the rotation angle parameter, map the compensated and corrected phase delay difference to the corresponding sampling point spatial position, and generate a surface defect distribution map.

5. The image big data-driven process optimization method for crankshaft machining according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Based on the spatial coordinates of pixels in the surface defect distribution map and the defect region identifier encoding, the phase delay parameters of adjacent pixels in the two-dimensional spatial axis are called, and the phase delay difference of adjacent pixel pairs is vector decomposed according to the horizontal and vertical directions to generate a phase gradient distribution sequence. S302: Based on the gradient magnitude and direction angle information of the phase gradient distribution sequence, retrieve the changing trend of the gradient at adjacent positions and filter the direction vector set of the continuous path. Weighted aggregate the magnitude of each vector in the path and the distribution weight of its direction to obtain the defect severity vector set. S303: Call the amplitude features and orientation distribution parameters of the defect severity vector set, use a similarity metric based on feature space distance to divide the vector set into categories, identify the defect pattern number corresponding to each category and map it to the matching quality level index, and generate the processing quality prediction level.

6. The image big data-driven process optimization method for crankshaft machining according to claim 5, characterized in that, The specific steps of S4 are as follows: S401: Based on the machining quality prediction level, multi-dimensional process parameters are constructed by combining cutting speed, feed rate, and depth of cut. The original parameter values ​​are assembled into column vectors according to their corresponding dimensional positions. Vector inner product and variance operations are performed, and the covariance is merged into coupling strength data to generate a set of parameter coupling coefficients. S402: Call the parameter coupling coefficient set, extract the coupling coefficients and compare them with the stored associated thresholds item by item, record the indexes with values ​​greater than the thresholds as parameter identifiers, and aggregate the identifiers to form an over-limit parameter identifier group; S403: Locate the corresponding value in the multi-dimensional process parameter set according to the over-limit parameter identification group, calculate the adjustment range in combination with the processing quality prediction level reference range, serialize the difference between the adjustment range and the original value, and generate a process parameter adjustment scheme.

7. The image big data-driven process optimization method for crankshaft machining according to claim 6, characterized in that, The parameter coupling coefficient refers to a dimensionless value obtained by normalizing the covariance between multidimensional process parameters. The association threshold refers to the numerical limit of whether the coupling coefficient of the determination parameter reaches the preset coupling determination condition.

8. The image big data-driven process optimization method for crankshaft machining according to claim 1, characterized in that, The method also includes step S5: S5: Calculate a new combination of cutting speed, feed rate, and depth of cut based on the process parameter adjustment scheme, solve the optimized values ​​of process parameters through a multi-objective function, verify the processing equipment capability constraints, screen parameters that simultaneously meet the surface roughness threshold and dimensional tolerance threshold, and generate crankshaft processing optimization instructions; The crankshaft machining process optimization instructions include optimizing cutting speed, optimizing feed rate, and optimizing depth of cut.

9. The image big data-driven process optimization method for crankshaft machining according to claim 8, characterized in that, The specific steps of S5 are as follows: S501: Based on the process parameter adjustment scheme, the cutting speed, feed rate and depth of cut are formed into a multi-dimensional parameter sequence, and the process coefficients are calculated and combined one by one. The multi-objective function constraint calculation is called to eliminate non-convergent combinations and generate an initial parameter combination set. S502: Call each set of cutting speed, feed rate, and depth of cut data in the initial parameter combination set, perform difference calculation and sign judgment on them with the power limit value and stroke boundary value in the processing equipment capacity constraint conditions, eliminate parameter combinations that exceed the limits, and obtain the feasible parameter set of the equipment; S503: Based on the combination data of the feasible parameter set of the equipment, perform absolute difference calculation on the measured surface roughness value and the surface roughness threshold, and perform difference judgment on the measured size value and the size tolerance threshold, eliminate any combination whose difference exceeds the threshold, and generate crankshaft machining process optimization instructions.

10. The image big data-driven process optimization method for crankshaft machining according to claim 9, characterized in that, The surface roughness threshold refers to the surface roughness limit value set according to the product design documents and processing technology standards; The dimensional tolerance threshold refers to the dimensional deviation limit set according to the product design documents and tolerance standards.

Images