An on-line monitoring method for a forging process of a variable cross-section automotive transmission member

Abstract

This invention relates to the field of metal plastic forming technology, specifically disclosing an online monitoring method for the forging process of automotive variable cross-section transmission parts. The method includes: collecting historical forging data from the production line and identifying defective batches; predicting the folding tendency of the current batch based on time-series analysis; conducting flow rate difference-folding correlation analysis on high-risk batches to determine the main detection areas; determining whether a first-level folding warning is triggered after threshold comparison; if triggered, simultaneously collecting forging temperature, establishing a linear folding coupling formula, constructing a linear folding deep learning model, generating the current theoretical flow rate difference, and making a second-level folding warning trigger decision; if triggered, analyzing the correlation between temperature and flow rate, constructing an adaptation model to assess the feasibility of temperature adjustment, and executing temperature control or manual intervention as needed. This method achieves predictive identification of linear folding defects, hierarchical and accurate early warning, and efficient targeted intervention, effectively avoiding batch defects and significantly reducing scrap rate and production losses.

Description

Technical Field

[0001] This invention relates to the field of metal plastic forming technology, specifically to an online monitoring method for the forging process of automotive variable cross-section transmission components. Background Technology

[0002] Variable cross-section transmission components are core and critical parts of automotive transmission systems. Their unique structure gives them advantages such as high transmission efficiency and lightweight design, making them widely used in various passenger and commercial vehicles. With the rapid development of the automotive industry towards lightweighting and high performance, significant progress has been made in the deployment of automated production lines and the improvement of production efficiency in areas such as high-quality titanium forgings and magnesium alloy forgings, indicating a very broad application prospect.

[0003] However, high-quality automotive variable cross-section transmission components, such as those made of titanium and magnesium alloys, have complex structures including large and small cross-section ends and slope transition sections. During forging, metal flow is prone to velocity imbalance, leading to frequent linear folding defects. Currently, the industry faces two major technical challenges in monitoring and controlling these defects: First, existing monitoring methods rely heavily on post-production quality inspection or single process parameter testing, lacking in-depth analysis of historical data and temporal pattern mining. This makes it difficult to predict and identify the occurrence trend of linear folding defects, resulting in low early warning accuracy and a high risk of batch defects. Second, existing intervention measures lack specificity, often involving blind shutdowns for debugging or uniform adjustments to process parameters. A quantitative correlation model between process parameters, temperature, and defect occurrence has not been established, making it impossible to develop appropriate intervention plans based on actual risk conditions. This easily leads to increased production losses and decreased production efficiency. This technical problem is particularly prominent in the manufacturing of high-quality titanium forgings and magnesium alloy forgings, severely hindering the improvement of yield rates and industrialization of high-end automotive variable cross-section transmission components.

[0004] Therefore, the present invention provides an online monitoring method for the forging process of automotive variable cross-section transmission components. Summary of the Invention

[0005] The purpose of this invention is to provide an online monitoring method for the forging process of automotive variable cross-section transmission components, so as to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions: An online monitoring method for the forging process of automotive variable cross-section transmission components includes: Acquire historical forging data of automotive variable cross-section transmission components, calibrate linear folding forging batches, analyze the time characteristics of linear folding forging batches, and predict the trend of current forging batch characteristics. If the current forging cycle characteristics tend to be linear folding, obtain historical parameter data of linear folding forging batches, perform flow rate difference folding batch analysis, determine the main detection areas, and simultaneously collect the current metal flow rate difference of the main detection areas. Through metal flow rate difference comparison analysis, determine whether a first-level folding warning has been triggered. If triggered, the current forging temperature data is collected synchronously, a linear folding coupling formula is established, a linear folding deep learning model is constructed, the current theoretical metal flow rate difference is determined, and a folding secondary warning trigger decision is made. If triggered, a temperature-flow rate difference correlation analysis is performed to determine the temperature-flow rate difference model. Combined with the current forging temperature, a feasibility analysis of forging temperature adjustment is conducted. Based on the results of the feasibility analysis, a forging intervention operation is executed.

[0007] As a further aspect of the present invention: Furthermore, the process of calibrating the linear folding forging batch and analyzing its time characteristics is as follows: If a linear folding band appears in the automotive variable cross-section transmission parts within a production batch during forging, the production batch will be marked as a linear folding forging batch. The linear folding sequence is obtained by arranging the number of linear folding bands in each batch of historical forging data according to the production time sequence. Identify consecutive batches with a non-zero number of folds within a sequence and define them as high-frequency batch clusters; The number of continuous batches in high-incidence batch clusters and the number of blank batches between adjacent clusters are counted separately, and the fluctuation ranges are extracted to obtain the first fluctuation range and the second fluctuation range. Autocorrelation analysis is used to calculate the autocorrelation coefficient of the sequence to determine the pattern of folded batch occurrences. If the autocorrelation coefficient of a certain lag order is significantly greater than 0, it is determined that the linear folding forging batch has periodicity and the lag order is the core period. Conversely, it is determined to be non-periodic.

[0008] Furthermore, the process for predicting the current forging batch characteristics is as follows: Real-time acquisition of the preceding time sequence of the current production batch: Taking the current batch as the endpoint, extract the linear folded sequence of the previous N batches, where N is the sum of the largest interval batch and the largest continuous batch of the historical high-occurrence batch cluster; Extract features from the preceding time sequence, including the number of blank batches from the previous high-incidence batch cluster; if the sequence has already shown consecutive non-zero batches, simultaneously count the number of current continuous batches. Set a tendency judgment condition. If any judgment condition is met, the current forging cycle characteristics are determined to be inclined to linear folding.

[0009] Furthermore, the process of setting the conditions for biased judgments is as follows: The first condition for a biased judgment is that the linear folding forging batches are not periodic, and the current continuous batch quantity is within the first fluctuation range. The second condition for a biased judgment is that the linear folding forging batches are not periodic and the number of blank batches is within the second fluctuation range. The third condition for a biased judgment is that the linear folding forging batches are periodic and the current production batch is within the core cycle. If the core period is T, and the interval between the current production batch and the starting batch of the previous linearly folded high-frequency batch cluster is equal to T, then the current production batch is determined to be within the core period.

[0010] Further, the process of performing flow rate difference folding batch analysis to determine the main detection sites is as follows: The historical parameter data of the linear folding forging batches were integrated according to different parts of the variable cross-section transmission component to obtain several historical data sequences of metal flow velocity difference. Calculate the mean of metal velocity difference and the frequency of linear folding in the historical data series of metal velocity difference; Among them, the linear folding frequency is the ratio of the number of linear folds in the linear folding forging batch to the total number of flow rate difference data; The components of the variable cross-section transmission component are arranged in descending order according to the mean flow velocity difference and the linear folding frequency, respectively, to obtain the metal flow velocity difference component sequence and the linear folding component sequence. Calculate the percentage of identical component positions in the two sequences and compare it with a threshold: If the proportion is greater than or equal to the threshold, it is determined that the metal flow rate difference is proportional to the probability of linear folding. The location of the first identical component in the sequence of metal flow rate difference components and the sequence of linear folding components is taken as the main detection area.

[0011] Furthermore, the process of determining whether a first-level folding warning has been triggered through comparative analysis of metal flow velocity differences is as follows: If the current metal flow rate difference is greater than or equal to the metal flow rate difference threshold, then trigger a first-level warning. The process for obtaining the metal flow rate difference threshold is as follows: Screen all forging samples with linear folding defects and locate the sampling point where the linear folding band first appears in each sample; Among them, the linear fold defect forging sample refers to a single automotive variable cross-section transmission component within a linear fold forging batch that exhibits a linear fold band. Extract the measured data of metal flow velocity difference of the main detection part corresponding to the time sequence node before the location acquisition point, and take the arithmetic mean to obtain the metal flow velocity difference threshold.

[0012] Furthermore, the characteristic is that the process of establishing the linear folding coupling formula is as follows: Constructing a linear folding coupling formula: ; Where a is a constant term, The main detection area is the velocity difference across the variable cross-section. , , For coupling weight terms, The average flow velocity at the large cross-section end. It is the cross-sectional ratio. The slope angle, This is the error term.

[0013] Furthermore, the process of constructing a linear folding deep learning model, determining the current theoretical metal flow velocity difference, and making a folding secondary warning trigger decision is as follows: A hybrid deep learning model of LSTM-CNN is constructed using the linear folding coupling formula as a framework. Input the average flow velocity, cross-section ratio, and slope angle at the large cross-section end, and output the predicted value of the flow velocity difference at the main detection location. Extract the corresponding input data within the statistical time period and divide it into training, validation, and test sets in a 7:2:1 ratio; The mean squared error is used as the loss function, and the average absolute error between the predicted and actual values ​​of the velocity difference at the main detection sites is used as the accuracy index for iterative training. Training is stopped and the model is saved when the mean absolute error is less than or equal to 0.1 mm / s. Input the current average flow velocity, cross-sectional ratio, and slope angle at the large cross-section end into the model, and output the current theoretical metal flow velocity difference; If the current theoretical metal flow rate difference is greater than or equal to the metal flow rate difference threshold, a level 2 warning will be triggered.

[0014] Furthermore, the process of performing temperature-flow-velocity difference correlation analysis and determining the temperature-flow-velocity difference model is as follows: Historical metal flow rate difference and forging temperature data of the main detection sites were extracted and aligned according to the time sequence of the acquisition interval to obtain the historical metal flow rate difference sequence and the historical forging temperature sequence, respectively. The differences between adjacent data in the historical metal flow rate difference sequence and the historical forging temperature sequence are calculated separately to generate the metal flow rate difference change sequence and the forging temperature difference sequence. The Pearson correlation coefficient between the metal flow rate difference sequence and the forging temperature difference sequence was calculated and its absolute value was taken to obtain the correlation coefficient. If the correlation coefficient is greater than or equal to the correlation coefficient threshold, a univariate linear model is fitted by the least squares method to obtain the temperature-flow-rate difference model. If the correlation coefficient is less than the correlation coefficient threshold, nonlinear models are constructed respectively: binary linear model, higher-order polynomial model, exponential model, power function model, and trigonometric function model. After fitting by the least squares method, the model with the largest determination coefficient is selected as the temperature-flow-rate difference model.

[0015] Furthermore, a feasibility analysis was conducted on forging temperature adjustment. Based on the results of the feasibility analysis, the process of implementing forging intervention was as follows: Calculate the difference between the current forging temperature and the maximum / minimum value of the forging temperature range, and substitute it into the temperature-flow-rate difference model to obtain the theoretical range of metal flow-rate difference variation. The forging temperature range refers to the allowable temperature range that meets the requirements of the forging process of automotive variable cross-section transmission parts. The adjustment range of the current metal flow rate difference is obtained by adding the current theoretical metal flow rate difference with the maximum and minimum values ​​of the theoretical metal flow rate difference variation range. If the adjustment range includes the metal flow rate difference threshold, it indicates that forging temperature adjustment is highly feasible, and forging intervention should be performed. Calculate the difference between the current theoretical metal flow rate difference and the metal flow rate difference threshold, substitute it into the temperature flow rate difference model to obtain the current forging temperature adjustment amount, and adjust the current forging temperature according to the current forging temperature adjustment amount. If the adjustment range does not include the metal flow rate difference threshold, it indicates that the feasibility of forging temperature adjustment is low, and manual intervention is required.

[0016] The beneficial effects of this invention are: (1) Improve the accuracy of early warning of linear folding defects in automotive variable cross-section transmission components and effectively avoid batch defects; by collecting historical forging full data, deeply analyze the characteristics of linear folding defects and batch timing patterns, and identify high-risk batches; then combine the linear folding deep learning model, integrate process, geometry and temperature parameters to construct quantitative correlation, generate theoretical metal flow rate difference as early warning benchmark, and realize early prediction and graded accurate early warning of defects. (2) Enhance the targeting of intervention measures during the forging process to reduce production and efficiency losses; construct a suitable temperature and flow rate difference model by analyzing the correlation characteristics between temperature and metal flow rate difference, and evaluate the feasibility of temperature adjustment; formulate graded intervention plans for different risk scenarios, implement precise temperature control when feasible, and take manual intervention when not feasible, avoiding the drawbacks of blind shutdown and unified debugging in traditional intervention, and reducing scrap rate and ineffective production losses while ensuring the stability of forging quality. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1This is a schematic diagram of an online monitoring method for the forging process of automotive variable cross-section transmission components; Figure 2 This is a logic diagram for an online monitoring method of the forging process of automotive variable cross-section transmission components. Detailed Implementation

[0019] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0020] Please see Figure 1 - Figure 2 As shown, this invention is an online monitoring method for the forging process of automotive variable cross-section transmission components. This invention primarily addresses the specific problems of linear folding defects easily caused by factors such as metal flow velocity imbalance due to the variable cross-section structure during the forging process of automotive variable cross-section transmission components, and the difficulty of existing monitoring methods in predicting and identifying defects and the lack of targeted intervention measures. First, based on historical forging data, defect characteristics are analyzed to predict the folding tendency of the current batch. Then, for high-risk batches, a first-level folding warning is triggered through flow velocity difference analysis. Next, theoretical flow velocity differences are generated by combining temperature data and a deep learning model to trigger a second-level folding warning. Finally, based on temperature-flow velocity correlation analysis, temperature adjustment or multi-dimensional manual intervention is performed as needed. Ultimately, this achieves full-process online monitoring, accurate early warning, and efficient intervention for linear folding defects in the forging process of automotive variable cross-section transmission components, effectively reducing scrap rate and production losses. Specifically, it includes the following steps: Step 1: Obtain historical forging data of automotive variable cross-section transmission components, conduct forging result characteristic analysis, calibrate linear folding forging batches, analyze the time characteristics of linear folding forging batches, and predict the trend of current forging batch characteristics. In step one, the historical forging data refers to: within the statistical period (the time period from the start of production on the current production line to the current moment), the batch identification data (a unique and traceable production batch number to ensure the continuity of batch sequence) of each production batch in the forging process of automotive variable cross-section transmission parts, the existence data of linear folding strips after forging of automotive variable cross-section transmission parts in each production batch (i.e., the determination result of whether linear folding strips appear in the forgings of this batch), and the number of automotive variable cross-section transmission parts with linear folding strips in each production batch. Historical forging data can be obtained by collecting the unique batch number of each forging production batch from the automotive variable cross-section transmission component forging production execution system (MES); For finished forgings from each production batch, the determination and quantity of linear folded strips are completed through visual inspection / manual verification, specifically as follows: A special test for linear folding bands was conducted on all forgings within the batch, and the presence of linear folding bands in the batch was recorded to form existence determination data. For batches found to contain linear folding bands, the specific number of forgings containing linear folding bands is counted to form a quantity statistics report; The existence determination data and quantity statistics of each batch are bound and archived with the corresponding batch identification data to form a complete historical forging data ledger. In step one, the process of calibrating the linear folding forging batch is as follows: If a linear folding band appears in the automotive variable cross-section transmission parts within a production batch during forging, the production batch will be marked as a linear folding forging batch. If the automotive variable cross-section transmission parts in the production batch are normal during forging, no marking is required. In step one, the process of obtaining the time characteristics of the linear folding forging batch is as follows: The number of automotive variable cross-section transmission parts with linear folding bands in any linear folding forging batch is counted and defined as the number of linear folds in the linear folding forging batch. During forging, the number of linear folds in a normal production batch is 0. The number of linear folds in each production batch in the historical forging data is arranged sequentially according to the production time to obtain the linear fold sequence; A cluster of consecutive batches (not appearing alone) with a non-zero number of folds in a linear folding sequence is defined as a high-frequency batch cluster. Count the number of consecutive batches in all high-incidence batch clusters and extract their fluctuation range (first fluctuation range). For example, in the linear folded sequence [5,4,0,0,1,6,5,0,0,0,0,0,0,8,5,2,4,0,0,0……]: 5,4 represents 2 consecutive batches, 8,5,2,4 represents 4 consecutive batches, and the fluctuation range is 2-4 batches.

[0021] Count the number of blank batches between two adjacent high-frequency batch clusters in the linear folded sequence (i.e., the number of batches with 0 consecutive folds between adjacent high-frequency batch clusters, such as 5,4 and 1,6,5 with 2 batches between them, and 1,6,5 and 8,5,2,4 with 6 batches between them), and extract the fluctuation range of the number of blank batches (second fluctuation range) (e.g., the fluctuation of the number of blank batches is between 2 and 6 batches). The autocorrelation coefficient of the linear folded sequence is calculated using autocorrelation analysis (ACF), and its occurrence pattern is analyzed: If the autocorrelation coefficient of a certain lag order (such as lag 15 batches or 20 batches) is significantly greater than 0 (P<0.05), then it is determined that the linear folding forging batches have periodicity, and this lag order is the core period. If no autocorrelation coefficient significantly greater than 0 is found (P<0.05), the linear folding forging batch is determined to be non-periodic. In step one, the process of predicting the tendency of the current forging cycle characteristics is as follows: Real-time acquisition of the preceding time sequence of the current production batch: Using the current batch as the endpoint, extract the linear folded sequence of the preceding N batches (N is the maximum interval batch + the maximum duration batch of historically high-frequency batch clusters, such as N=6+4=10 in this case), and extract the sequence features: It should be noted that N is the sum of the largest interval batch and the largest continuous batch in the historical high-incidence batch cluster. This ensures that the preceding time series sequence can completely cover the time characteristics of high defect incidence and avoids prediction bias caused by incomplete sequence truncation.

[0022] The number of blank batches in the preceding time sequence (i.e., the interval batches from the previous high-frequency batch cluster); If the preceding time series sequence has already shown consecutive non-zero batches, count the number of the current continuous batches; Preferred judgment condition 1: If the linear folding forging batches do not have a periodic occurrence pattern, the current continuous batch number in the preceding time sequence is within the first fluctuation range; Second biased judgment condition: If the linear folding forging batches do not have a periodic occurrence pattern, the number of blank batches in the preceding time sequence is within the second fluctuation range; Preferred judgment condition three: If the linear folding forging batches exhibit a periodic occurrence pattern, and the current production batch is located within the core cycle; Among them, if the core period is T, and the interval between the current production batch and the starting batch of the previous linear folded high-frequency batch cluster is equal to T, the current production batch is determined to be within the core period. If any of the tendency judgment conditions are met, the current forging cycle characteristics are determined to tend towards linear folding; If none of the predisposition judgment conditions are met, the current forging cycle characteristics are determined to be normal. Step 2: If the current forging cycle characteristics tend to be linear folding, obtain historical parameter data of linear folding forging batches, perform flow rate difference folding batch analysis, determine the main detection areas, and simultaneously collect the current metal flow rate difference of the main detection areas. Through metal flow rate difference comparison analysis, determine whether a first-level folding warning has been triggered. In step two, the historical parameter data of the linear folding forging batch refers to the flow velocity difference data of each part (large cross-section end, small cross-section end, slope transition section, etc.) of the variable cross-section transmission component within each production batch; Historical parameter data can be obtained by deploying high-speed cameras to perform non-contact velocity measurement on the metal surface of various parts during the forging process, capturing the metal flow velocity in real time at different locations such as the large cross-section end, small cross-section end, and slope of the variable cross-section, and then processing the ratio with the acquisition interval to obtain the flow velocity difference data. Metal flow rate difference data of linear folding forging batches were extracted from the Manufacturing Execution System (MES) for forging of automotive variable cross-section transmission components and integrated according to different parts of the variable cross-section transmission components to obtain several historical data sequences of metal flow rate difference. In step two, the flow velocity difference folding batch analysis refers to the correlation batch analysis between flow velocity difference and linear folding defects. In step two, the process of performing flow rate difference folding batch analysis to determine the main detection locations is as follows: Calculate the mean value of metal velocity difference based on any historical data sequence of metal velocity difference. Arrange the various parts of the variable cross-section transmission component in descending order of the corresponding average metal flow velocity difference to obtain the metal flow velocity difference component sequence; Calculate the linear folding frequency based on any historical data sequence of metal flow velocity difference; Linear folding frequency = number of linear folds in the linear folding forging batch / total number of metal flow rate difference data; The total number of metal velocity difference data refers to the number of data points in the historical data sequence of metal velocity difference. Arrange the various parts of the variable cross-section transmission component in descending order of their corresponding linear folding frequencies to obtain a sequence of linear folding components; Calculate the proportion of identical component positions in the metal flow rate difference component sequence and the linear folding component sequence; For example, in the slope transition section, the area (A), the upper area (B), the large cross-section end sidewall (C), the lower area (D), the small cross-section end sidewall (E), the large cross-section end face (F), and the small cross-section end face (G), the sequence of metal flow velocity difference components is (A, B, D, C, E, F, G), and the sequence of linear folding components is (A, B, C, D, E, F, G). Among the 7 parts, A, B, E, F, and G are completely consistent in position, and the proportion of identical component positions is approximately 71.4% (5 / 7).

[0023] Compare the proportion of identical component locations with a proportion threshold: The percentage threshold is set according to the relevant clauses on forging forming defect control in Chapter 5 "Technical Requirements" of GB / T 33520-2017 "T-type end face teeth of transmission shafts"; If the proportion of the same component location is less than the proportion threshold, it indicates that the difference in metal flow velocity is not proportional to the probability of linear folding. The statistical time period should be extended and the proportion of the same component location should be recalculated. If the proportion of the same component position is greater than or equal to the proportion threshold, it indicates that the difference in metal flow velocity is proportional to the probability of linear folding. The location of the first identical component in the metal flow rate difference component sequence and the linear folding component sequence is used as the main detection area; In (A, B, D, C, E, F, G) and (A, B, C, D, E, F, G), A is the variable cross-section transmission part of the first identical component at the same component location: it is the region (A) in the slope transition section.

[0024] In step two, the process of determining whether a first-level folding warning has been triggered through metal flow rate difference comparison and analysis is as follows: Compare the current metal flow rate difference at the main detection site with the metal flow rate difference threshold: The method for obtaining the metal flow rate difference threshold is as follows: By traversing the entire forging process of historical automotive variable cross-section transmission components, all forging samples that have experienced linear folding defects were screened and identified, and a historical sample set of linear folding defects was established. For each defective sample, the acquisition time and acquisition point corresponding to the first identification of the linear folding zone during its forging process are located. Based on the first sampling point of the linear folding band in each defect sample, trace back to the previous time sequence sampling node immediately adjacent to the sampling point, and extract the measured data of metal flow velocity difference (critical state data before the linear folding defect is about to occur) of the corresponding main detection part under that node. The metal flow rate difference threshold is obtained by arithmetically averaging the measured data of the metal flow rate difference at the main detection sites. If the current metal flow rate difference is greater than or equal to the metal flow rate difference threshold, then trigger a first-level warning. If the current metal flow rate difference is less than the metal flow rate difference threshold, the first-level warning will not be triggered. It should be noted that if the current metal flow rate difference is greater than or equal to the metal flow rate difference threshold, the reason for triggering the first-level folding warning is that: the forging of automotive variable cross-section transmission parts is a metal plastic forming process. The variable cross-section structure will cause natural differences in the metal flow resistance and flow path in different parts. Uniform metal rheology is the core prerequisite for ensuring the forming quality of forgings. When the metal flow rate difference of the main detection part exceeds the threshold, step two supplements the warning causal mechanism, explaining that when the metal flow rate difference of the main detection part exceeds the threshold, the imbalance of metal flow rate will lead to local material accumulation and tensile tearing, directly forming a linear folding band, thus triggering the first-level warning.

[0025] The physical significance of triggering the first-level folding warning is that this warning is a pre-process preventive warning in the forging process, rather than a post-process alarm after the defect has occurred. Its main physical significance is to identify in real time that the forging process has deviated from the safe rheological process window and the current metal flow state has entered the high-risk range of linear folding defects, prompting production and process personnel to intervene in time to suppress the generation of linear folding defects from the source and reduce scrap rate and production loss.

[0026] Step 3: If a first-level folding warning is triggered, the current forging temperature data is collected synchronously, a linear folding coupling formula is established, a linear folding deep learning model is constructed, the current theoretical metal flow rate difference is determined, and a second-level folding warning trigger decision is made. In step three, the method for synchronously collecting the current forging temperature data can be: when a first-level folding warning is triggered, real-time data is collected through deployed contact thermocouples; In step three, the process of establishing the linear folding coupling formula is as follows: Constructing a linear folding coupling formula: ; Where a is a constant term, The variable cross-section flow velocity difference (the axial average flow velocity difference of metal from the large cross-section end to the small cross-section end) is the main detection point. , , For coupling weight terms, The average flow velocity at the large cross-section end (the actual average axial flow velocity of metal at the large cross-section end of a variable cross-section). It is the cross-sectional ratio (the ratio of the area of ​​the larger cross-section to the area of ​​the smaller cross-section in a variable cross-section). It is the slope angle (the angle between the slope section of the variable cross section and the axis of the part). This is the error term; The cross-section ratio and slope angle are directly read from the design drawings of automotive variable cross-section transmission components; It should be noted that the logic behind establishing the linear folding coupling formula is as follows: the constant term 'a' is used to correct the baseline deviation between the production line equipment and the process system; C1·v1 quantifies the driving effect of the flow velocity at the large cross-section end on the velocity difference of the variable cross-section; C2·(γ−1) reflects the amplification effect of the degree of cross-sectional contraction on flow resistance and velocity difference; (γ−1) is used because γ is the ratio of the large cross-section to the small cross-section, which is greater than 1, and subtracting it from 1 amplifies the characteristic of the ratio of the large cross-section to the small cross-section; C3·tanθ reflects the effect of the slope steepness on the metal flow path. The disturbance effect is represented by C1, which characterizes the driving response amplitude of the flow velocity difference at the large cross-section end to the flow velocity difference at the variable cross-section. C2 characterizes the amplification and control weight of the flow velocity difference by the degree of cross-section contraction. C3 reflects the intensity of the disturbance effect of the slope steepness on the metal flow and flow velocity difference. The coupling weight represents the contribution ratio of different variable cross-section geometric features and flow features to the flow velocity difference. tanθ refers to the quantitative value of the slope. The larger the tanθ, the more proportional it is to the size of θ (the slope is normally between 0-90°). The error term ε includes random disturbances such as billet fluctuation and lubrication changes.

[0027] In step three, the process of constructing the linear folding deep learning model is as follows: Using the linear folding coupling formula as a framework, a deep learning model is constructed to capture the coupling weight terms in the linear folding coupling formula: Input features: average flow velocity at the large cross-section end, cross-sectional ratio, and slope angle; Output characteristics: Predicted values ​​of velocity difference at the main detection locations; The average flow velocity, cross-section ratio, and slope angle of the large cross-section end within the statistical time period are extracted from the automotive variable cross-section transmission component forging production execution system (MES) and combined according to the same time period as a set of model inputs; The inputs of each model are divided into training set, validation set and test set in a ratio of 7:2:1. The training set is used for model parameter learning, the validation set is used for hyperparameter tuning, and the test set is used for final model accuracy verification. A hybrid deep learning model of LSTM-CNN was selected (CNN captures the spatial features of the flow velocity difference in different parts, and LSTM captures the temporal variation features of each input data). The model is iteratively trained using mean squared error (MSE) as the loss function (based on model parameter adjustment) and mean absolute error (MAE) as the accuracy evaluation index (based on iteration termination). The mean absolute error refers to the average absolute error between the predicted value of the variable cross-section flow velocity difference of the main detection part and the actual value of the variable cross-section flow velocity difference of the main detection part. When the mean absolute error is less than or equal to the error threshold (preferably 0.1 mm / s), stop the iteration and save the trained linear folding deep learning model; In step three, the process of determining the current theoretical metal flow rate difference is as follows: The current average flow velocity at the large cross-section end, the current cross-section ratio, and the current slope angle are collected as inputs to the linear folding deep learning model, and the current theoretical metal flow velocity difference is output. It should be noted that the beneficial effects of obtaining the current theoretical metal flow rate difference are as follows: obtaining the current theoretical metal flow rate difference can generate a benchmark value based on the quantitative mapping relationship between process and geometric parameters, effectively calibrating random interference in the actual flow rate difference measurement, and accurately quantifying the degree of process deviation from the safe rheological window by comparing with the measured value. This can not only assist in triggering the secondary early warning of folding and reduce the false alarm rate, but also provide a clear basis for adjusting process parameters. At the same time, combined with the temporal characteristics of deep learning models, it can realize risk prediction, improve the accuracy of linear folding defect early warning, and reduce production losses and scrap rate.

[0028] In step three, the process of triggering the folding level two early warning decision is as follows: Compare the current theoretical metal velocity difference with the metal velocity difference threshold; If the current theoretical metal flow rate difference is greater than or equal to the metal flow rate difference threshold, then a folded level two warning will be triggered. If the current theoretical metal flow rate difference is less than the metal flow rate difference threshold, the second-level warning will not be triggered. It should be noted that if the current theoretical metal flow rate difference is less than the metal flow rate difference threshold, the second-level warning will not be triggered. The reason is that the current theoretical metal flow rate difference is a benchmark value generated by a deep learning model based on process and geometric parameters such as the average flow rate at the large cross-section end, cross-sectional ratio, and slope angle. It reflects the expected safe rheological state under the current process parameter matching. When this theoretical value is less than the threshold, it means that from the quantitative mapping relationship between process input and geometric constraints, the current metal flow state is still within the safe rheological process window. The triggering of the first-level warning is more likely to be due to occasional interference in the actual measurement process (such as instantaneous equipment noise or local fluctuations in billet) rather than a systemic process deviation. Therefore, there is no need to upgrade to the second-level warning to avoid excessive intervention in the production rhythm and reduce ineffective downtime and debugging costs.

[0029] Step 4: If a level 2 folding warning is triggered, perform a temperature-flow rate difference correlation analysis to determine the temperature-flow rate difference model. Combined with the current forging temperature, conduct a feasibility analysis of forging temperature adjustment. Based on the feasibility analysis results, execute the forging intervention operation. In step four, the process of performing temperature-flow-rate difference correlation analysis is as follows: Historical metal flow rate difference data and historical forging temperature data of the main detection areas were extracted and integrated to obtain historical metal flow rate difference sequences and historical forging temperature sequences. It should be noted that the historical metal flow rate difference data and historical forging temperature data in the historical metal flow rate difference sequence and the historical forging temperature sequence are time-aligned according to the acquisition interval.

[0030] Calculate the difference between each adjacent historical metal flow rate difference in the historical metal flow rate difference sequence and the difference between each adjacent historical forging temperature in the historical forging temperature sequence, and integrate them into a metal flow rate difference change sequence and a forging temperature difference sequence, respectively. Calculate the Pearson correlation coefficient between the metal flow rate difference sequence and the forging temperature difference sequence, and take the absolute value to obtain the correlation coefficient; If the correlation coefficient is greater than or equal to the correlation coefficient threshold (preferably 0.7, a commonly used industry standard for determining linear correlation), it indicates that there is a linear correlation between the metal flow rate difference and the forging temperature. A univariate linear model is constructed by substituting the metal flow rate difference sequence and the forging temperature difference sequence into the model according to their corresponding relationship, and then fitting the model using the least squares method to obtain the temperature flow rate difference model. If the correlation coefficient is less than the correlation coefficient threshold, it indicates that there is no linear relationship between the metal flow rate difference and the forging temperature. Nonlinear models were constructed respectively: bivariate linear model, higher-order polynomial model, exponential model, power function model and trigonometric function model; The least squares method was used for fitting, the determination coefficient of the nonlinear model was calculated, and the nonlinear model corresponding to the maximum value of the determination coefficient was taken as the temperature-flow-rate difference model. In step four, a feasibility analysis of forging temperature adjustment is performed. Based on the results of the feasibility analysis, the process of executing the forging intervention operation is as follows: Calculate the difference between the current forging temperature and the maximum value of the forging temperature range (the allowable temperature range that meets the requirements of the forging process of automotive variable cross-section transmission parts), and substitute it into the temperature flow rate difference model to obtain the theoretical metal flow rate difference variation range. The adjustment range of the current metal flow rate difference is obtained by adding the current theoretical metal flow rate difference with the maximum and minimum values ​​of the theoretical metal flow rate difference variation range. The forging temperature range can be obtained by reading it from the current production line equipment. If the adjustment range includes the metal flow rate difference threshold, it indicates that the feasibility of forging temperature adjustment is high. Forging intervention operation is then performed (calculate the difference between the current theoretical metal flow rate difference and the metal flow rate difference threshold, substitute it into the temperature flow rate difference model to obtain the current forging temperature adjustment amount, and adjust the current forging temperature according to the current forging temperature adjustment amount). If the adjustment range does not include the metal flow rate difference threshold, it indicates that the feasibility of forging temperature adjustment is low, and manual intervention should be carried out (implementing comprehensive manual intervention, specifically including mold inspection, repair or replacement, manual adjustment of core process parameters, billet pretreatment optimization, and shutdown isolation and investigation when necessary). The working principle of this invention is as follows: Based on historical forging data, linear folding forging batches are calibrated, and their time characteristics are analyzed to predict the defect tendency of the current batch. If linear defects are expected, the main detection areas are first determined by correlating the flow rate difference with the folding frequency, and the current flow rate difference is compared with the critical threshold to trigger a first-level warning. After triggering, an LSTM-CNN model is constructed using a coupling formula as a framework, and key forging parameters are input to predict the theoretical flow rate difference and trigger a second-level warning. Finally, the correlation between the flow rate difference and forging temperature is analyzed to verify the feasibility of temperature adjustment, and corresponding interventions (automatic temperature adjustment or manual intervention) are executed to achieve precise defect prevention and control.

[0031] 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 present invention should still fall within the scope of the present invention.

Claims

1. An online monitoring method for the forging process of automotive variable cross-section transmission components, characterized in that: include: Acquire historical forging data of automotive variable cross-section transmission components, calibrate linear folding forging batches, analyze the time characteristics of linear folding forging batches, and predict the trend of current forging batch characteristics. If the current forging cycle characteristics tend to be linear folding, obtain historical parameter data of linear folding forging batches, perform flow rate difference folding batch analysis, determine the main detection areas, and simultaneously collect the current metal flow rate difference of the main detection areas. Through metal flow rate difference comparison analysis, determine whether a first-level folding warning has been triggered. If triggered, the current forging temperature data is collected synchronously, a linear folding coupling formula is established, a linear folding deep learning model is constructed, the current theoretical metal flow rate difference is determined, and a folding secondary warning trigger decision is made. If triggered, a temperature-flow rate difference correlation analysis is performed to determine the temperature-flow rate difference model. Combined with the current forging temperature, a feasibility analysis of forging temperature adjustment is conducted. Based on the results of the feasibility analysis, a forging intervention operation is executed.

2. The online monitoring method for the forging process of a variable cross-section transmission component for automobiles according to claim 1, characterized in that: The process of calibrating linear folding forging batches and analyzing their time characteristics is as follows: If a linear folding band appears in the automotive variable cross-section transmission parts within a production batch during forging, the production batch will be marked as a linear folding forging batch. The linear folding sequence is obtained by arranging the number of linear folding bands in each batch of historical forging data according to the production time sequence. Identify consecutive batches with a non-zero number of folds within a sequence and define them as high-frequency batch clusters; The number of continuous batches in high-incidence batch clusters and the number of blank batches between adjacent clusters are counted separately, and the fluctuation ranges are extracted to obtain the first fluctuation range and the second fluctuation range. Autocorrelation analysis is used to calculate the autocorrelation coefficient of the sequence to determine the pattern of folded batch occurrences. If the autocorrelation coefficient of a certain lag order is significantly greater than 0, it is determined that the linear folding forging batch has periodicity and the lag order is the core period. Conversely, it is determined to be non-periodic.

3. The online monitoring method for the forging process of a variable cross-section transmission component for automobiles according to claim 2, characterized in that: The process for predicting the tendency of the current forging batch characteristics is as follows: Real-time acquisition of the preceding time sequence of the current production batch: Taking the current batch as the endpoint, extract the linear folded sequence of the previous N batches, where N is the sum of the largest interval batch and the largest continuous batch of the historical high-occurrence batch cluster; Extract features from the preceding temporal sequence, including the number of blank batches from the current time to the previous high-incidence batch cluster; If the sequence has already shown consecutive non-zero batches, the current number of consecutive batches is counted simultaneously. Set a tendency judgment condition. If any judgment condition is met, the current forging cycle characteristics are determined to be inclined to linear folding.

4. The online monitoring method for the forging process of a variable cross-section transmission component for automobiles according to claim 3, characterized in that: The process of setting the conditions for biased judgment is as follows: The first condition for a biased judgment is that the linear folding forging batches are not periodic, and the current continuous batch quantity is within the first fluctuation range. The second condition for a biased judgment is that the linear folding forging batches are not periodic and the number of blank batches is within the second fluctuation range. The third condition for a biased judgment is that the linear folding forging batches are periodic and the current production batch is within the core cycle. If the core period is T, and the interval between the current production batch and the starting batch of the previous linearly folded high-frequency batch cluster is equal to T, then the current production batch is determined to be within the core period.

5. The online monitoring method for the forging process of a variable cross-section transmission component for automobiles according to claim 1, characterized in that: The process of performing flow velocity difference folding batch analysis to determine the main detection sites is as follows: The historical parameter data of the linear folding forging batches were integrated according to different parts of the variable cross-section transmission component to obtain several historical data sequences of metal flow velocity difference. Calculate the mean of metal velocity difference and the frequency of linear folding in the historical data series of metal velocity difference; Among them, the linear folding frequency is the ratio of the number of linear folds in the linear folding forging batch to the total number of flow rate difference data; The components of the variable cross-section transmission component are arranged in descending order according to the mean flow velocity difference and the linear folding frequency, respectively, to obtain the metal flow velocity difference component sequence and the linear folding component sequence. Calculate the percentage of identical component positions in the two sequences and compare it with a threshold: If the proportion is greater than or equal to the threshold, it is determined that the metal flow rate difference is proportional to the probability of linear folding. The location of the first identical component in the sequence of metal flow rate difference components and the sequence of linear folding components is taken as the main detection area.

6. The online monitoring method for the forging process of a variable cross-section transmission component for automobiles according to claim 1, characterized in that: The process of determining whether a first-level folding warning has been triggered by comparing and analyzing the metal flow velocity difference is as follows: If the current metal flow rate difference is greater than or equal to the metal flow rate difference threshold, then trigger a first-level warning. The process for obtaining the metal flow rate difference threshold is as follows: Screen all forging samples with linear folding defects and locate the sampling point where the linear folding band first appears in each sample; Among them, the linear fold defect forging sample refers to a single automotive variable cross-section transmission component within a linear fold forging batch that exhibits a linear fold band. Extract the measured data of metal flow velocity difference of the main detection part corresponding to the time sequence node before the location acquisition point, and take the arithmetic mean to obtain the metal flow velocity difference threshold.

7. The online monitoring method for the forging process of a variable cross-section transmission component for automobiles according to claim 1, characterized in that: The process of establishing the linear folding coupling formula is as follows: Constructing a linear folding coupling formula: ； Where a is a constant term, The main detection area is the velocity difference across the variable cross-section. , , For coupling weight terms, The average flow velocity at the large cross-section end. It is the cross-sectional ratio. The slope angle, This is the error term.

8. The online monitoring method for the forging process of an automotive variable cross-section transmission component according to claim 7, characterized in that: The process of constructing a linear folding deep learning model, determining the current theoretical metal flow velocity difference, and making a folding secondary warning trigger decision is as follows: A hybrid deep learning model of LSTM-CNN is constructed using the linear folding coupling formula as a framework. Input the average flow velocity, cross-section ratio, and slope angle at the large cross-section end, and output the predicted value of the flow velocity difference at the main detection location. Extract the corresponding input data within the statistical time period and divide it into training, validation, and test sets in a 7:2:1 ratio; The mean squared error is used as the loss function, and the average absolute error between the predicted and actual values ​​of the velocity difference at the main detection sites is used as the accuracy index for iterative training. Training is stopped and the model is saved when the mean absolute error is less than or equal to 0.1 mm / s. Input the current average flow velocity, cross-sectional ratio, and slope angle at the large cross-section end into the model, and output the current theoretical metal flow velocity difference; If the current theoretical metal flow rate difference is greater than or equal to the metal flow rate difference threshold, a level 2 warning will be triggered.

9. The online monitoring method for the forging process of a variable cross-section transmission component for automobiles according to claim 1, characterized in that: The process of performing temperature-flow-velocity difference correlation analysis and determining the temperature-flow-velocity difference model is as follows: Historical metal flow rate difference and forging temperature data of the main detection sites were extracted and aligned according to the time sequence of the acquisition interval to obtain the historical metal flow rate difference sequence and the historical forging temperature sequence, respectively. The differences between adjacent data in the historical metal flow rate difference sequence and the historical forging temperature sequence are calculated separately to generate the metal flow rate difference change sequence and the forging temperature difference sequence. The Pearson correlation coefficient between the metal flow rate difference sequence and the forging temperature difference sequence was calculated and its absolute value was taken to obtain the correlation coefficient. If the correlation coefficient is greater than or equal to the correlation coefficient threshold, a univariate linear model is fitted by the least squares method to obtain the temperature-flow-rate difference model. If the correlation coefficient is less than the correlation coefficient threshold, nonlinear models are constructed respectively: binary linear model, higher-order polynomial model, exponential model, power function model, and trigonometric function model. After fitting by the least squares method, the model with the largest determination coefficient is selected as the temperature-flow-rate difference model.

10. The online monitoring method for the forging process of a variable cross-section transmission component for automobiles according to claim 1, characterized in that: A feasibility analysis was conducted on forging temperature regulation. Based on the results of the feasibility analysis, the process of implementing forging intervention operations is as follows: Calculate the difference between the current forging temperature and the maximum / minimum value of the forging temperature range, and substitute it into the temperature-flow-rate difference model to obtain the theoretical range of metal flow-rate difference variation. The forging temperature range refers to the allowable temperature range that meets the requirements of the forging process of automotive variable cross-section transmission parts. The adjustment range of the current metal flow rate difference is obtained by adding the current theoretical metal flow rate difference with the maximum and minimum values ​​of the theoretical metal flow rate difference variation range. If the adjustment range includes the metal flow rate difference threshold, it indicates that forging temperature adjustment is highly feasible, and forging intervention should be performed. Calculate the difference between the current theoretical metal flow rate difference and the metal flow rate difference threshold, substitute it into the temperature flow rate difference model to obtain the current forging temperature adjustment amount, and adjust the current forging temperature according to the current forging temperature adjustment amount. If the adjustment range does not include the metal flow rate difference threshold, it indicates that the feasibility of forging temperature adjustment is low, and manual intervention is required.

Images