Method and system for predicting element life of hydraulic multi-axis coordinated feeding device of forging machine

Abstract

The application discloses a kind of forging machine hydraulic multi-axis collaborative feed device element life prediction method, comprising: obtaining the real working data of element in forging machine hydraulic multi-axis collaborative feed device;Build element simulation model, verify simulation model and collect simulation data;Using simulation data to construct source domain dataset, extract core feature variables consistent with source domain data structure in real working data and construct target domain dataset after data preprocessing;Combined with source domain sample and target domain sample, construct mixed domain;Build life prediction model, life prediction model includes feature extractor, predictor, field adaptation module and joint optimization module, and train life prediction model, obtain trained life prediction model;The residual life of element in future time period is predicted using trained life prediction model, and the prediction result is output.The method significantly improves the prediction accuracy and generalization robustness during migration, further improves the accuracy of hydraulic component life prediction.

Description

Technical Field

[0001] This invention relates to the field of hydraulic technology for engineering machinery, specifically to a method, system, terminal, and medium for predicting the lifespan of components in a hydraulic multi-axis coordinated feeding device for forging machines. Background Technology

[0002] In large-scale engineering machinery such as forging machines, hydraulic multi-axis cooperative feeding devices are crucial actuators for achieving high-precision forming. These devices typically consist of core components such as hydraulic cylinders, servo valves, and hydraulic motors. Through coordinated operation between multiple axes, they enable synchronous clamping, displacement adjustment, and force control of the workpiece, thereby ensuring forming accuracy and production efficiency during the forging process. Compared to single-axis drives, multi-axis cooperative feeding not only achieves synchronous motion with multiple degrees of freedom under complex working conditions but also effectively distributes loads, improving system stability and reliability. However, in the high-pressure, high-load, and frequent-impact forging environment, hydraulic components are prone to wear, fatigue, or performance degradation over long-term operation, affecting the overall operational safety of the equipment. Failure of hydraulic components often leads to machine shutdown, production interruption, and even safety accidents. Therefore, accurate prediction of the remaining lifespan of components in the hydraulic multi-axis cooperative feeding device of a forging machine is of great significance for the reliable operation and maintenance of the equipment.

[0003] Existing life prediction methods can be broadly classified into three categories: (1) stress-based life prediction, which predicts life based on the dynamic characteristics of the failure mechanism, which is closer to actual engineering problems, but is difficult to apply under complex working conditions; (2) probability-statistic-based life prediction, which establishes a statistical model by accumulating experimental data, and predicts the expected life by determining the distribution of life characteristic values ​​over time and the failure probability, but requires a large amount of experimental data; (3) artificial intelligence-based life prediction, especially the deep learning methods that have been widely used in recent years, such as convolutional neural networks, recurrent neural networks and temporal convolutional networks, which can automatically extract complex degradation features from the original monitoring data and have strong nonlinear capabilities and prediction accuracy. However, these methods also have prominent problems: (1) they are highly dependent on large-scale, high-quality labeled samples, while the complete failure samples that can be obtained in actual engineering scenarios are extremely limited; (2) the models are generally trained under a single device or working condition, with poor generalization ability, and are difficult to adapt to the distribution differences between different devices, different working conditions and even different data sources.

[0004] To overcome the aforementioned limitations, transfer learning offers a solution for predicting the lifespan of hydraulic components. By pre-training the model using ample simulation or historical data from the source domain and then transferring it to a small-sample environment in the target domain, the dependence on large amounts of labeled data can be significantly reduced, improving the model's adaptability in different scenarios. However, existing transfer learning methods often neglect the weight coordination issue between the source domain supervision loss and the domain adaptation loss, especially in prediction scenarios with complex operating conditions and high nonlinearity, where fixed weights struggle to balance the optimization progress of both. When domain alignment is optimal, the source domain loss may not yet have converged; conversely, when the source domain loss is minimized, the domain adaptation effect may be weakened due to overfitting, leading to a bounce in feature distribution distance and affecting generalization performance. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method, system, terminal, and medium for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for forging machines. This enables the lifespan prediction model to flexibly balance source domain supervision and domain alignment losses during training, thereby significantly improving prediction accuracy and generalization robustness during domain migration and further enhancing the accuracy of hydraulic component lifespan prediction.

[0006] In a first aspect, the present invention provides a method for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for a forging machine, comprising:

[0007] Obtain the actual working data of the components in the hydraulic multi-axis coordinated feeding device of the forging machine;

[0008] Build component simulation models, verify simulation models, and collect simulation data;

[0009] The source domain dataset is constructed using simulation data. Core feature variables with the same data structure as the source domain are extracted from real working data and preprocessed to construct the target domain dataset.

[0010] A hybrid domain is constructed by combining source domain samples and target domain samples;

[0011] A lifespan prediction model is constructed, comprising a feature extractor, a predictor, a domain adaptation module, and a joint optimization module. The feature extractor is constructed using a fusion algorithm of TCN and Transformer. Trajectory data from the source and target domains are linearly interpolated to generate a mixed-domain input. The source, target, and mixed-domain data are input into the feature extractor, which extracts hidden-layer features and mixed-domain features. The hidden-layer features include both source and target-domain features. These features are then fused using a mix-up method to obtain enhanced mixed features. The source features are input into the predictor for supervised training to obtain the source-domain training loss. The mixed-domain features and the mixed-domain features are output by the predictor and synthesized to output the domain adaptation loss. In the domain adaptation module, the maximum mean difference is used to measure the distance between the feature distributions of the source, target, and mixed domains. In the joint optimization module, dynamic weights are used to adaptively adjust the balance between the source and domain adaptation losses, while simultaneously updating the mix-up interpolation weights in reverse and training the lifespan prediction model to obtain a trained lifespan prediction model.

[0012] The trained lifetime prediction model is used to predict the remaining lifetime of components over a future period, and the prediction results are output.

[0013] Secondly, embodiments of the present invention provide a component life prediction system for a forging machine hydraulic multi-axis cooperative feeding device, used to implement the component life prediction method for a forging machine hydraulic multi-axis cooperative feeding device described in the above embodiments. The system includes: a data acquisition module, a simulation model building module, a dataset construction module, a hybrid domain construction module, a prediction model construction module, and an output module.

[0014] The data acquisition module is used to acquire the actual working data of the components in the hydraulic multi-axis collaborative feeding device of the forging machine;

[0015] The simulation model building module is used to build component simulation models, verify simulation models, and collect simulation data.

[0016] The dataset construction module uses simulation data to construct the source domain dataset, extracts core feature variables that are consistent with the source domain data structure from real working data, and constructs the target domain dataset after data preprocessing.

[0017] The hybrid domain construction module is used to combine source domain samples and target domain samples to construct a hybrid domain;

[0018] The prediction model construction module is used to construct a lifespan prediction model. The lifespan prediction model includes a feature extractor, a predictor, a domain adaptation module, and a joint optimization module. The feature extractor is constructed using a fusion algorithm of TCN and Transformer. Trajectory data from the source domain and the target domain are linearly interpolated to generate a mixed domain input. The source domain data, target domain data, and mixed domain data are input into the feature extractor. The feature extractor extracts hidden layer features and mixed domain features. The hidden layer features include source domain features and target domain features. The source domain features and target domain features are fused using a mix-up method to obtain enhanced mixed features. The source domain features are input into the predictor for supervised training to obtain the source domain training loss. The mixed domain features and the mixed features are output by the predictor respectively and synthesized to output the domain adaptation loss. In the domain adaptation module, the maximum mean difference is used to measure the distance between the feature distributions of the source domain, target domain, and mixed domain. In the joint optimization module, dynamic weights are used to adaptively adjust the balance between the source domain loss and the domain adaptation loss. At the same time, the mix-up interpolation weights are updated in reverse, and the lifespan prediction model is trained to obtain a trained lifespan prediction model.

[0019] The output module uses a trained lifetime prediction model to predict the remaining lifetime of the component over a future time period and outputs the prediction results.

[0020] Thirdly, an embodiment of the present invention provides a smart terminal, including a processor, an input device, an output device, and a memory. The processor, input device, output device, and memory are interconnected. The memory is used to store a computer program, the computer program including program instructions, and the processor is configured to call the program instructions to execute the method described in the above embodiments.

[0021] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform the methods described in the above embodiments.

[0022] The beneficial effects of this invention are:

[0023] 1. This invention provides a method for obtaining a more effective domain adaptation loss by mixing the inputs and feature distributions of the source and target domains based on the Temporal Convolutional Network (TCN)-Transformer feature extractor. Simultaneously, by combining a dynamic weight adaptive adjustment strategy based on feature distribution distance, the model can flexibly balance the source domain supervision and domain alignment losses during training, thereby significantly improving the prediction accuracy and generalization robustness during domain transfer, and further enhancing the accuracy of hydraulic component life prediction.

[0024] 2. The component life prediction system, terminal and medium of the hydraulic multi-axis cooperative feeding device for forging machine provided in the embodiments of the present invention have the same inventive concept and the same beneficial effects as the component life prediction method of the hydraulic multi-axis cooperative feeding device for forging machine provided in the embodiments of the present invention, and will not be repeated here. Attached Figure Description

[0025] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0026] Figure 1 A flowchart of a method for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for a forging machine, provided in the first embodiment of the present invention;

[0027] Figure 2 This is a framework diagram of a lifetime prediction model;

[0028] Figure 3 A flowchart illustrating the specific method for obtaining the actual working data of components in a hydraulic multi-axis coordinated feed device for forging machines;

[0029] Figure 4 A flowchart outlining the specific methods for building a hydraulic motor simulation model, validating the model, and collecting simulation data.

[0030] Figure 5 A flowchart illustrating the specific methods for data feature extraction and prediction in order to build feature extractors and predictors;

[0031] Figure 6 A flowchart illustrating the specific method for constructing a hybrid domain by combining source domain samples and target domain samples;

[0032] Figure 7 The flowchart illustrates the specific method for adaptively adjusting the balance between source domain loss and domain adaptation loss using dynamic weights, while simultaneously updating the Mix-up interpolation weights in reverse and training the lifetime prediction model.

[0033] Figure 8 The diagram below shows a structural block diagram of a component life prediction system for a hydraulic multi-axis cooperative feeding device for a forging machine, provided in another embodiment of the present invention. Detailed Implementation

[0034] 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, not all, of the embodiments of the present invention. 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.

[0035] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0036] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0037] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention. The components of the hydraulic multi-axis cooperative feeding device for forging machines include core components such as servo valves, hydraulic cylinders, piston pumps, and hydraulic motors. In this embodiment, the hydraulic motor is used as the target object for life prediction, but the life prediction target of the present invention is not limited to the hydraulic motor, and can also be applied to other hydraulic components in the multi-axis cooperative feeding device.

[0038] like Figure 1 As shown, a method for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for a forging machine, provided in the first embodiment of the present invention, is illustrated. This method includes the following steps:

[0039] Obtain the actual working data of the components in the hydraulic multi-axis coordinated feeding device of the forging machine;

[0040] Build component simulation models, verify simulation models, and collect simulation data;

[0041] The source domain dataset is constructed using simulation data. Core feature variables with the same data structure as the source domain are extracted from real working data and preprocessed to construct the target domain dataset.

[0042] A hybrid domain is constructed by combining source domain samples and target domain samples;

[0043] A lifespan prediction model is constructed, comprising a feature extractor, a predictor, a domain adaptation module, and a joint optimization module. The feature extractor is constructed using a fusion algorithm of TCN (Temporal Convolutional Network) and Transformer. Trajectory data from the source and target domains are linearly interpolated to generate a mixed-domain input. The source, target, and mixed-domain data are input into the feature extractor, which extracts hidden-layer features and mixed-domain features. These hidden-layer features include both source and target-domain features. The source and target-domain features are fused using a mix-up approach to obtain enhanced mixed features. The source features are input into the predictor for supervised training to obtain the source-domain training loss. The mixed-domain features and the mixed-domain features are output by the predictor and synthesized to output the domain adaptation loss. In the domain adaptation module, the maximum mean difference is used to measure the distance between the source, target, and mixed-domain feature distributions. In the joint optimization module, dynamic weights are used to adaptively adjust the balance between the source and domain adaptation losses, while simultaneously updating the mix-up interpolation weights in reverse and training the lifespan prediction model to obtain a trained lifespan prediction model.

[0044] The trained lifetime prediction model is used to predict the remaining lifetime of components over a future period, and the prediction results are output.

[0045] like Figure 2 As shown, the lifetime prediction model includes a feature extractor, a predictor, a domain adaptation module, and a joint optimization module. Trajectory data from the source and target domains are linearly interpolated to generate a mixed-domain input, which is then fed into a feature extractor consisting of a linear layer, a TCN, a positional encoder, and Transformer encoders to extract hidden features with long-term and short-term dependencies. Subsequently, the source and target domain features output by the feature extractor are further fused using a mix-up approach to obtain enhanced mixed-domain features. The source domain features are used for supervised training of the predictor, while the mixed and enhanced features are output by the predictor and synthesized into a domain adaptation loss. In the domain adaptation module, the maximum mean difference is used to measure the distance between the feature distributions of the source, target, and mixed domains, achieving effective alignment of the three in the feature space. Through joint optimization, dynamic weights are used to adaptively adjust the balance between the source domain loss and the domain adaptation loss, while the mix-up interpolation weights are updated in reverse, improving the model's cross-domain generalization ability and adaptive performance.

[0046] In this invention, the goal of the hydraulic motor life prediction task is to predict the future remaining life of the motor based on its historical operating data. Specifically, the model learns from degradation patterns (e.g., the decreasing trend of volumetric efficiency over time) in a large number of historical samples to infer the time interval between the motor's current state and a preset failure threshold. Finally, the model's prediction results are compared with the actual retirement time to evaluate its prediction accuracy and generalization ability across operating conditions.

[0047] Each sample sequence is defined as consisting of historical observations at time steps (tobs). The historical data of the i-th hydraulic motor sample can be represented as:

[0048]

[0049] in, This represents the volumetric efficiency, inlet and outlet pressure, speed, oil temperature, and other characteristics collected at the t-th time step, where i is the ID of the motor sequence. Volumetric efficiency η v The calculation formula is:

[0050]

[0051] Among them, V ii V represents the motor input displacement at no-load pressure, in mL / r. ie q represents the motor input displacement at the test pressure, in mL / r. Vii q represents the motor input flow rate at no-load pressure, in L / min. Vie n is the motor input flow rate at test pressure, in L / min. i n is the motor speed under no-load pressure, in r / min; e Motor speed during test pressure, in r / min; q Voi q represents the motor output flow rate at no-load pressure, in L / min; Vdi q represents the motor leakage flow rate under no-load pressure, in L / min. Voe Motor output flow rate at test pressure, in L / min; q Vde The motor leakage flow rate is measured in L / min during the test pressure.

[0052] Correspondingly, the life expectancy prediction target for the future time period can be expressed as:

[0053]

[0054] in, This represents the predicted remaining lifetime starting from time point t.

[0055] For cross-domain lifetime prediction tasks, there exists a sufficient set of simulation source domain datasets. and a small number of target domain datasets from real experiments Predictive models need to be based on source domain data. and The training process learns domain-invariant features.

[0056] To address this, a feature extractor G(·) is designed to extract and align features from the source and target domains, minimizing the data distribution differences between the two domains, i.e.:

[0057]

[0058] This allows target domain data to be better utilized for prediction using models trained in the source domain. Furthermore, by introducing a predictor F(·) to decode features into predicted values, a complete prediction model is ultimately constructed that, after training in the source domain and domain adaptation, can generalize to the target domain.

[0059]

[0060] During the testing phase, the domain adaptation model is deployed to the new target domain, using the target domain input. Generate output if Compared with the true target domain trajectory Y T A smaller gap between the values ​​indicates that the model has good domain adaptability.

[0061] like Figure 3 As shown, the specific steps for obtaining the actual working data of the components in the hydraulic multi-axis cooperative feeding device of the forging machine include:

[0062] Operating configuration of hydraulic multi-axis coordinated feed device for forging machine;

[0063] Sensor placement and accuracy calibration are performed at the inlet and outlet locations of components.

[0064] The time series data obtained from continuous experiments are exported and outliers are removed.

[0065] Specifically, the test preparation work is completed on the hydraulic multi-axis coordinated feeding device of the forging machine, including basic debugging steps such as system pressure adjustment, accumulator pre-charging, and cooling device configuration, and test parameters such as load pressure, impact frequency and typical operating conditions are set according to the requirements of the life prediction task.

[0066] Pressure, flow, and speed sensors are installed at key locations such as the inlet and outlet of the hydraulic motor to ensure the collection of complete and continuous operating data. The collected signals mainly include the inlet and outlet flow rate, displacement, speed of the hydraulic motor, and relevant characteristic information during no-load operation.

[0067] Continuous monitoring and real-time data recording were conducted during the experiment to ensure data integrity and validity. After the experiment, the obtained raw time-series data underwent unified processing, including outlier removal and signal correction, thereby providing a higher-quality input data foundation for subsequent feature extraction and lifetime modeling.

[0068] like Figure 4 As shown, the specific steps for building a hydraulic motor simulation model, verifying the model, and collecting simulation data include:

[0069] Define the requirements and operating conditions in the simulation task;

[0070] Build component models in the simulation platform;

[0071] The key parameters in the model were initially calibrated, and the accuracy of the simulation model was verified by comparing it with the experiment.

[0072] Design parameter combination schemes, generate simulation tasks under different usage conditions in batches, monitor the simulation process, and eliminate non-convergent or unreasonable data.

[0073] Export the simulation data and perform data preprocessing.

[0074] Specifically, firstly, the actual operating conditions of the hydraulic motor in the hydraulic multi-axis collaborative feeding device of the forging machine are clarified. Based on the actual application requirements, the boundary conditions of the simulation task are set, including key parameters such as pressure range, impact frequency, and operating cycle, to ensure that the established model has good engineering adaptability and representativeness.

[0075] Build a hydraulic motor system model in the simulation platform. The model should include components such as the hydraulic motor body, inlet and outlet pipelines, and load unit, and set the corresponding physical parameters such as oil properties, volumetric efficiency, leakage channels, and elastic elements.

[0076] Subsequently, the key parameters in the model (such as sealing gap, leakage coefficient, friction coefficient, etc.) were initially calibrated, and benchmark verification was carried out in combination with actual experimental data to ensure that the simulation output and the measured data maintain a high degree of consistency in indicators such as efficiency and flow rate.

[0077] After model validation, design parameter perturbation combinations are used to generate simulation tasks in batches covering different operating conditions (such as load changes, speed fluctuations, and ambient temperature differences). During the simulation, the convergence status is monitored in real time, and outlier samples with calculation divergence are removed.

[0078] Finally, the simulation data is exported and preprocessed, including standardizing the time step, removing outliers, and smoothing the signal, to provide a standardized, high-quality source domain data foundation for subsequent feature extraction and lifetime prediction modeling.

[0079] The source domain dataset and target domain dataset are constructed as follows: In constructing the source domain dataset, simulation-generated data is used as the primary basis. The obtained data is sliced ​​using a sliding window method, with a window length of 40 and a step size of 1. Within each window, the first 30 time steps are used as historical observation sequences as input, and the last 10 time steps are used as future lifespan prediction targets as output, constructing input-output sample pairs in a uniform format. The generated sample dataset is further divided into training, validation, and test sets in a 6:1:3 ratio to ensure the completeness and consistency of model training, tuning, and evaluation.

[0080] The construction of the target domain dataset is based on actual collected experimental data of hydraulic motors. Core feature variables consistent with the source domain data structure are extracted from the raw monitoring data, and operations such as outlier removal, missing value imputation, and time alignment are completed according to a strict data preprocessing process to ensure the accuracy and usability of the data.

[0081] To eliminate the influence of different feature dimensions and improve the stability and convergence efficiency of model training, all source and target domain data are uniformly processed using the Z-Score standardization method before modeling, so that the mean of each feature dimension is 0 and the standard deviation is 1, thereby ensuring that the data enters the feature extractor for subsequent learning under a uniform scale.

[0082] like Figure 5 As shown, by using the TCN and Transformer fusion algorithm to construct a feature extractor and predictor, the specific methods for data feature extraction and prediction include:

[0083] The linear mapping layer processes the raw time-series data;

[0084] Multi-layer stacked TCN modules effectively extract local dynamic features of the trajectory;

[0085] The multi-layer Transformer encoder performs in-depth processing on the TCN output features;

[0086] The extracted composite features are decoded by a predictor with a linear layer.

[0087] Specifically, the original time-series data is projected onto a unified latent space through a linear mapping layer to standardize the input feature dimensions and improve feature representation capabilities.

[0088] By employing multi-layered stacked TCN modules, local dynamic features of the trajectory can be effectively extracted, enabling the modeling of short-term motion patterns.

[0089] A multi-layer Transformer encoder is introduced to perform in-depth processing of the TCN output features, further enhancing the ability to model long-term temporal dependencies in trajectory sequences.

[0090] The extracted comprehensive features are decoded by a predictor with a linear layer to achieve accurate prediction of the trajectory position at future time points, thus completing the supervised trajectory prediction task.

[0091] The TCN-Transformer structure fully combines the efficiency of TCN in local dynamic modeling with the strong expressive power of Transformer for global temporal dependencies, thereby constructing a unified trajectory representation that has both prediction accuracy and cross-domain generalization ability, which is convenient for subsequent lifetime prediction tasks.

[0092] like Figure 6 As shown, the specific methods for constructing a hybrid domain by combining source domain samples and target domain samples include:

[0093] A hybrid domain is constructed by combining source domain samples and target domain samples through linear interpolation, and the features of the source domain and target domain are aligned to the hybrid domain respectively.

[0094] MMD is used to calculate the differences between the source domain and the hybrid domain, as well as the differences between the target domain and the hybrid domain;

[0095] Adaptive calculation of weighting factors characterizes the similarity between the mixed domain and the source domain;

[0096] Dynamically adjust the mix-up ratio;

[0097] Optimize domain adaptation loss.

[0098] Specifically, an intermediate hybrid domain is constructed by combining a large number of source domain samples and a small number of target domain samples through linear interpolation. The features of the source domain and the target domain are then aligned to this hybrid domain to achieve a smoother domain transition.

[0099] A mix-up regularization strategy is introduced to enhance samples from mixed domains, effectively improving the model's generalization ability and robustness in the target domain. For source domain samples... and target domain samples The specific method for constructing the hybrid domain is as follows:

[0100]

[0101] Where λ is the mix-up ratio, and These are the input and label of the mixing domain, respectively. The feature-level mix-up... This process is performed between the feature extractor and the predictor F(·), allowing the output of the feature extractor to directly interact with the predictor F(·).

[0102]

[0103] Here, λ is the mix-up ratio, similar to the mix-up in the input mixing process. A larger λ will make the sample closer to the source domain, while a smaller λ will make the sample closer to the target domain.

[0104] In the mix-up strategy, the weight of the proportion λ controls the interpolation intensity, thus determining the degree of knowledge transfer. A stepwise sampling strategy is adopted, and the MMD (maximum mean difference) distance is used to measure the difference d(D) between the source domain and the mixture domain. S D mix ) and the difference d(D) between the target domain and the mixed domain. T D mix The basic idea of ​​MMD is to compare two distributions by using the mean embedding in the reproducing kernel Hilbert space, as shown in the formula:

[0105]

[0106] Where P and Q represent two probability distributions, which can be the source domain distribution and the mixture domain distribution, and the target domain distribution and the mixture domain distribution, respectively; H is the reproducing kernel Hilbert space (RKHS); and φ(·) represents the feature mapping function that maps the original sample to the RKHS. H Let x denote the norm in RKHS. i ~P,y i ~Q are data points sampled from the two distributions, respectively.

[0107] In the domain adaptation process, the initial mixed-domain distribution is closer to the target domain. Subsequently, by gradually adjusting the mix-up ratio, the mixed domain gradually moves closer to the source domain, thus achieving domain alignment between the source and target domains. This approach, starting from the target domain and gradually transitioning to the source domain, helps to gradually narrow the cross-domain gap and ensures stable knowledge transfer between the two domains. To this end, a weighting factor q is introduced to characterize the similarity between the mixed domain and the source domain.

[0108]

[0109] Where T is the control parameter, set to 0.05. The initial expected value q is small and is applied to λ using a moving average:

[0110]

[0111] Where N is the total number of iterations, n is the current iteration index, and λ n is the mixing ratio for the nth iteration.

[0112] Uniform distribution U and random perturbation are used to stabilize the training process:

[0113]

[0114] Among them, σ is the local perturbation range, which is fixed at 0.2. is the value after being perturbed randomly and truncated, that is, the final output result. U(λ n -σ, λ n +σ) represents uniform distribution random sampling in the interval [λ n -σ, λ n +σ]. Clamp(x, min, max): truncation function, which limits the value of x within the range [min, max]. If x < min, then take min; if x > max, then take max; otherwise, keep x itself. Sampled from the uniform distribution and limited within the range of. During the backpropagation process, through adaptive adjustment, the model can automatically perceive the distribution relationship between the source domain and the target domain, dynamically adjust the composition ratio of the two in the mixed samples, so that the mixed domain continuously adapts to the current cross-domain feature state, thereby realizing a flexible and stable cross-domain migration mechanism, and finally improving the overall domain alignment effect and the prediction performance of the target domain.

[0115] As Figure 7 shown, the specific method of using dynamic weight to adaptively adjust the balance between the source domain loss and the domain adaptation loss, and at the same time update the Mix-up interpolation weight in the reverse direction, and train the life prediction model includes:

[0116] Dual-domain batch sampling;

[0117] Constructing mixed-domain samples;

[0118] Calculating two types of losses;

[0119] Dynamic weight update;

[0120] Jointly optimizing the total loss.

[0121] Specifically, by constructing a mixed domain to indirectly align the source domain and target domain features, it helps to alleviate the problem of insufficient coordination between the source domain training loss and the domain adaptation loss in the joint optimization process, making the overall feature alignment process smoother and effectively reducing the fluctuation of the domain adaptation loss.

[0122] To further solve the problem of the migration distance rebound caused by the overfitting of the domain adaptation loss, an adaptive weight adjustment mechanism based on the change of the distribution distance is proposed to dynamically balance the source domain training loss and the domain adaptation loss, so as to achieve more stable and effective joint optimization. During the training process, the source domain training loss Loss1 is:

[0123]

[0124] Where MSE is the error evaluation function. Similarly, the domain adaptation loss Loss2 is:

[0125]

[0126] in, and These represent the input and label of the mixed domain, respectively. A weight adjustment parameter β is introduced into the mechanism:

[0127]

[0128] Where ε is the stability constant, set to 1e -6 β∈[0,1]. Then the total loss is:

[0129] Loss = (1-β)*Loss1 + β*Loss2

[0130] In the early stages of domain adaptation, the distribution of the mixed domain is closer to the target domain. Therefore, a larger weight needs to be assigned to the domain adaptation loss to guide it towards the source domain. As training progresses, the distance between the mixed domain and the source domain distributions gradually decreases, and the weight adjustment mechanism tends to balance, maintaining the synergistic optimization of source domain learning and domain alignment. If a distribution "bounce" phenomenon occurs later, the weight adjustment parameter β can sense changes in the migration distance in real time and dynamically increase the weight of the domain adaptation loss to promote distribution realignment.

[0131] In this example, the model uses the Adam optimizer for parameter updates, with an initial learning rate of 0.00001 and a random number seed of 2021 to ensure reproducibility. The activation function is ReLU, the batch input sample size is 128, the dropout rate is 0.1, and each training cycle consists of 200 epochs.

[0132] Throughout the training process, the global trend of loss weights gradually decreases, and in the later stages of training, more emphasis is placed on fitting the source domain task, thereby improving the model's robustness and generalization ability in the target domain.

[0133] This invention proposes a method for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for forging machines. Based on the TCN-Transformer feature extractor, this method obtains a more effective domain adaptation loss by mixing the inputs and feature distributions of the source and target domains. Simultaneously, by combining a dynamic weight adaptive adjustment strategy based on feature distribution distance, the model can flexibly balance the source domain supervision and domain alignment losses during training, thereby significantly improving the prediction accuracy and generalization robustness during the transfer learning process, and further enhancing the accuracy of hydraulic component lifespan prediction.

[0134] like Figure 8As shown, another embodiment of the present invention provides a component life prediction system for a forging machine hydraulic multi-axis cooperative feeding device, used to implement the component life prediction method for a forging machine hydraulic multi-axis cooperative feeding device described in the first embodiment. The system includes: a data acquisition module, a simulation model building module, a dataset construction module, a hybrid domain construction module, a prediction model construction module, and an output module.

[0135] The data acquisition module is used to acquire the actual working data of the components in the hydraulic multi-axis collaborative feeding device of the forging machine;

[0136] The simulation model building module is used to build component simulation models, verify simulation models, and collect simulation data.

[0137] The dataset construction module uses simulation data to construct the source domain dataset, extracts core feature variables that are consistent with the source domain data structure from real working data, and constructs the target domain dataset after data preprocessing.

[0138] The hybrid domain construction module is used to combine source domain samples and target domain samples to construct a hybrid domain;

[0139] The prediction model construction module is used to construct a lifespan prediction model. The lifespan prediction model includes a feature extractor, a predictor, a domain adaptation module, and a joint optimization module. The feature extractor is constructed using a fusion algorithm of TCN and Transformer. Trajectory data from the source domain and the target domain are linearly interpolated to generate a mixed domain input. The source domain data, target domain data, and mixed domain data are input into the feature extractor. The feature extractor extracts hidden layer features and mixed domain features. The hidden layer features include source domain features and target domain features. The source domain features and target domain features are fused using a mix-up method to obtain enhanced mixed features. The source domain features are input into the predictor for supervised training to obtain the source domain training loss. The mixed domain features and the mixed features are output by the predictor respectively and synthesized to output the domain adaptation loss. In the domain adaptation module, the maximum mean difference is used to measure the distance between the feature distributions of the source domain, target domain, and mixed domain. In the joint optimization module, dynamic weights are used to adaptively adjust the balance between the source domain loss and the domain adaptation loss. At the same time, the mix-up interpolation weights are updated in reverse, and the lifespan prediction model is trained to obtain a trained lifespan prediction model.

[0140] The output module uses a trained lifetime prediction model to predict the remaining lifetime of the component over a future time period and outputs the prediction results.

[0141] The data acquisition module includes a working condition configuration unit, a setting unit, and a data export unit;

[0142] The operating condition configuration unit is used for configuring the operating conditions of the hydraulic multi-axis coordinated feed device of the forging machine;

[0143] The setting unit is used to deploy sensors at the inlet and outlet positions of the components and to perform accuracy calibration.

[0144] The data export unit is used to export and process the time-series data obtained from continuous experiments.

[0145] The third embodiment of the present invention provides a smart terminal, which includes a processor, an input device, an output device, and a memory. The processor, input device, output device, and memory are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to call the program instructions to execute the method described in the first embodiment above.

[0146] It should be understood that, in the embodiments of the present invention, the processor may be a Central Processing Unit (CPU), but it may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0147] Input devices may include touchpads, microphones, etc., while output devices may include displays (LCDs, etc.), speakers, etc.

[0148] The memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory. For example, the memory may also store information about the device type.

[0149] In specific implementations, the processor, input device, and output device described in the embodiments of the present invention can execute the implementation methods described in the method embodiments of the present invention, or they can execute the implementation methods described in the system embodiments of the present invention, which will not be repeated here.

[0150] The present invention also provides an embodiment of a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method described in the above embodiments.

[0151] The computer-readable storage medium can be an internal storage unit of the terminal described in the foregoing embodiments, such as the terminal's hard drive or memory. The computer-readable storage medium can also be an external storage device of the terminal, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the terminal. Furthermore, the computer-readable storage medium can include both internal storage units and external storage devices of the terminal. The computer-readable storage medium is used to store the computer program and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0152] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0153] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the terminals and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0154] In the several embodiments provided in this application, it should be understood that the disclosed terminals and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices or units, or may be electrical, mechanical or other forms of connection.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for a forging machine, characterized in that, include: Obtain the actual working data of the components in the hydraulic multi-axis coordinated feeding device of the forging machine; Build component simulation models, verify simulation models, and collect simulation data; The source domain dataset is constructed using simulation data. Core feature variables with the same data structure as the source domain are extracted from real working data and preprocessed to construct the target domain dataset. A hybrid domain is constructed by combining source domain samples and target domain samples; A lifespan prediction model is constructed, comprising a feature extractor, a predictor, a domain adaptation module, and a joint optimization module. The feature extractor is constructed using a fusion algorithm of TCN and Transformer. Trajectory data from the source and target domains are linearly interpolated to generate a mixed-domain input. The source, target, and mixed-domain data are input into the feature extractor, which extracts hidden-layer features and mixed-domain features. The hidden-layer features include both source and target-domain features. These features are then fused using a mix-up method to obtain enhanced mixed features. The source features are input into the predictor for supervised training to obtain the source-domain training loss. The mixed-domain features and the mixed-domain features are output by the predictor and synthesized to output the domain adaptation loss. In the domain adaptation module, the maximum mean difference is used to measure the distance between the feature distributions of the source, target, and mixed domains. In the joint optimization module, dynamic weights are used to adaptively adjust the balance between the source and domain adaptation losses, while simultaneously updating the mix-up interpolation weights in reverse and training the lifespan prediction model to obtain a trained lifespan prediction model. The trained lifetime prediction model is used to predict the remaining lifetime of components over a future period, and the prediction results are output.

2. The method for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for forging machines as described in claim 1, characterized in that, The specific method for obtaining the actual working data of the components in the hydraulic multi-axis coordinated feeding device of the forging machine includes: Operating configuration of hydraulic multi-axis coordinated feed device for forging machine; Sensors were installed at the inlet and outlet positions of the components and their accuracy was calibrated. The time series data obtained from the continuous experiments are exported and processed.

3. The method for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for forging machines as described in claim 1, characterized in that, The specific methods for building component simulation models, verifying simulation models, and collecting simulation data include: Define the requirements and operating conditions in the simulation task; Build component models in the simulation platform; The key parameters in the model were initially calibrated, and the accuracy of the simulation model was verified by comparing it with the experiment. Design parameter combination schemes, generate simulation tasks under different usage conditions in batches, monitor the simulation process, and eliminate non-convergent or unreasonable data. Export the simulation data and perform data preprocessing.

4. The method for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for forging machines as described in claim 1, characterized in that, The specific method for constructing the source domain dataset using simulation data includes: The simulated sequence data was sliced ​​using a sliding window method, with a window length of 40 and a step size of 1. The first 30 time steps were used as historical observation data, and the last 10 time steps were used as future prediction data.

5. The method for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for forging machines as described in claim 4, characterized in that, The target domain dataset is divided into training set, validation set and test set in a ratio of 6:1:

3.

6. The method for predicting the lifespan of components in a hydraulic multi-axis cooperative feeding device for forging machines as described in claim 1, characterized in that, The specific method for constructing a hybrid domain by combining source domain samples and target domain samples includes: A hybrid domain is constructed by combining source domain samples and target domain samples through linear interpolation, and the features of the source domain and target domain are aligned to the hybrid domain respectively. MMD is used to calculate the differences between the source domain and the hybrid domain, as well as the differences between the target domain and the hybrid domain; Adaptive calculation of weighting factors characterizes the similarity between the mixed domain and the source domain; Dynamically adjust the mix-up ratio to optimize the domain adaptation loss.

7. A component life prediction system for a hydraulic multi-axis cooperative feeding device of a forging machine, characterized in that, The system is used to implement the component life prediction method of the hydraulic multi-axis cooperative feeding device for forging machines as described in any one of claims 1-6, the system comprising: a data acquisition module, a simulation model building module, a dataset construction module, a hybrid domain construction module, a prediction model construction module, and an output module; The data acquisition module is used to acquire the actual working data of the components in the hydraulic multi-axis collaborative feeding device of the forging machine; The simulation model building module is used to build component simulation models, verify simulation models, and collect simulation data. The dataset construction module uses simulation data to construct the source domain dataset, extracts core feature variables that are consistent with the source domain data structure from real working data, and constructs the target domain dataset after data preprocessing. The hybrid domain construction module is used to combine source domain samples and target domain samples to construct a hybrid domain; The prediction model construction module is used to construct a lifespan prediction model. The lifespan prediction model includes a feature extractor, a predictor, a domain adaptation module, and a joint optimization module. The feature extractor is constructed using a fusion algorithm of TCN and Transformer. Trajectory data from the source domain and the target domain are linearly interpolated to generate a mixed domain input. The source domain data, target domain data, and mixed domain data are input into the feature extractor. The feature extractor extracts hidden layer features and mixed domain features. The hidden layer features include source domain features and target domain features. The source domain features and target domain features are fused using a mix-up method to obtain enhanced mixed features. The source domain features are input into the predictor for supervised training to obtain the source domain training loss. The mixed domain features and the mixed features are output by the predictor respectively and synthesized to output the domain adaptation loss. In the domain adaptation module, the maximum mean difference is used to measure the distance between the feature distributions of the source domain, target domain, and mixed domain. In the joint optimization module, dynamic weights are used to adaptively adjust the balance between the source domain loss and the domain adaptation loss. At the same time, the mix-up interpolation weights are updated in reverse, and the lifespan prediction model is trained to obtain a trained lifespan prediction model. The output module uses a trained lifetime prediction model to predict the remaining lifetime of the component over a future time period and outputs the prediction results.

8. The component life prediction system for the hydraulic multi-axis cooperative feeding device of a forging machine as described in claim 7, characterized in that, The data acquisition module includes a working condition configuration unit, a setting unit, and a data export unit; The operating condition configuration unit is used for configuring the operating conditions of the hydraulic multi-axis coordinated feed device of the forging machine; The setting unit is used to deploy sensors at the inlet and outlet positions of the components and to perform accuracy calibration. The data export unit is used to export and process the time-series data obtained from continuous experiments.

9. A smart terminal, comprising a processor, an input device, an output device, and a memory, wherein the processor, input device, output device, and memory are interconnected, and the memory is used to store a computer program, the computer program comprising program instructions, characterized in that, The processor is configured to invoke the program instructions to perform the method as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer storage medium stores a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform the method as described in any one of claims 1 to 6.

Images