High-precision thread machining process optimization method and system applied to hot runner system

Abstract

The invention relates to the field of machining process optimization based on machine learning, and particularly discloses a high-precision thread machining process optimization method and system applied to a hot runner system.The method comprises the steps that machining parameters (the nominal thread diameter, the thread pitch and the like) and working condition parameters (the material linear expansion coefficient, the working temperature range and the like) are obtained; the parameters are input into a pre-trained neural network model, an ideal machining compensation angle or quantity is output, and the model is formed by training historical, simulation or experimental data; calculating the radial offset of the cutter path of the numerical control machine tool according to the compensation angle or quantity; and a machine tool is controlled to machine according to the offset, so that the cold state of the thread contains pre-compensation eccentricity, and the hot state tends to be coaxial or contact stress is uniform. The high-temperature service reliability of the threads is improved, the tooth engagement risk is reduced, and the method is adaptive to an existing machine tool and high in practicability.

Description

Technical Field

[0001] This invention relates to the field of machining process optimization based on machine learning, specifically to a high-precision thread machining process optimization method and system applied to hot runner systems. Background Technology

[0002] In high-temperature operating scenarios such as hot runner systems, threaded connectors (such as the connection between the manifold and the hot nozzle) are core force transmission and sealing components, and their connection reliability directly affects the system's operational stability. However, existing thread processing technology has significant technical defects, making it difficult to adapt to the requirements of high-temperature operation and prone to "thread seizing" failures (high-pressure seizing after thermal expansion, and inability to disassemble after cooling). Specific problems are as follows: Thermal eccentricity and uneven stress: Threaded fasteners are mostly made of different materials. The difference in thermal expansion coefficients causes the components to expand asynchronously at high temperatures, resulting in slight eccentricity of the thread axis. At the same time, it is difficult for the mounting holes to be completely coaxial during assembly. Even a deviation of tens of micrometers can accumulate to form uneven stress, causing one side of the thread to be too tight and the other side to be too loose.

[0003] Traditional machining datum adaptation failure: Traditional thread machining (turning, milling) takes the workpiece geometric center as the datum, and the tool path is strictly coaxial with the axis. Although it can ensure the geometric accuracy in the cold state, it ignores the eccentric deformation in the hot state. During service, the thread accumulates micro-strain due to thermal cycling, and the eccentricity gradually amplifies, resulting in the machining accuracy being out of sync with the actual working conditions.

[0004] Insufficient reliability in long-term use: Repeated heating-cooling thermal cycles will continuously accumulate micro-strain, causing the thread eccentricity to increase with the number of uses. Traditional processes do not have targeted compensation designs, and the risk of thread breakage increases significantly after long-term service. This not only increases maintenance costs, but may also cause system downtime due to connector failure, affecting production efficiency.

[0005] In summary, existing processes cannot solve the problem of thermal eccentricity of threads under high temperature and multi-cycle conditions. There is an urgent need for a high-precision thread machining optimization solution that can adapt to thermal requirements and improve long-term reliability. Summary of the Invention

[0006] The purpose of this invention is to provide a high-precision thread machining process optimization method and system for hot runner systems, so as to solve the problems mentioned in the background art.

[0007] This invention provides a method for optimizing high-precision thread machining processes in hot runner systems, comprising the following steps: S1, Obtain the machining parameters and working condition parameters of the target threaded connector; the machining parameters include the nominal thread diameter, pitch, thread angle, and effective connection length; the working condition parameters include the first material linear expansion coefficient, the second material linear expansion coefficient, the working temperature range, and the preload. S2, the processing parameters and working condition parameters are input into a pre-trained neural network model, and the ideal processing compensation angle or ideal processing compensation amount is output; wherein, the neural network model is trained using historical data, simulation data or experimental data with the processing parameters and working condition parameters as input features and the ideal processing compensation angle or ideal processing compensation amount used to compensate for thermal eccentricity as output labels; S3, Based on the ideal machining compensation angle or ideal machining compensation amount, calculate the radial offset of the tool path of the CNC machine tool; S4, control the CNC machine tool to perform thread machining according to the radial offset of the tool path, so that the tool path is radially offset relative to the theoretical center axis of the workpiece, thereby machining a thread that has pre-compensated eccentricity in the cold state and tends to be coaxial or has uniform contact stress distribution at the working temperature.

[0008] Preferably, the temperature rise is calculated based on the operating temperature range, and the difference in thermal expansion coefficients of the materials is calculated based on the linear expansion coefficient of the materials; if the temperature rise is higher than a first threshold, and / or the difference in thermal expansion coefficients of the materials is higher than a second threshold, then a weighting coefficient greater than 1 is applied to the preload parameter input to the neural network model to amplify the influence weight of the preload on the compensation amount in the model.

[0009] Preferably, the method further includes acquiring dynamic operating parameters, the dynamic operating parameters including the expected number of thermal cycles; and introducing a safety compensation increment that is positively correlated with the expected number of thermal cycles in the dynamic operating parameters based on the radial offset of the toolpath.

[0010] Preferably, the neural network model is a feedforward neural network, comprising an input layer, at least one hidden layer, and an output layer, wherein the hidden layer uses the ReLU activation function and the output layer uses the linear activation function.

[0011] Preferably, when training the neural network model, mean squared error is used as the loss function, and at least one of L2 regularization, early stopping strategy and k-fold cross-validation is used to improve the generalization ability of the model.

[0012] Preferably, the simulation data is obtained through thermo-mechanical coupled finite element analysis. Specifically, under multiple sets of input parameters, the ideal machining compensation angle that makes the contact pressure distribution of the threaded connector most uniform or the contact pressure difference the smallest at the working temperature is found through reverse search, and this value is used as the output label of the corresponding input parameter.

[0013] Preferably, the ideal machining compensation angle is expressed in radians, and the product of the radian compensation angle and the nominal radius of the thread is used as the radial offset.

[0014] Preferably, when the compensation has directional requirements, the calculated radial offset is multiplied by the cosine and sine values ​​of a pre-specified circumferential angle based on the direction of thermal expansion or assembly deviation, thereby decomposing the offset components in the mutually perpendicular X-axis and Y-axis directions.

[0015] Preferably, before training the neural network model, the input features used are numericalized and normalized.

[0016] In another aspect, this application also provides a high-precision thread machining process optimization system for hot runner systems, comprising: The parameter acquisition module is used to acquire the machining parameters and working condition parameters of the target threaded connector; the machining parameters include the nominal diameter of the thread, the pitch, the thread angle, and the effective connection length; the working condition parameters include the first material linear expansion coefficient, the second material linear expansion coefficient, the working temperature range, and the preload. The model output module is used to input the processing parameters and working condition parameters into a pre-trained neural network model and output the ideal processing compensation angle or ideal processing compensation amount; wherein, the neural network model is trained using historical data, simulation data or experimental data with the processing parameters and working condition parameters as input features and the ideal processing compensation angle or ideal processing compensation amount used to compensate for thermal eccentricity as output labels; The offset calculation module is used to calculate the radial offset of the tool path of the CNC machine tool based on the ideal machining compensation angle or ideal machining compensation amount. The control module is used to control the CNC machine tool to perform thread machining according to the radial offset of the tool path, so that the tool path is radially offset relative to the theoretical center axis of the workpiece, thereby machining a thread that has pre-compensated eccentricity in the cold state and tends to be coaxial or has a uniform contact stress distribution at the working temperature.

[0017] This invention collects multi-dimensional parameters and predicts the ideal compensation amount through a feedforward neural network. Combining small-angle approximation and directional decomposition, the compensation amount is converted into a radial offset that can be executed by the machine tool. It can also amplify the preload weight for high-risk working conditions and introduce a safety increment based on the number of thermal cycles to achieve cold pre-compensation. The processed threads tend to be coaxial and have uniform contact stress at the working temperature, which significantly reduces the risk of "tooth biting", improves long-term service reliability, reduces maintenance costs and system downtime losses, and at the same time requires no additional hardware modification, is compatible with existing CNC machine tools, and balances accuracy and practicality. Attached Figure Description

[0018] Figure 1 A flowchart of a high-precision thread machining process optimization method for hot runner systems provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the neural network model structure provided in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the workflow of the neural network model provided in this embodiment of the invention; Figure 4 This is a schematic diagram of the weighting coefficient application process provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of a high-precision thread machining process optimization system applied to a hot runner system, provided as an embodiment of the present invention. Detailed Implementation

[0019] 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.

[0020] like Figure 1 As shown, this embodiment of the invention discloses a high-precision thread machining process optimization method 100 applied to a hot runner system, comprising the following method steps: S1, Obtain the machining parameters and working condition parameters of the target threaded connector; the machining parameters include the nominal thread diameter, pitch, thread angle, and effective connection length; the working condition parameters include the first material linear expansion coefficient, the second material linear expansion coefficient, the working temperature range, and the preload. S2, the processing parameters and working condition parameters are input into a pre-trained neural network model, and the ideal processing compensation angle or ideal processing compensation amount is output; wherein, the neural network model is trained using historical data, simulation data or experimental data with the processing parameters and working condition parameters as input features and the ideal processing compensation angle or ideal processing compensation amount used to compensate for thermal eccentricity as output labels; S3, Based on the ideal machining compensation angle or ideal machining compensation amount, calculate the radial offset of the tool path of the CNC machine tool; S4, control the CNC machine tool to perform thread machining according to the radial offset of the tool path, so that the tool path is radially offset relative to the theoretical center axis of the workpiece, thereby machining a thread that has pre-compensated eccentricity in the cold state and tends to be coaxial or has uniform contact stress distribution at the working temperature.

[0021] In some embodiments, for S1, the machining parameters focus on the geometric features of the thread itself and determine the basic structure of the thread, which may include the nominal thread diameter (D), pitch (P), thread angle (α), and effective connection length (L).

[0022] Specifically, for the nominal diameter (D) of the thread, the basic size of the major diameter of the thread (e.g., D=12mm for M12) determines the thread's load-bearing capacity and radius R (R=D / 2, which is needed for subsequent calculation of radial offset); the nominal size can be extracted from the design drawings (e.g., CAD model annotation), or if it is a replica of an existing workpiece, the average value of the major diameter can be measured using a micrometer or laser diameter gauge; For the pitch (P), it is the distance between corresponding points of two adjacent threads in the axial direction, which affects the number of threads engaged and the uniformity of force distribution. The standard pitch can be directly read from the design drawings (e.g., M12×1.75 corresponds to P=1.75mm); or it can be measured one by one with a pitch gauge or tool microscope during actual measurement. For the thread profile angle (α), the included angle between the two sides of the thread profile (e.g., α=60° for ordinary metric threads) determines the contact area and force direction when the threads are engaged. For example, it can be determined according to the standard selection during the design stage (e.g., α=55° for pipe threads), or the thread profile section can be observed using a universal angle ruler or metallographic microscope during actual measurement. For the effective length of the connection (L), the actual engagement length of the thread (not the total length of the workpiece, the invalid engagement section needs to be deducted) affects the overall load-bearing stiffness and thermal strain accumulation of the thread; it can be obtained by marking the effective engagement length on the design drawings (e.g., L=15mm); or, during actual measurement, the actual engagement length of the thread after assembly can be measured with a depth gauge, or the length of the mating section can be read from the CAD assembly model.

[0023] In one embodiment, the operating condition parameters focus on the thermal environment and assembly condition of the thread during service, which are the core causes of thread seizing problems in traditional processes. These parameters may include the linear expansion coefficients of the first / second materials (α1 / α2) and the operating temperature range (T). min ~T max ), preload (F) pre ); The linear expansion coefficients (α1 / α2) of the first and second materials respectively correspond to the core materials of the threaded connectors (e.g., material α1 for hot nozzles and material α2 for manifolds), describing the length expansion rate of the material per 1°C increase (unit: ×10). -6 / K). The difference between the two (Δα=|α1-α2|) is the main reason for the asynchronous material expansion, determining the initial trend of thread eccentricity under hot conditions. For example, it can be extracted from the material report provided by the material supplier (e.g., α≈11.5×10 for H13 steel). -6 / K, P20 steel α≈12.1×10 -6 / K); For the operating temperature range, the threaded fastener is in the actual service temperature range (e.g., room temperature 25℃ to operating temperature 350℃), where the temperature rise ΔT = T max -T min Ambient temperature determines the amount of material expansion (expansion = original length × α × ΔT), which can be determined according to the process requirements of the hot runner system (e.g., the hot nozzle temperature needs to be 300~320℃ when injection molding PP material); or the temperature of the threaded connection area can be measured at multiple points during normal operation (e.g., the junction of the manifold and the hot nozzle). For preload (F) pre Preload is the tightening force applied during thread assembly (unit: N), or the preload torque converted by a torque wrench (unit: N·m). Preload determines the tightness of the thread engagement; too much or too little preload increases the risk of thread jamming under heat. Excessive preload leads to excessive contact pressure during thermal expansion, while insufficient preload results in increased clearance after cooling. The preload range can be determined through mechanical calculations based on thread load requirements (such as the sealing pressure of a hot runner system) (e.g., 5~8 kN preload for M12 threads); alternatively, the preload value can be collected in real-time during assembly, or the actual preload can be calculated using a torque-preload conversion formula (combined with the thread friction coefficient).

[0024] In some embodiments, the neural network model is specifically a feedforward neural network, the specific structure of which can be found in [reference needed]. Figure 2 The diagram shows the structure of a neural network model, including an input layer, a hidden layer, and an output layer.

[0025] Specifically, for the input layer, for example, the number of nodes is the same as the number of input features (e.g., 4 processing parameters and 4 working condition parameters collected in step S1, for a total of 8 core features); the standardized parameter features are received and transmitted to provide a basis for the nonlinear calculation of the subsequent hidden layer; Preferably, the input data is first processed by numericalization and normalization (e.g., converting a tooth angle of 60° and a preload of 5000N into standardized values ​​in the range of 0 to 1; the specific processing steps will not be described in this embodiment); normalization can avoid the impact of differences in parameter magnitudes (e.g., a diameter of 12mm vs. an expansion coefficient of 11.5×10). -6 / K) causes the model to be overly sensitive to large numerical parameters. For the hidden layer, for example, there is at least one hidden layer, preferably two layers; the number of neurons is 16-32 per layer (which can be increased to 64 when the sample size exceeds 1000 groups); the activation function is ReLU (rectified linear unit), with the formula f(x)=max(0,x); the ReLU activation function realizes nonlinear mapping and solves the complex coupling relationship between processing parameters and compensation amount (such as the interaction between temperature rise and the difference in material expansion coefficient); the multi-layer design can improve the model's ability to fit complex laws, and the scale of 16-32 neurons can balance the fitting accuracy and the risk of overfitting; For the output layer, there is one node (outputting a single value, i.e., the ideal machining compensation angle Δθ, in degrees or radians; or the ideal machining compensation amount Δx, in mm); the activation function is a linear activation function, with the formula f(x)=x; the output is a true compensation value without nonlinear transformation. Since the compensation angle / amount is a continuous value (e.g., 0.02°, 3μm), linear activation ensures that the output value directly corresponds to the physical meaning of the actual requirement, without additional scaling deviation; it avoids the truncation or scaling of the compensation amount by nonlinear activation functions (e.g., Sigmoid, Tanh), ensuring that the output Δθ / Δx can be directly used for subsequent geometric conversions (e.g., converted into machine tool radial offset).

[0026] In one embodiment, the neural network model is trained using historical data, simulation data, or experimental data with the processing parameters and working condition parameters as input features and the ideal processing compensation angle or ideal processing compensation amount for compensating for thermally induced eccentricity as output labels.

[0027] Specifically, the input features are the numerical and normalized data of the eight core parameters corresponding to S1, as described in the aforementioned embodiment, and will not be repeated here.

[0028] The output label is the target value that the model needs to learn. In this embodiment, the label is labeled by historical data backtracking, simulation calculation or experimental trial and error to ensure the physical validity of the label. Specifically, for annotation based on historical data, for example, extracting the parameters of threaded connections that did not exhibit seizing phenomena from the maintenance records of hot runner systems, and retrospectively tracing back the possible compensation angles during their machining, for example, a certain M12 thread (D=12mm) at T_max=350℃ and α1=11.5×10 -6 / K、α2=12.1×10 -6 Under the / K condition without biting, the optimal Δθ (e.g., 0.03°) is calculated by back-calculating the disassembly torque and hot contact pressure records as a label.

[0029] Preferably, the annotation is based on simulation calculations. The simulation data is obtained through thermo-mechanical coupled finite element analysis. Specifically, under multiple sets of input parameters, the ideal machining compensation angle that makes the contact pressure distribution of the threaded connector most uniform or the contact pressure difference the smallest at the working temperature is found through reverse search, and this value is used as the output label of the corresponding input parameter.

[0030] Specifically, for labels based on simulation data, for example, labels are generated through thermo-mechanical coupled finite element analysis (FEA). Given input features (e.g., D=16mm, ΔT=325℃), different compensation angles Δθ are adjusted, the hot thread contact pressure distribution corresponding to each Δθ is calculated, and the Δθ that minimizes the contact pressure difference (e.g., 0.045°) is searched in reverse, and this value is used as the label for that set of input features.

[0031] For labeling based on experimental data, for example, labeling can be done through small-batch physical tests (10-30 samples), processing multiple sets of threaded samples with the same parameters but different Δθ (e.g., Δθ=0.02°, 0.03°, 0.04°), assembling them and heating them to the working temperature, measuring the disassembly torque and contact uniformity of each set of samples, and selecting the Δθ (e.g., 0.03°) with the smoothest disassembly and the most uniform contact pressure as the label.

[0032] In one embodiment, preferably, when training the neural network model, mean squared error is used as the loss function, and at least one of L2 regularization, early stopping strategy and k-fold cross-validation is used to improve the generalization ability of the model.

[0033] Specifically, the formula for mean squared error (MSE) is: ,in For accurate compensation of angle / quantity labels, Here, n represents the model's predicted value, and n is the number of samples. MSE is more sensitive to small errors and can accurately penalize the model's prediction deviation for small compensation amounts (such as 3μm or 0.02°). Thermally induced eccentricity compensation requires accuracy at the μm and angular levels. MSE can ensure that the model focuses on the accurate prediction of small compensation amounts, avoiding processing failure due to error accumulation.

[0034] The optimizer uses mini-batch SGD or Adam, with an initial learning rate set to 1e. -3 SGD is suitable for scenarios with small sample sizes and can gradually approach the minimum of the loss function; Adam, combining a momentum term and an adaptive learning rate, is suitable for scenarios with complex parameter coupling. During training, the learning rate can be adjusted according to the decreasing trend of the loss (e.g., reducing it to 1e when the loss stagnates). -4 This ensures efficient model convergence.

[0035] In one embodiment, L2 regularization is used, which involves adding the sum of squares of the model weights to the loss function, as shown in the formula: , where λ is the regularization coefficient (for example, λ≈1e -4 ), where w represents the connection weights of each layer in the model. This can suppress excessively large model weights and prevent the model from overfitting noise in the training data.

[0036] In one embodiment, an early stopping strategy is adopted to avoid overfitting caused by continuous iteration of the model on the training set. K-fold cross-validation is used, which involves randomly dividing the entire dataset into k non-overlapping subsets (e.g., k=5), using one subset as the validation set and the remaining k-1 subsets as the training set, repeating the training and validation process k times, and finally taking the average MSE of the k validations as the evaluation index of the model's generalization ability.

[0037] In one embodiment, when the trained model is put into use, such as Figure 3 The diagram shown illustrates the workflow of a neural network model, which may include: S201, Parameter Input: Normalize the processing parameters and working condition parameters collected in step S1 according to the standard during training (e.g., map the working temperature 25℃~350℃ to 0~1) and input them into the model input layer. S202, forward computation, the input features are passed from the input layer to the hidden layer, undergo nonlinear transformation through the ReLU activation function, and then passed to the output layer to output the normalized compensation amount through the linear activation function; S203, denormalization, restores the normalized output value to the ideal processing compensation angle / amount in a physical sense. For example, if Δθ=0°~0.1° is normalized to 0~1 during training, and the model output is 0.3, then the corresponding denormalization is Δθ=0.03°. S204, Output Selection: Output the corresponding form according to subsequent machining requirements. If the radial offset of the computer tool is required, Δθ (angle) will be output first. If it is directly used for milling CAM path correction, Δx (radial compensation amount, mm) can be output. The two can be interchanged through geometric relationships, which will not be elaborated further.

[0038] Preferably, the temperature rise is calculated based on the operating temperature range, and the difference in thermal expansion coefficients of the materials is calculated based on the linear expansion coefficient of the materials; if the temperature rise is higher than a first threshold, and / or the difference in thermal expansion coefficients of the materials is higher than a second threshold, then a weighting coefficient greater than 1 is applied to the preload parameter input to the neural network model to amplify the influence weight of the preload on the compensation amount in the model.

[0039] Specifically, by quantitatively judging the risk level of the working condition, a weighting coefficient greater than 1 is applied to the preload parameter to amplify its weight in the model prediction, ensuring that the compensation amount under extreme working conditions can adapt to the coupling effect of preload and thermal expansion, and further reduce the risk of "gripping".

[0040] Specifically, for the calculation of temperature rise (ΔT), the temperature difference between ambient temperature and maximum operating temperature during the service life of threaded fasteners is a direct indicator of the magnitude of thermal expansion. A larger temperature rise indicates greater material expansion and more significant thermally induced eccentricity. The operating temperature range (T) collected in step S1 is used as the basis for this calculation. min ~T max ), where T min The temperature is usually ambient (default 25℃; if there are special low-temperature conditions, the actual value T will be used). min ), T max This represents the highest temperature under operating conditions. ΔT = T max -T min ; For calculating the difference in thermal expansion coefficients (Δα), for example, the difference in linear expansion coefficients between the two core materials of a threaded connector (such as the hot nozzle material α1 and the manifold material α2) is a key indicator for measuring the degree of asynchronous material expansion. The larger the difference, the more significant the difference in expansion between the two materials under hot conditions, and the more severe the thread eccentricity. The linear expansion coefficients (α1) and (α2) of the first and second materials, collected in step S1, are both in units of ×10⁻⁶. -6 / K. The calculation formula is, Δα=|α1-α2| (take the absolute value, only focus on the magnitude of the difference, not on whether it is positive or negative); In one embodiment, the first threshold (temperature rise threshold) and the second threshold (expansion coefficient difference threshold) are set based on the actual operating conditions and historical fault data of the hot runner system to ensure accurate identification of high-risk operating conditions. For example, the specific setting logic and reference values ​​may include... For the temperature rise threshold, based on the historical statistics of "biting" failures, the minimum temperature rise of threaded fasteners that have previously exhibited "biting" phenomenon is calculated, and 90% of this value is taken as the first threshold (with a safety margin); for the expansion coefficient difference threshold, based on the safety difference range of material combinations, the maximum expansion coefficient difference among commonly used material combinations that has not exhibited "biting" is calculated, and 80% of this value is taken as the second threshold. It is understandable that if a new "biting" fault occurs in the future, the ΔT and Δα of the faulty part need to be recalculated and the threshold adjusted. In addition, the thresholds for different hot runner application scenarios (such as injection molding high-temperature nylon vs. low-temperature PE materials) need to be set separately to avoid misjudgment caused by a general threshold. This embodiment will not elaborate further.

[0041] In one embodiment, when a high-risk condition is identified (ΔT>ΔT0 or Δα>Δα0), a weighting coefficient (k>1) is applied to the preload parameter input to the neural network. The core function of the weighting coefficient is to amplify the influence of the preload in the model prediction. That is, under high-risk conditions, the preload is more sensitive to the "gripping" effect (e.g., high temperature rise and high preload can easily lead to excessive contact pressure), making the model pay more attention to the preload parameter. For example, the process of determining the weighting coefficient (k) may include... For a single high risk, i.e. only ΔT>ΔT0 or only Δα>Δα0, for example, the weighting coefficient is 1.2~1.5. For a medium risk level, the preload weight is appropriately increased. The coefficient is positively correlated with the proportion exceeding the threshold: Exceeding proportion = (actual value - threshold) / threshold. The larger the exceeding proportion, the larger the coefficient. For dual high-risk conditions (ΔT>ΔT0 and Δα>Δα0), for example, the weighting coefficient is 1.6~2.0. The risk level is high, and the preload weight is significantly amplified. The coefficient is the superposition of the single high-risk coefficient (but does not exceed 2.0 to avoid excessive amplification leading to parameter imbalance).

[0042] In one embodiment, such as Figure 4 The diagram shown illustrates the weighting coefficient application process, which may include: S401, parameter extraction, for example, extracting the original preload value (F) from the working condition parameters collected in step S1. pre (Unit: N), e.g., F pre =6000N; S402, Working condition judgment: Calculate ΔT and Δα, compare with the threshold to determine whether it is a high-risk working condition. For example, if ΔT=325℃>ΔT0=280℃ (single high risk), determine k=1.3. S403, weighted calculation: multiply the original preload value by a weighting coefficient to obtain the weighted preload value (F). pre '), F pre '=F pre ×k; S404, normalization process, the weighted F preNormalize the data along with other input parameters (such as diameter and pitch) (map to the 0~1 range) before inputting it into the neural network model. Note that the normalization needs to be recalculated based on the weighted data range. For example, if the original preload range is 3000~8000N, the weighted range may become 3600~16000N. Normalization should be performed according to the new range to avoid imbalance of parameter magnitudes.

[0043] Therefore, under high-risk operating conditions, the weighted preload parameter has a higher proportion in the weight update of the model training. For example, when the model learns the parameter combination of high temperature rise and high preload, it will tend to adjust the weights related to preload to make the output compensation amount more suitable for the contact pressure change caused by preload. If no weighting coefficient is applied, the model may pay too much attention to the temperature rise and the difference in expansion coefficient and ignore the influence of preload, resulting in a small compensation amount. For example, under high temperature rise, the thread itself expands in volume. If the preload is too high, the model without amplified weights may output a small compensation angle, which cannot offset the excessive contact pressure and still result in "gritting teeth".

[0044] The weighting coefficient only applies to the preload parameter and does not impose additional weights on other parameters (such as diameter and expansion coefficient) to avoid breaking the original mapping relationship between input features. For example, if the weights of diameter and preload are increased at the same time, the model may misjudge the influence of diameter on the compensation amount and output an incorrect compensation angle. The weighting coefficient shall not exceed 2.0. If it exceeds this value, the preload parameter may have too high a weight in the model, which may mask the influence of other key parameters (such as temperature rise). For example, when k=3.0, the model may only predict the compensation amount based on the preload and ignore the large expansion caused by the temperature rise, which may increase the risk of "biting teeth".

[0045] In some embodiments, for S3, the ideal machining compensation angle / amount output by the neural network is converted into a radial offset of the tool path (μm-level accuracy) that can be recognized and executed by the CNC machine tool, and decomposed into X / Y axis components when there are directional requirements. When the thermal eccentricity of the thread has no obvious directionality (such as uniform assembly deviation and symmetrical thermal expansion direction), only a single radial offset needs to be calculated. Based on the small angle approximation principle, the angle compensation is converted into a linear offset. For example, the specific implementation is as follows.

[0046] First, the ideal machining compensation angle (Δθ) is output by the neural network and is in degrees, typically ranging from 0.01° to 0.1° (small angle, satisfying subsequent approximation conditions); the nominal thread diameter (D) comes from the machining parameters collected in step S1 and is in millimeters (mm), used to calculate the nominal thread radius (R). The nominal radius is the core geometric reference for converting angle compensation into radial offset, and the formula is R=D / 2. Next, preferably, the ideal machining compensation angle is expressed in radians, and the radian compensation angle is multiplied by the nominal radius of the thread. The resulting product is the radial offset.

[0047] Specifically, the compensation angle is converted from "degrees" to "radians". The radian system directly relates angles to arc length (1 radian corresponds to an arc length of radius), which can be simplified to the following formula: ,in, Indicates the compensated angle in radians. The angle output by the neural network is represented by the compensation angle in radians. Next, the radial offset (e) is approximated based on a small angle. Since the compensation angle Δθ is typically <0.1° (small angle), it meets the approximation condition (error <0.005%, fully satisfying the μm-level accuracy requirements for thread machining). Therefore, the radial offset can be directly calculated using the nominal radius × radian compensation angle. Where e is the radial offset of the tool path, in millimeters (mm), which is converted to micrometers (μm) in practical applications to facilitate precise execution by CNC machine tools (the minimum adjustment unit of the machine tool is usually 0.1μm). At small angles, the arc length corresponding to the angular offset in the thread circumferential direction is approximately equal to the radial linear offset, that is, after the tool is radially offset by e, the machined thread can just return to the coaxial state under hot conditions.

[0048] In one embodiment, when the thermally induced eccentricity of the thread has a clear direction (such as a fixed coaxial deviation when assembling the manifold and the hot nozzle, or when the thermal expansion of the material is more significant along a specific direction), the single radial offset e needs to be decomposed into offset components of the X / Y axes of the CNC machine tool (e... x e y To ensure the tool offsets along the compensation direction, the specific steps are as follows: First, determine the circumferential angle (φ) of the compensation direction. The circumferential angle φ is the core benchmark for direction decomposition, defined as the angle between the compensation direction and the positive X-axis (clockwise is positive, range 0°~360°), which needs to be determined in advance based on engineering experience or simulation data; for example, Among them, based on the direction of thermal expansion, for example, if two materials (such as a hot nozzle α1=11.5×10 -6 / K, Diverter plate α2 = 12.1 × 10 -6 The difference in the coefficient of thermal expansion of / K causes the thread to shift towards the manifold side when hot (assuming this direction corresponds to a 45° deviation in the positive X-axis direction), then φ=45°; Based on the direction of assembly deviation, for example, if the manifold mounting hole is offset relative to the hot nozzle hole in the positive Y-axis direction during actual assembly (i.e., compensation needs to be performed in the positive Y-axis direction), then φ = 90°. Among them, φ can be determined by thermal-mechanical coupling FEA simulation (outputting the offset direction of the thread axis under hot state) or historical maintenance records (statistically calculating the eccentric direction of the "biting" parts). It is necessary to ensure that the error of φ is <5° (to avoid the compensation failure caused by directional deviation).

[0049] Next, the X / Y axis offset components are calculated. Based on trigonometric relationships, the radial offset e is decomposed into the X-axis component e along the circumferential angle φ. x (Left and right direction) and Y-axis component e y (Front-back direction), the formula is as follows: X-axis component: e x =e×cosφ, where φ is the inscribed angle and needs to be converted to radians before calculation; Y-axis component: e y =e×sinφ;e x A positive value indicates that the tool shifts in the positive X-axis direction, while a negative value indicates that it shifts in the negative X-axis direction; e y Similarly, it conforms to the CNC machine tool coordinate system specification (right-hand coordinate system).

[0050] Therefore, if thermal eccentricity is only along a certain direction (such as the negative X-axis direction), compensation can be achieved solely through the X-axis component, avoiding the misalignment of compensation offset and actual eccentricity direction caused by the lack of directional compensation. When there are multi-directional assembly deviations between the manifold and the hot nozzle (such as 2μm in the positive X-axis direction and 3μm in the negative Y-axis direction), directional correction can be achieved by adjusting φ to make the compensation direction opposite to the deviation direction.

[0051] In some embodiments, for S4, the CNC machine tool is controlled to perform thread machining according to the radial offset of the tool path, so that the tool path is radially offset relative to the theoretical center axis of the workpiece, thereby machining a thread that has pre-compensated eccentricity in the cold state and tends to be coaxial or has a uniform contact stress distribution at the working temperature.

[0052] Preferably, the method further includes acquiring dynamic operating parameters, the dynamic operating parameters including the expected number of thermal cycles; and introducing a safety compensation increment that is positively correlated with the expected number of thermal cycles in the dynamic operating parameters based on the radial offset of the toolpath.

[0053] Introducing a safety compensation increment positively correlated with the expected number of thermal cycles, an additional compensation margin is reserved on top of the basic radial offset (to offset single thermal eccentricity) to cope with the eccentricity amplification caused by the superposition of thermal cycles during long-term service, further reducing the risk of "later thread breakage" and making the thread machining accuracy adapt to the use requirements of the entire life cycle.

[0054] For example, dynamic operating parameters focus on the long-term use scenarios of threaded fasteners. The parameter is the expected number of thermal cycles, determined in conjunction with the actual service requirements and maintenance cycles of the hot runner system. Specifically, the threaded fastener undergoes the number of complete thermal cycles at room temperature, operating temperature, and room temperature within its design service life (1 thermal cycle = 1 process of heating to operating temperature and cooling to room temperature). For example, if an injection mold is planned to run 2 thermal cycles per day (heating in the morning and cooling in the evening), operating for 300 days a year, with a design life of 5 years, then the expected number of thermal cycles = 2 × 300 × 5 = 3000 cycles. Each thermal cycle will generate micro-strain accumulation (the thermal expansion and contraction of the material causes slight deformation of the thread meshing surface). The more cycles, the more significant the superposition of micro-strain. The thermal eccentricity will gradually amplify with the increase of the number of cycles. Therefore, if the basic offset is calculated only based on the eccentricity of a single thermal cycle, the "tooth bite" will reappear after long-term use due to the amplified eccentricity.

[0055] In one embodiment, the expected number of thermal cycles can be obtained by extracting it from the design specifications or user requirements document of the hot runner system. If it is an existing device, the average number of thermal cycles per year can be calculated from the device's historical maintenance records. Combined with the user's planned device upgrade cycle (e.g., remaining 3 years of service life), the remaining expected number of thermal cycles can be calculated. If there is no specific requirement, it can be determined according to industry-standard practices. This invention does not impose any limitations.

[0056] In one embodiment, the calculation of the safety compensation increment is positively correlated with the number of thermal cycles. The safety compensation increment (Δe) is an additional offset added to the base radial offset (e, referring to the aforementioned embodiment). The key point is that the more thermal cycles expected, the larger Δe will be. For example, the specific calculation formula is as follows: ,in, Safety compensation increment coefficient (unit: μm / thousand cycles), N is the expected number of thermal cycles (times), N / 100 converts the number of cycles to thousands, and Δe is in μm.

[0057] Among them, the safety compensation increment coefficient It can be determined based on material properties and historical test data, reflecting the eccentricity increment caused by each thousand thermal cycles. Different materials have different thermal fatigue resistance, with lower thermal fatigue resistance (such as ordinary carbon steel). The larger the value, the higher the resistance (e.g., high-temperature alloy H13). The smaller the value; reference values ​​are shown in the table below. Table 1:

[0058] In one embodiment, the basic radial offset (e) is superimposed with the safety compensation increment (Δe) to obtain the final actual radial offset (e) used for the CNC machine tool. final In scenarios without directional requirements, efinal =e + Δe, directly superimposed, with the same direction as e; In scenarios with directional requirements, the direction of the safety compensation increment is the same as the direction of the basic offset (the direction of the eccentric amplification caused by micro-strain accumulation is consistent with the initial thermally induced eccentricity direction). Therefore, the X / Y axis components are superimposed separately. Specifically, e final_x =e x +Δe x , where Δe x =Δe×cosφ, where φ is the inscribed angle of the basic offset direction. Similarly, e final_y =e x +Δe y , where Δe y =Δe×sinφ; Final offset (e) final or e final_x e final_y The machine tool input method is consistent with the basic offset. Specifically, For turning operations, for example, e final Input the X-axis tool compensation value; For milling operations, for example, setting e in CAM software final_x e final_y This represents the offset of the tool center.

[0059] Figure 5 A high-precision thread machining process optimization system 500 for use in hot runner systems is illustrated. The system embodiment is similar to... Figure 1 Corresponding to the illustrated method embodiments, the specific methods include: The parameter acquisition module 501 is used to acquire the processing parameters and working condition parameters of the target threaded connector; the processing parameters include the nominal diameter of the thread, the pitch, the thread angle, and the effective connection length; the working condition parameters include the first material linear expansion coefficient, the second material linear expansion coefficient, the working temperature range, and the preload. The model output module 502 is used to input the processing parameters and working condition parameters into a pre-trained neural network model and output the ideal processing compensation angle or ideal processing compensation amount; wherein, the neural network model is trained using historical data, simulation data or experimental data with the processing parameters and working condition parameters as input features and the ideal processing compensation angle or ideal processing compensation amount used to compensate for thermal eccentricity as output labels; The offset calculation module 503 is used to calculate the radial offset of the tool path of the CNC machine tool based on the ideal machining compensation angle or the ideal machining compensation amount. The control module 504 is used to control the CNC machine tool to perform thread machining according to the radial offset of the tool path, so that the tool path is radially offset relative to the theoretical center axis of the workpiece, thereby machining a thread that has pre-compensated eccentricity in the cold state and tends to be coaxial or has a uniform contact stress distribution at the working temperature.

[0060] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for optimizing high-precision thread machining processes applied to hot runner systems, characterized in that, Includes the following steps: S1, Obtain the machining parameters and working condition parameters of the target threaded connector; the machining parameters include the nominal thread diameter, pitch, thread angle, and effective connection length; the working condition parameters include the first material linear expansion coefficient, the second material linear expansion coefficient, the working temperature range, and the preload. S2, the processing parameters and working condition parameters are input into a pre-trained neural network model, and the ideal processing compensation angle or ideal processing compensation amount is output; wherein, the neural network model is trained using historical data, simulation data or experimental data with the processing parameters and working condition parameters as input features and the ideal processing compensation angle or ideal processing compensation amount used to compensate for thermal eccentricity as output labels; S3, Based on the ideal machining compensation angle or ideal machining compensation amount, calculate the radial offset of the tool path of the CNC machine tool; S4, control the CNC machine tool to perform thread machining according to the radial offset of the tool path, so that the tool path is radially offset relative to the theoretical center axis of the workpiece, thereby machining a thread that has pre-compensated eccentricity in the cold state and tends to be coaxial or has uniform contact stress distribution at the working temperature.

2. The method for optimizing high-precision thread machining in a hot runner system according to claim 1, characterized in that, The temperature rise is calculated based on the operating temperature range, and the difference in thermal expansion coefficients of the materials is calculated based on the linear expansion coefficient of the materials. If the temperature rise exceeds a first threshold, and / or the difference in thermal expansion coefficients of the materials exceeds a second threshold, a weighting coefficient greater than 1 is applied to the preload parameter input to the neural network model to amplify the influence weight of the preload on the compensation amount in the model.

3. The method for optimizing high-precision thread machining in a hot runner system according to claim 1, characterized in that, It also includes acquiring dynamic operating parameters, which include the expected number of thermal cycles; and introducing a safety compensation increment that is positively correlated with the expected number of thermal cycles in the dynamic operating parameters, based on the radial offset of the toolpath.

4. The method for optimizing high-precision thread machining in a hot runner system according to claim 1, characterized in that, The neural network model is a feedforward neural network, which includes an input layer, at least one hidden layer, and an output layer, wherein the hidden layer uses the ReLU activation function and the output layer uses the linear activation function.

5. The method for optimizing high-precision thread machining in a hot runner system according to claim 4, characterized in that, When training the neural network model, mean squared error is used as the loss function, and at least one of L2 regularization, early stopping strategy and k-fold cross-validation is used to improve the generalization ability of the model.

6. The method for optimizing high-precision thread machining in a hot runner system according to claim 1, characterized in that, The simulation data was obtained through thermo-mechanical coupled finite element analysis. Specifically, under multiple sets of input parameters, the ideal machining compensation angle was found by reverse search to make the contact pressure distribution of the threaded connector most uniform or the contact pressure difference the smallest at the working temperature, and this value was used as the output label of the corresponding input parameter.

7. The method for optimizing high-precision thread machining in a hot runner system according to claim 1, characterized in that, The ideal machining compensation angle is expressed in radians. The product of the radian compensation angle and the nominal radius of the thread is the radial offset.

8. The method for optimizing high-precision thread machining in a hot runner system according to claim 7, characterized in that, It also includes, When the compensation has directional requirements, the calculated radial offset is multiplied by the cosine and sine of a pre-specified circumferential angle based on the direction of thermal expansion or assembly deviation, thereby decomposing the offset components in the mutually perpendicular X-axis and Y-axis directions.

9. The method for optimizing high-precision thread machining in a hot runner system according to claim 1, characterized in that, Before training the neural network model, the input features used are numericalized and normalized.

10. A high-precision thread machining process optimization system applied to hot runner systems, characterized in that, include: The parameter acquisition module is used to acquire the machining parameters and working condition parameters of the target threaded connector; the machining parameters include the nominal diameter of the thread, the pitch, the thread angle, and the effective connection length; the working condition parameters include the first material linear expansion coefficient, the second material linear expansion coefficient, the working temperature range, and the preload. The model output module is used to input the processing parameters and working condition parameters into a pre-trained neural network model and output the ideal processing compensation angle or ideal processing compensation amount; wherein, the neural network model is trained using historical data, simulation data or experimental data with the processing parameters and working condition parameters as input features and the ideal processing compensation angle or ideal processing compensation amount used to compensate for thermal eccentricity as output labels; The offset calculation module is used to calculate the radial offset of the tool path of the CNC machine tool based on the ideal machining compensation angle or ideal machining compensation amount. The control module is used to control the CNC machine tool to perform thread machining according to the radial offset of the tool path, so that the tool path is radially offset relative to the theoretical center axis of the workpiece, thereby machining a thread that has pre-compensated eccentricity in the cold state and tends to be coaxial or has a uniform contact stress distribution at the working temperature.

Images