Adaptive calculation method for assembly parameters of reduction gearbox housing

Abstract

The present application relates to the technical field of reducer automation assembly, and specifically discloses a reducer housing assembly parameter adaptive calculation method, which collects a press-fit force value and a displacement value in real time to construct an original time series data set; performs dynamic characteristic decomposition on the original time series data set, extracts an energy contribution rate of an intrinsic dynamic component to form a dynamic characteristic vector; performs similarity measurement on the dynamic characteristic vector and a pre-stored deformation recovery feature library and a real jamming feature library respectively to obtain two matching degrees; determines whether the force fluctuation originates from housing deformation recovery or assembly jamming according to the matching degree comparison result, and correspondingly generates control parameters for inhibiting position correction superimposed micro-amplitude vibration or activating position backtracking correction; adjusts the press-fit process in real time according to the control parameters, and optimizes the corresponding feature library by taking the current press-fit data as a new sample; the present application realizes accurate identification and differentiated control of the source of force fluctuation, avoids misjudgment caused by resistance during the deformation recovery period, and improves assembly continuity and adaptive ability of the feature library.

Description

Technical Field

[0001] This invention relates to the field of automated assembly technology for speed reducers, and more specifically to an adaptive calculation method for assembly parameters of a speed reducer housing. Background Technology

[0002] Gearbox housing assembly is a critical process in automotive transmission manufacturing, employing an interference fit to press the bearing outer ring or oil seal into the bearing bore of the housing. To ensure assembly accuracy, existing automated pressing equipment generally adopts an adaptive control method based on force-position feedback. This method dynamically adjusts the feed speed and pressing trajectory by monitoring resistance changes in real time during the pressing process to avoid problems such as jamming or incomplete pressing.

[0003] Existing adaptive press-fitting methods are typically based on impedance control principles, directly and linearly mapping real-time force errors to position corrections. Their core flaw lies in their ability to perceive only the superficial "resistance anomaly" from the perspective of temporal amplitude, failing to decouple the physical source of force fluctuations from the deeper dimension of dynamic evolution. When faced with two distinctly different but potentially overlapping force signals in the temporal domain—the deformation recovery process after vacuum adsorption and release of a thin-walled shell (characterized by low-amplitude, long-period, exponentially decaying force fluctuations) and the actual assembly jamming process (characterized by high-amplitude, short-period, exponentially increasing force peaks)—traditional algorithms, lacking the ability to identify the inherent evolutionary patterns of force fluctuations, inevitably misjudge normal resistance during the deformation recovery period as interference or deviation, triggering ineffective deceleration and yielding. This not only fails to solve the assembly problems caused by deformation but also introduces additional efficiency losses and process oscillations due to erroneous actions. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive calculation method for the assembly parameters of a gearbox housing, so as to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: An adaptive calculation method for gearbox housing assembly parameters includes the following steps: S1: During the pressing process, the pressing force value and its corresponding displacement value are collected in real time at a fixed sampling frequency to construct the original time series dataset reflecting the changes of force and displacement over time; S2: Perform dynamic feature decomposition on the original time series dataset, extract multiple intrinsic dynamic components that represent the intrinsic evolution law of force fluctuation, and calculate the energy contribution rate of each intrinsic dynamic component. Use the energy contribution rate to construct the dynamic feature vector of the current pressing process. S3: Perform similarity measurements between the dynamic feature vector and the pre-stored deformation recovery feature library and the real jamming feature library to obtain the deformation recovery matching degree and the real jamming matching degree. S4: Compare the deformation recovery matching degree with the actual jamming matching degree. If the deformation recovery matching degree is greater than the actual jamming matching degree and exceeds the deformation recovery judgment threshold, it is determined that the current force fluctuation originates from the deformation recovery of the shell. A deformation suppression control parameter with suppressed position correction and superimposed axial micro-vibration is generated. If the actual jamming matching degree is greater than the deformation recovery matching degree and exceeds the actual jamming judgment threshold, it is determined that the current force fluctuation originates from assembly jamming. A jamming avoidance control parameter with activated position backtracking correction is generated. S5: Based on the deformation suppression control parameters or jamming avoidance control parameters, adjust the feed speed and attitude of the current pressing process in real time, and use the force-displacement data of the entire pressing process as new samples to optimize the corresponding deformation recovery feature library or real jamming feature library respectively.

[0006] As a further aspect of the present invention: S2 specifically includes: Based on the preset delayed embedding dimension, the original time series dataset is reconstructed into a Hankel matrix that reflects the evolution trajectory of force fluctuations; The Hankel matrix is ​​orthogonally projected to solve the principal feature space that characterizes the dynamic evolution of force fluctuations, and multiple intrinsic dynamic components are extracted from the principal feature space. Calculate the proportion of each intrinsic dynamic component in the total energy, and use the proportion to construct the dynamic characteristic vector of the current pressing process.

[0007] As a further aspect of the present invention: the extraction of multiple intrinsic dynamic components from the principal feature space specifically includes: Covariance operations are performed on the Hankel matrix to construct a covariance matrix reflecting the energy distribution of force fluctuations, and eigenvalue decomposition is performed on the covariance matrix to obtain the orthogonal projection operator. By using the orthogonal projection operator to project the Hankel matrix onto a low-dimensional feature space, a dimension-reduced characterization matrix reflecting the evolutionary trend of force fluctuations is obtained. Temporal evolution analysis was performed on the dimensionality-reduced representation matrix to extract the intrinsic dynamic components of the intrinsic evolution law of multiple representation force fluctuations.

[0008] As a further aspect of the present invention: the construction process of the deformation recovery feature library is as follows: Force-displacement time-series data of the shell deformation recovery stage during historical press fitting are collected. Dynamic feature decomposition is performed on the force-displacement time-series data to obtain the deformation recovery feature vector of each historical sample. Cluster analysis is performed on each deformation recovery feature vector to obtain multiple deformation recovery cluster centers. The deformation recovery feature library is constructed using the deformation recovery cluster centers.

[0009] As a further aspect of the present invention: the construction process of the real card resistance feature library is as follows: Force-displacement time-series data of actual jamming stages during historical pressing processes are collected. Dynamic feature decomposition is performed on the force-displacement time-series data to obtain the actual jamming feature vectors of each historical sample. Cluster analysis is performed on each actual jamming feature vector to obtain multiple actual jamming cluster centers. The actual jamming feature library is constructed using the actual jamming cluster centers.

[0010] As a further aspect of the present invention: obtaining the deformation recovery matching degree and the actual jamming resistance matching degree specifically includes: Calculate the distance between the dynamic feature vector and each deformation recovery cluster center in the deformation recovery feature library, and use the minimum distance as the deformation recovery matching degree. Calculate the distance between the dynamic feature vector and each real card block cluster center in the real card block feature library, and use the minimum distance as the real card block matching degree.

[0011] As a further aspect of the present invention: S5 specifically includes: The force-displacement time series data of the entire pressing process were subjected to the same dynamic feature decomposition to obtain new feature vectors, and the corresponding deformation recovery category or real jamming category was labeled for the new feature vectors. Calculate the distance between the newly added feature vector and each existing cluster center in the corresponding feature library, and determine the new feature vector to belong to the nearest existing cluster center based on the distance; The newly added feature vectors are merged into the existing cluster centers to which they belong, the coordinate values ​​of the existing cluster centers are updated, and the original cluster centers are replaced with the updated cluster centers to complete the optimization of the corresponding feature library.

[0012] As a further aspect of the present invention: the method of obtaining new feature vectors by performing the same dynamic feature decomposition on the force-displacement time series data of the entire pressing process specifically includes: Complete force-displacement time series data from the start to the end of the pressing process were extracted, and abnormal data points in the unstable contact stages at the beginning and end were removed to obtain pure pressing segment time series data. The pure pressing section time series data is reconstructed into the Hankel matrix to be decomposed, and the same orthogonal projection transformation is performed on the Hankel matrix to be decomposed to obtain the principal feature space to be decomposed. Extract multiple intrinsic dynamic components of the same number as S2 from the principal feature space to be decomposed, calculate the energy contribution rate of each intrinsic dynamic component, and use the energy contribution rate to form a new feature vector.

[0013] The beneficial effects of this invention are: (1) By constructing a deformation recovery feature library and a real jamming feature library, the present invention performs dynamic feature decomposition and matching identification on real-time press-fit force-displacement data. It can accurately distinguish between low-amplitude attenuation force fluctuations caused by shell deformation recovery and high-amplitude growth force peaks caused by real assembly jamming. This avoids the frequent deceleration or yielding triggered by the traditional method misjudging the resistance during the deformation recovery period as interference and excess tolerance, thus improving the continuity of the press-fit process and the assembly cycle.

[0014] (2) The present invention uses the force-displacement data of each pressing process as new samples after decomposing the same features. The corresponding feature library is updated online through the weighted fusion method of cluster centers, so that the feature library can continuously shift towards the real distribution of actual working conditions as the production batch increases. This realizes the dynamic evolution and self-optimization of the identification process and effectively adapts to the working condition drift caused by factors such as shell material characteristics and environmental temperature changes in different batches. Attached Figure Description

[0015] The invention will now be further described with reference to the accompanying drawings.

[0016] Figure 1 This is a flowchart of the adaptive calculation method for the assembly parameters of the gearbox housing according to the present invention; Figure 2 This is a flowchart illustrating the construction process of the deformation recovery feature library in this invention; Figure 3 This is a flowchart illustrating the construction process of the real card resistance feature library in this 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] Please see Figure 1 As shown, the present invention is an adaptive calculation method for the assembly parameters of a gearbox housing, comprising the following steps: S1: During the pressing process, the pressing force value and its corresponding displacement value are collected in real time at a fixed sampling frequency to construct the original time series dataset reflecting the changes of force and displacement over time; S2: Perform dynamic feature decomposition on the original time series dataset, extract multiple intrinsic dynamic components that represent the intrinsic evolution law of force fluctuation, and calculate the energy contribution rate of each intrinsic dynamic component. Use the energy contribution rate to construct the dynamic feature vector of the current pressing process. S3: Perform similarity measurements between the dynamic feature vector and the pre-stored deformation recovery feature library and the real jamming feature library to obtain the deformation recovery matching degree and the real jamming matching degree. S4: Compare the deformation recovery matching degree with the actual jamming matching degree. If the deformation recovery matching degree is greater than the actual jamming matching degree and exceeds the deformation recovery judgment threshold, it is determined that the current force fluctuation originates from the deformation recovery of the shell. A deformation suppression control parameter with suppressed position correction and superimposed axial micro-vibration is generated. If the actual jamming matching degree is greater than the deformation recovery matching degree and exceeds the actual jamming judgment threshold, it is determined that the current force fluctuation originates from assembly jamming. A jamming avoidance control parameter with activated position backtracking correction is generated. S5: Based on the deformation suppression control parameters or jamming avoidance control parameters, adjust the feed speed and attitude of the current pressing process in real time, and use the force-displacement data of the entire pressing process as new samples to optimize the corresponding deformation recovery feature library or real jamming feature library respectively.

[0019] In S1, during the pressing process, the pressing force and its corresponding displacement are collected in real time at a fixed sampling frequency to construct an original time-series dataset reflecting the changes in force and displacement over time, specifically including: First, a pressing force sensor is installed between the pressing head and the spindle of the pressing equipment to sense the axial force generated during the pressing process in real time; at the same time, a displacement sensor is installed on one side of the pressing head to detect the distance the pressing head moves relative to the reference position in real time.

[0020] Before starting the pressing operation, the data acquisition sampling frequency is set to 1000 Hz to ensure that the instantaneous changes in force and displacement can be captured.

[0021] The moment the pressing head begins to feed and contacts the housing, the pressing force sensor and displacement sensor are simultaneously triggered to start working. The pressing force sensor continuously collects the instantaneous value of the pressing force at each moment at a set sampling frequency, and the displacement sensor synchronously collects the instantaneous value of the displacement at each moment. The controller records the collection time corresponding to each set of force and displacement values.

[0022] As the pressing head continues to feed until the pressing process ends, the controller arranges all the pressing force values, displacement values ​​and corresponding acquisition times collected during the entire pressing process in chronological order, forming a two-dimensional time sequence consisting of multiple data points. Each data point in this sequence contains the pressing force value and displacement value at the same moment.

[0023] In S2, dynamic feature decomposition is performed on the original time-series dataset to extract multiple intrinsic dynamic components that characterize the inherent evolution of force fluctuations. The energy contribution rate of each intrinsic dynamic component is then calculated, and the energy contribution rate forms the dynamic feature vector of the current pressing process. Specifically, this includes: In this embodiment, the original time-series dataset is first reconstructed. The delayed embedding dimension is set to ten, a value determined based on phase space reconstruction theory, used to expand the one-dimensional time series into a high-dimensional evolutionary trajectory. Specifically, starting from the first data point in the original time-series dataset, ten consecutive data points are sequentially selected to form the first row; then, one data point is slid to the right, and the second to eleventh data points are selected to form the second row; this process continues until all data points are covered, thus constructing a Hankel matrix with multiple rows and ten columns. Each row of this Hankel matrix represents a force-displacement evolution segment within a local time window, and all rows together constitute the high-dimensional evolutionary trajectory of force fluctuations on the time axis.

[0024] Subsequently, an orthogonal projection transformation is performed on the Hankel matrix to extract its dominant eigenspace. The specific process is as follows: First, the covariance matrix of the Hankel matrix is ​​calculated. Each element of this covariance matrix represents the degree of correlation between different columns in the Hankel matrix, reflecting the overall energy distribution characteristics of force fluctuations. Next, eigenvalue decomposition is performed on the covariance matrix to obtain a series of eigenvalues ​​and their corresponding eigenvectors. These eigenvalues ​​are arranged in descending order, and the eigenvectors corresponding to the five largest eigenvalues ​​are selected. These eigenvectors are then combined into an orthogonal projection operator. This orthogonal projection operator projects the original high-dimensional Hankel matrix into a low-dimensional eigenspace spanned by the dominant eigenvectors.

[0025] The Hankel matrix is ​​projected using the orthogonal projection operator described above to obtain a dimension-reduced representation matrix. This dimension-reduced representation matrix has the same number of rows as the Hankel matrix, but the number of columns is reduced to five. Each column represents a dominant evolutionary trend component, and each row reflects the intensity value of each dominant trend component at the corresponding time. The dimension-reduced representation matrix completely preserves the main dynamic evolutionary information in the original Hankel matrix while eliminating noise and secondary perturbations.

[0026] Next, a time-series evolution analysis is performed on the dimensionality-reduced representation matrix. Specifically, each column in the matrix is ​​analyzed independently, and the linear evolution relationship between each column and its own time-delay sequence is solved. By solving this linear relationship, the evolution coefficients corresponding to each column are obtained. These coefficients reflect the intrinsic evolution law of the trend component represented by that column on the time axis. Columns with significant evolution coefficients are extracted as intrinsic dynamic components. Each intrinsic dynamic component is a one-dimensional time series, representing a specific intrinsic evolutionary pattern of force fluctuations.

[0027] Finally, the energy contribution rate of each intrinsic dynamic component is calculated. Specifically, the sum of squares of each intrinsic dynamic component is calculated as its energy value; then, the energy values ​​of all intrinsic dynamic components are summed to obtain the total energy value; finally, the energy value of each intrinsic dynamic component is divided by the total energy value to obtain its energy contribution rate, which is a value between zero and one. The energy contribution rates of the five intrinsic dynamic components are arranged in corresponding order to form a five-dimensional vector, which is the dynamic characteristic vector of the current pressing process. This dynamic characteristic vector quantitatively describes the inherent pattern of force fluctuations during the current pressing process in the form of a numerical distribution.

[0028] In S3, the dynamic feature vectors are compared with the pre-stored deformation recovery feature library and the real jamming feature library to obtain the deformation recovery matching degree and the real jamming matching degree, specifically including: Please see Figure 2 As shown, the construction process of the deformation recovery feature library is as follows: First, force-displacement time-series data were collected during the deformation recovery phase of the shell during historical press-fitting processes. The deformation recovery phase refers to the gradual recovery of residual deformation in the bearing bore after the thin-walled shell is released by vacuum adsorption, due to the elastic hysteresis effect of the material. Force-displacement data in this phase are characterized by low amplitude, long period, and an exponential decay trend. During data collection, continuous data segments corresponding to this phase were extracted from each historical press-fitting record to ensure data integrity and purity.

[0029] Then, each segment of force-displacement time-series data is processed according to the dynamic eigenvalue decomposition method described in S2. Specifically, each segment of time-series data is reconstructed into a Hankel matrix based on the same delay embedding dimension. An orthogonal projection transformation is performed on this Hankel matrix to obtain the principal feature space. The same number of intrinsic dynamic components are extracted from the principal feature space, and the energy contribution rate of each intrinsic dynamic component is calculated. These energy contribution rates are then arranged in a fixed order to form the deformation recovery feature vector of the historical sample. This process is repeated until all historical samples are processed, resulting in multiple deformation recovery feature vectors.

[0030] Next, cluster analysis is performed on the obtained deformation recovery feature vectors. This implementation uses the K-means clustering algorithm, with the following steps: First, set the number of clusters to five, and randomly select five deformation recovery feature vectors as initial cluster centers; second, calculate the Euclidean distance from each deformation recovery feature vector to the five cluster centers, and assign the feature vector to the cluster center with the smallest distance; third, for each cluster, calculate the arithmetic mean of all feature vectors within that cluster, and use this mean as the new cluster center; fourth, repeat steps two and three until the change in cluster centers is less than a preset convergence threshold, at which point five final cluster centers are obtained. These five final cluster centers are saved as a deformation recovery feature library.

[0031] Please see Figure 3 As shown, the construction process of the real card-blocking feature library is as follows: First, force-displacement time-series data were collected during the actual jamming stage of the historical pressing process. The actual jamming stage refers to the period of abnormally increased resistance during pressing caused by factors such as excessive interference fit, misalignment during assembly, or foreign object obstruction. Force-displacement data during this stage are characterized by high amplitude, short period, and an exponential growth trend. During data collection, continuous data segments corresponding to this stage were extracted from each historical pressing record.

[0032] Then, each segment of force-displacement time-series data is processed using the same dynamic feature decomposition method described in S2 to obtain the true jamming feature vectors of each historical sample. The parameters used in the processing, such as the delayed embedding dimension and the number of intrinsic dynamic components, are completely consistent with those used in the construction of the deformation recovery feature library to ensure the comparability of the feature vectors.

[0033] Next, the same clustering analysis was performed on the multiple real jamming feature vectors obtained. The K-means clustering algorithm was also used, with five clusters. After iterative calculation, five final cluster centers were obtained, and these five final cluster centers were saved as the real jamming feature library.

[0034] At this point, the two reference feature libraries have been constructed, each containing five cluster center vectors representing typical patterns of force fluctuations of this type.

[0035] The calculation process for deformation recovery matching degree and true jamming resistance matching degree is as follows: After obtaining the dynamic feature vector of the current pressing process, it is necessary to calculate its matching degree with the two feature libraries respectively.

[0036] First, calculate the deformation recovery matching degree. The specific steps are as follows: calculate the Euclidean distance between the current dynamic feature vector and the five deformation recovery cluster centers in the deformation recovery feature library. The formula for calculating the Euclidean distance is as follows: ;in, Indicates the current dynamic feature vector and the first The distance between the deformation recovery cluster centers is The value range is from 1 to 5; Represents the th element of the current dynamic feature vector. One portion, The value ranges from 1 to 5, corresponding to five energy contribution rates; Indicates the first The deformation recovery cluster center of the first Each component has a distance value. After calculation, five distance values ​​are obtained. to The minimum value among these five distance values ​​is taken as the deformation recovery matching degree, i.e., the deformation recovery matching degree. The smaller the matching degree value, the closer the current force fluctuation pattern is to the typical pattern in the deformation recovery feature library.

[0037] Next, the true card-stop matching degree is calculated. The specific steps are: calculate the Euclidean distance between the current dynamic feature vector and the five true card-stop cluster centers in the true card-stop feature library. The first... Each cluster center is denoted as Its weight is The distance calculation formula is as follows: ;in Indicates the current dynamic feature vector and the first The distance between the real Karnofsky Performances (KMP) cluster centers The value range is from 1 to 5; Represents the th element of the current dynamic feature vector. One component; Indicates the first The first real cardioid cluster center Each component has a distance value. After calculation, five distance values ​​are obtained. to The minimum value among these five distance values ​​is taken as the true card resistance matching degree, i.e., the true card resistance matching degree. The smaller the matching degree value, the closer the current force fluctuation pattern is to the typical pattern in the real jamming feature library.

[0038] Through the above calculations, two matching degree values ​​were obtained, namely the deformation recovery matching degree. Matching degree with actual card resistance These two matching degrees will be used in subsequent steps to determine the source of the current force fluctuation, thus providing a basis for differentiated control.

[0039] In S4, the deformation recovery matching degree is compared with the actual jamming matching degree. If the deformation recovery matching degree is greater than the actual jamming matching degree and exceeds the deformation recovery judgment threshold, the current force fluctuation is determined to originate from the deformation recovery of the shell, and deformation suppression control parameters with suppressed position correction and superimposed axial micro-vibration are generated. If the actual jamming matching degree is greater than the deformation recovery matching degree and exceeds the actual jamming judgment threshold, the current force fluctuation is determined to originate from assembly jamming, and jamming avoidance control parameters with activated position backtracking correction are generated, specifically including: In this embodiment, two judgment thresholds need to be preset: a deformation recovery judgment threshold and a true jamming judgment threshold. The deformation recovery judgment threshold is set as follows: multiple sets of samples known to be in the deformation recovery stage are selected from historical pressing data. The deformation recovery matching degree of each set of samples is calculated according to the method described in S3. The arithmetic mean of the deformation recovery matching degrees of all samples is calculated, and this arithmetic mean plus three times the standard deviation is used as the deformation recovery judgment threshold. The true jamming judgment threshold is set as follows: multiple sets of samples known to be in the true jamming stage are selected from historical pressing data. The true jamming matching degree of each set of samples is calculated. The arithmetic mean of the true jamming matching degrees of all samples is calculated, and this arithmetic mean plus three times the standard deviation is used as the true jamming judgment threshold.

[0040] After setting the threshold, the determination is made based on the deformation recovery matching degree and the actual jamming resistance matching degree calculated by S3. The specific determination logic is as follows: The first step is to compare the deformation recovery matching degree with the actual jamming matching degree, and simultaneously determine whether the deformation recovery matching degree exceeds the deformation recovery judgment threshold. If the deformation recovery matching degree is less than the actual jamming matching degree, and also less than the deformation recovery judgment threshold, then the current force fluctuation is determined to originate from the deformation recovery of the shell. It should be noted that a smaller matching degree value indicates a higher degree of similarity to the corresponding feature library. Therefore, the deformation recovery matching degree must be less than both the actual jamming matching degree and its own threshold to confirm that it belongs to the deformation recovery category.

[0041] The second step involves generating deformation suppression control parameters if the deformation is classified as deformation recovery. These parameters include two instructions: the first is a position correction suppression instruction, which prevents the impedance control algorithm from retracting its position based on force error; the second is a superimposed axial micro-vibration instruction, which controls the press head to superimpose periodic micro-vibrations with an amplitude of 0.02 mm and a frequency of 50 Hz in the feed direction to overcome the additional resistance caused by surface adhesion during the deformation recovery stage.

[0042] The third step is to compare the actual jamming resistance matching degree with the deformation recovery matching degree, and simultaneously determine whether the actual jamming resistance matching degree exceeds the actual jamming resistance judgment threshold. If the actual jamming resistance matching degree is less than the deformation recovery matching degree, and the actual jamming resistance matching degree is less than the actual jamming resistance judgment threshold, then the current force fluctuation is determined to be caused by assembly jamming.

[0043] The fourth step, if the problem is determined to be assembly jamming, generates jamming avoidance control parameters. These parameters include an activation position backtracking correction command, specifically: based on the current displacement value, reverse the position by 0.5 mm at a speed of 5 mm per second, then pause for 0.2 seconds, and then refeed at a low speed of 2 mm per second. During the refeeding process, the feed speed is adjusted proportionally according to the real-time force error until the force value returns to a safe range.

[0044] Fifth step: If neither the deformation recovery matching degree nor the actual jamming matching degree meets any of the above judgment conditions, that is, both are greater than their respective judgment thresholds, or the relationship between the two does not meet the category classification conditions, then it is judged as an unidentifiable force fluctuation state. At this time, default control parameters are generated, that is, the current feed speed is kept unchanged, no position correction or vibration superposition is performed, and only the current state is recorded for subsequent analysis.

[0045] In S5, based on deformation suppression control parameters or jamming avoidance control parameters, the feed speed and attitude of the current pressing process are adjusted in real time. The force-displacement data of the entire pressing process are used as new samples to optimize the corresponding deformation recovery feature library or real jamming feature library, specifically including: In this embodiment, the current pressing process is adjusted in real time based on the deformation suppression control parameters or jamming avoidance control parameters generated by S4. Specifically, when the deformation suppression control parameters are received, the pressing controller immediately pauses the position correction function module based on force error in the impedance control algorithm, and simultaneously sends a vibration superposition command to the servo driver, causing the pressing head to superimpose an axial periodic micro-amplitude vibration with an amplitude of 0.02 mm and a frequency of 50 Hz on the basis of the current feed speed. This vibration continues until the end of the deformation recovery stage. When the jamming avoidance control parameters are received, the pressing controller immediately interrupts the current feed and executes a position backtracking correction command: based on the current displacement value, it retreats 0.5 mm in the opposite direction at a speed of 5 mm per second, then pauses for 0.2 seconds, and then resumes feeding at a low speed of 2 mm per second. During the refeeding process, the feed speed is dynamically adjusted according to the deviation ratio between the real-time pressing force value and the preset target force value until the pressing force value falls back to the safe threshold range.

[0046] While completing the above real-time adjustments, the post-processing workflow for this pressing data is initiated for online optimization of the feature library. First, the force-displacement time-series data recorded throughout the entire pressing process is retrieved from memory. This data includes all sampling points from the start of the pressing head contacting the housing to the end of the pressing process. The retrieved raw time-series data is preprocessed: the first fifty data points before the pressing head has stabilized in contact with the housing during the initial pressing stage, and the last fifty data points after the pressing head has disengaged during the final pressing stage, are identified and removed. The continuous data segment during the stable pressing stage is retained and used as the pure pressing segment time-series data.

[0047] Then, the time-series data of the pure pressing section is processed using the same dynamic eigenvalue decomposition method described in S2. First, based on the same delay embedding dimension of ten as in S2, the time-series data of the pure pressing section is reconstructed into a Hankel matrix to be decomposed. This matrix has ten rows (nine fewer than the total number of data points) and ten columns, with each row consisting of ten consecutive data points. Second, the covariance matrix of the Hankel matrix to be decomposed is calculated, and eigenvalue decomposition is performed on this covariance matrix. The eigenvectors corresponding to the five largest eigenvalues ​​are selected in descending order of eigenvalues ​​and combined into an orthogonal projection operator. Third, this orthogonal projection operator is used to project the Hankel matrix to be decomposed, obtaining a dimension-reduced representation matrix. This matrix has the same number of rows as the Hankel matrix to be decomposed and five columns. Fourth, time-series evolution analysis is performed on each column of the dimension-reduced representation matrix to extract five intrinsic dynamic components. The fifth step is to calculate the sum of squares of the five intrinsic dynamic components as the energy value of each component, sum the energy values ​​of all components to obtain the total energy value, divide the energy value of each component by the total energy value to obtain the energy contribution rate of each component, and arrange these five energy contribution rates in a fixed order to form a new feature vector.

[0048] Based on the determination result of S4, a category label is added to the new feature vector. If step 4 determines that the current force fluctuation originates from the shell deformation recovery, then the new feature vector is labeled with the deformation recovery category; if S4 determines that the current force fluctuation originates from assembly jamming, then the new feature vector is labeled with the actual jamming category.

[0049] Next, the feature library is updated. Taking the deformation recovery feature library as an example, if a newly added feature vector is labeled as a deformation recovery category, the Euclidean distance between the newly added feature vector and the five existing cluster centers in the deformation recovery feature library is calculated. Specifically, the five components of the newly added feature vector are subtracted sequentially from the five components of each cluster center, the squares of the differences in each dimension are summed, and the square root is taken to obtain five distance values. These five distance values ​​are compared, and the cluster center corresponding to the minimum value is taken as the cluster center to which the newly added feature vector belongs.

[0050] After determining the affiliation, the affiliation cluster center is updated. The specific update method is as follows: Record the original number of samples for each cluster center. Let the original number of samples be N. After adding the new feature vector, the new number of samples becomes N plus one. Multiply each component of the original cluster center by N, add each component corresponding to the new feature vector, and divide the sum by N plus one to obtain the new component values ​​for each cluster center. Use this set of new component values ​​to form the updated cluster center, replacing the original cluster center, and store it in the deformation recovery feature library.

[0051] For newly added feature vectors labeled as true jamming categories, the distance between them and the existing cluster centers in the true jamming feature library is calculated in the same way to determine the affiliation relationship, and a weighted fusion update is performed to replace the original cluster centers with the updated cluster centers.

[0052] Through the above steps, after each assembly is completed, the corresponding feature library will absorb the actual data of this assembly, so that the cluster center gradually shifts towards the real distribution in actual production, thereby realizing the continuous optimization of the feature library.

[0053] The working principle of this invention is as follows: A raw time-series dataset is constructed by real-time acquisition of force and displacement values ​​during the pressing process; dynamic feature decomposition is performed on the raw time-series dataset to extract multiple intrinsic dynamic components and calculate the energy contribution rate of each component, thus forming a dynamic feature vector for the current pressing process; the dynamic feature vector is compared with a pre-stored deformation recovery feature library and a real jamming feature library to obtain the deformation recovery matching degree and the real jamming matching degree; based on the comparison results, if the deformation recovery matching degree is dominant and exceeds the deformation recovery judgment threshold, the force fluctuation is determined to originate from the shell. Deformation recovery is performed, and deformation suppression control parameters that suppress position correction and superimpose axial micro-vibration are generated. If the actual jamming matching degree is dominant and exceeds the actual jamming judgment threshold, the force fluctuation is determined to be caused by assembly jamming, and jamming avoidance control parameters that activate position backtracking correction are generated. The feed speed and attitude of the pressing process are adjusted in real time according to the generated control parameters, and the force-displacement data of the entire pressing process are used as new samples. After obtaining new feature vectors through the same dynamic feature decomposition, the categories are labeled according to the judgment results and integrated into the corresponding feature library to achieve online continuous optimization of the feature library.

[0054] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An adaptive calculation method for gearbox housing assembly parameters, characterized in that, Includes the following steps: S1: During the pressing process, the pressing force value and its corresponding displacement value are collected in real time at a fixed sampling frequency to construct the original time series dataset reflecting the changes of force and displacement over time; S2: Perform dynamic feature decomposition on the original time series dataset, extract multiple intrinsic dynamic components that represent the intrinsic evolution law of force fluctuation, and calculate the energy contribution rate of each intrinsic dynamic component. Use the energy contribution rate to construct the dynamic feature vector of the current pressing process. S3: Perform similarity measurements between the dynamic feature vector and the pre-stored deformation recovery feature library and the real jamming feature library to obtain the deformation recovery matching degree and the real jamming matching degree. S4: Compare the deformation recovery matching degree with the actual jamming matching degree. If the deformation recovery matching degree is greater than the actual jamming matching degree and exceeds the deformation recovery judgment threshold, it is determined that the current force fluctuation originates from the deformation recovery of the shell. The deformation suppression control parameters for suppressing position correction and superimposing axial micro-vibration are generated. If the actual jamming matching degree is greater than the deformation recovery matching degree and exceeds the actual jamming judgment threshold, then the current force fluctuation is determined to be caused by assembly jamming, and jamming avoidance control parameters for activation position backtracking correction are generated. S5: Based on the deformation suppression control parameters or jamming avoidance control parameters, adjust the feed speed and attitude of the current pressing process in real time, and use the force-displacement data of the entire pressing process as new samples to optimize the corresponding deformation recovery feature library or real jamming feature library respectively.

2. The adaptive calculation method for gearbox housing assembly parameters according to claim 1, characterized in that, S2 specifically includes: Based on the preset delayed embedding dimension, the original time series dataset is reconstructed into a Hankel matrix that reflects the evolution trajectory of force fluctuations; The Hankel matrix is ​​orthogonally projected to solve the principal feature space that characterizes the dynamic evolution of force fluctuations, and multiple intrinsic dynamic components are extracted from the principal feature space. Calculate the proportion of each intrinsic dynamic component in the total energy, and use the proportion to construct the dynamic characteristic vector of the current pressing process.

3. The adaptive calculation method for gearbox housing assembly parameters according to claim 2, characterized in that, The extraction of multiple intrinsic dynamic components from the main feature space specifically includes: Covariance operations are performed on the Hankel matrix to construct a covariance matrix reflecting the energy distribution of force fluctuations, and eigenvalue decomposition is performed on the covariance matrix to obtain the orthogonal projection operator. By using the orthogonal projection operator to project the Hankel matrix onto a low-dimensional feature space, a dimension-reduced characterization matrix reflecting the evolutionary trend of force fluctuations is obtained. Temporal evolution analysis was performed on the dimensionality-reduced representation matrix to extract the intrinsic dynamic components of the intrinsic evolution law of multiple representation force fluctuations.

4. The adaptive calculation method for gearbox housing assembly parameters according to claim 1, characterized in that, The process of constructing the deformation recovery feature library is as follows: Force-displacement time-series data of the shell deformation recovery stage during historical press fitting are collected. Dynamic feature decomposition is performed on the force-displacement time-series data to obtain the deformation recovery feature vector of each historical sample. Cluster analysis is performed on each deformation recovery feature vector to obtain multiple deformation recovery cluster centers. The deformation recovery feature library is constructed using the deformation recovery cluster centers.

5. The adaptive calculation method for gearbox housing assembly parameters according to claim 1, characterized in that, The construction process of the real card-blocking feature library is as follows: Force-displacement time-series data of actual jamming stages during historical pressing processes are collected. Dynamic feature decomposition is performed on the force-displacement time-series data to obtain the actual jamming feature vectors of each historical sample. Cluster analysis is performed on each actual jamming feature vector to obtain multiple actual jamming cluster centers. The actual jamming feature library is constructed using the actual jamming cluster centers.

6. The adaptive calculation method for gearbox housing assembly parameters according to claim 1, characterized in that, The process of obtaining the deformation recovery matching degree and the actual jamming resistance matching degree specifically includes: Calculate the distance between the dynamic feature vector and each deformation recovery cluster center in the deformation recovery feature library, and use the minimum distance as the deformation recovery matching degree. Calculate the distance between the dynamic feature vector and each real card block cluster center in the real card block feature library, and use the minimum distance as the real card block matching degree.

7. The adaptive calculation method for gearbox housing assembly parameters according to claim 1, characterized in that, S5 specifically includes: The force-displacement time series data of the entire pressing process were subjected to the same dynamic feature decomposition to obtain new feature vectors, and the corresponding deformation recovery category or real jamming category was labeled for the new feature vectors. Calculate the distance between the newly added feature vector and each existing cluster center in the corresponding feature library, and determine the new feature vector to belong to the nearest existing cluster center based on the distance; The newly added feature vectors are merged into the existing cluster centers to which they belong, the coordinate values ​​of the existing cluster centers are updated, and the original cluster centers are replaced with the updated cluster centers to complete the optimization of the corresponding feature library.

8. The adaptive calculation method for gearbox housing assembly parameters according to claim 7, characterized in that, The process of obtaining new feature vectors by performing the same dynamic feature decomposition on the force-displacement time-series data of the entire pressing process specifically includes: Complete force-displacement time series data from the start to the end of the pressing process were extracted, and abnormal data points in the unstable contact stages at the beginning and end were removed to obtain pure pressing segment time series data. The pure pressing section time series data is reconstructed into the Hankel matrix to be decomposed, and the same orthogonal projection transformation is performed on the Hankel matrix to be decomposed to obtain the principal feature space to be decomposed. Extract multiple intrinsic dynamic components of the same number as S2 from the principal feature space to be decomposed, calculate the energy contribution rate of each intrinsic dynamic component, and use the energy contribution rate to form a new feature vector.

Images