Construction method of digital twinborn model of full life cycle of automobile parts

Abstract

The invention discloses a method for constructing a digital twinborn model of a full life cycle of an automobile part, which relates to the technical field of digital twinborn, and comprises the following steps: inputting a life cycle data set into an initial life digital twinborn model, carrying out time domain local feature fusion by a feature extraction layer, capturing a degradation trend by a time sequence analysis layer, and constructing a digital twinborn model. Extracting an actual detection value in the life cycle data set; calculating a deviation value between the life prediction value and the actual detection value of the automobile part; carrying out sequential relation mapping on the deviation value to generate a dynamic causal atlas; carrying out graph traversal on the dynamic causal atlas to obtain key degradation path nodes; the dynamic causal reasoning is used for carrying out correlation analysis on the key degradation path nodes, degradation characteristic parameters are generated, the depth-first search and analytic hierarchy process are adopted, dynamic causal correlation precise modeling of the life digital twin model is achieved, and the robustness of the life digital twin model under the actual complex working condition is remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the field of digital twin technology, and in particular to a method for constructing a digital twin model of an automobile component throughout its entire life cycle. Background Art

[0002] With the rapid development of intelligent manufacturing, digital twin technology, as a core means of interaction between physical entities and virtual models, has demonstrated significant value in the full lifecycle management of automotive parts. As a cyber-physical system integration tool, digital twin models create virtual replicas of physical products and continuously update and optimize them using real-time data. This enables precise monitoring, predictive maintenance, and performance optimization of automotive parts throughout their lifecycle, significantly improving product reliability and service life.

[0003] However, existing technologies suffer from two key flaws: First, traditional digital twin model construction methods typically rely on static feature extraction and single-dimensional time series analysis, making it difficult to accurately capture the nonlinear causal relationships of component degradation under complex operating conditions, resulting in large prediction errors in the constructed digital twin models. Second, when it comes to optimizing and updating digital twin model parameters, existing solutions often use offline training or global optimization strategies, lacking a mechanism for dynamically adjusting the priority of key degradation features, making it difficult for the constructed digital twin models to adapt to the rapid changes in actual operating conditions. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for constructing a digital twin model of the entire life cycle of automotive parts to solve the problems of large prediction deviation and insufficient adaptability of traditional digital twin models.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a method for constructing a digital twin model of the entire life cycle of an automotive component, which comprises: The life cycle dataset is input into the initial life digital twin model. The feature extraction layer performs time domain local feature fusion, and the time series analysis layer captures degradation trends to form the life prediction value of the automotive parts. Extract the actual test values ​​from the life cycle data set, calculate the deviation between the life prediction value and the actual test value of the automotive parts, map the deviation to a time series relationship, and generate a dynamic cause-effect graph; Traverse the dynamic causal graph to obtain key degradation path nodes, use dynamic causal reasoning to perform correlation analysis on key degradation path nodes, and generate degradation characteristic parameters; perform multi-dimensional decoupling and weight optimization on the degradation characteristic parameters to form an optimized life digital twin model construction plan; Based on the optimized life digital twin model construction plan, the hierarchical analysis method is used to sort the feature importance and obtain the priority weight tensor. According to the priority weight tensor, the Adam optimizer is used to incrementally update the parameters of the initial life digital twin model.

[0007] As a preferred solution of the method for constructing a digital twin model of the entire life cycle of automotive parts described in the present invention, the life cycle data set includes part design parameters, part production quality inspection data, part usage data and part actual life data.

[0008] As a preferred solution of the method for constructing a digital twin model of the entire life cycle of automobile parts according to the present invention, the method of forming a life prediction value of automobile parts specifically includes the following steps: The feature extraction layer and the timing analysis layer are stacked at multiple scales through a bidirectional gated recurrent unit to construct a global lifespan digital twin model. The life cycle dataset is input into the global life digital twin model. The feature extraction layer uses multi-resolution convolution kernels and ELU activation functions to fuse local features in the time domain and generate a health index curve. The time series analysis layer uses bidirectional gating and multi-head attention mechanisms to capture degradation trends and generate long-term dependency features; The health index curve and long-term dependency characteristics are spliced ​​in a channel cascade to form the life prediction value of automobile parts.

[0009] As a preferred solution of the method for constructing a digital twin model of the entire life cycle of an automobile component according to the present invention, the calculation of the deviation between the life prediction value and the actual detection value of the automobile component specifically includes the following operations: Spline interpolation is used to dynamically segment the life cycle dataset at multiple scales to extract actual detection values; The mean absolute error method is used to accumulate and average the absolute deviations between the life prediction value and the actual detection value of automobile parts to obtain the deviation amount.

[0010] As a preferred solution of the method for constructing a digital twin model of the entire life cycle of automobile parts according to the present invention, the generating of a dynamic causal graph specifically includes the following steps: The deviations are aligned in time series and decomposed into frequency domain energy using fast Fourier transform to form a causal intensity vector. Using the time series graph version control method, we perform spatiotemporal correlation analysis of causal chain on causal strength vectors to obtain contribution quantification indicators. According to the contribution quantification indicators, the timing relationship mapping is performed using the timing graph database to generate a dynamic causal graph.

[0011] As a preferred solution of the method for constructing a digital twin model of the entire life cycle of automobile parts according to the present invention, the optimized life digital twin construction solution specifically includes the following steps: Use depth-first search to traverse the dynamic causal graph and obtain key degradation path nodes; Dynamic causal reasoning is used to perform correlation quantification on key degradation path nodes to generate a correlation causal matrix; principal component decomposition is performed on the correlation causal matrix to obtain degradation characteristic parameters; The degradation characteristic parameters are decoupled in multiple dimensions through the tensor decomposition method to obtain the decoupling features; the decoupling features are weighted optimized and weighted fused to form an optimized life digital twin model construction plan.

[0012] As a preferred solution of the method for constructing a digital twin model of the entire life cycle of automobile parts according to the present invention, the obtaining of the priority weight tensor specifically includes the following steps: The obtaining of the priority weight tensor specifically includes the following steps: Perform feature space mapping on the optimized lifespan digital twin model construction scheme to obtain multidimensional feature vectors. Use the analytic hierarchy process to perform hierarchical division and feature importance ranking on the multidimensional feature vectors to generate a hierarchical weight priority matrix. The entropy method is applied to assign multi-dimensional feature weights to the hierarchical weight priority matrix to obtain the priority weight tensor.

[0013] As a preferred solution of the method for constructing a digital twin model of the entire life cycle of automotive parts according to the present invention, wherein: the parameter incremental update of the initial life digital twin model by the Adam optimizer specifically includes the following steps: The loss function is weighted according to the priority weight tensor, and backpropagation is performed using the weighted loss function to obtain the global gradient tensor; The gradient of the global gradient tensor is calculated by exponentially weighted moving average to obtain the gradient mean and gradient square mean; The Adam optimizer is used to perform deviation correction on the gradient mean and the gradient square mean to generate an incremental parameter matrix; based on the incremental parameter matrix, gradient clipping is used to perform incremental parameter updates, and the optimized life digital twin model is output synchronously.

[0014] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the method for constructing a digital twin model of the entire life cycle of an automotive component as described in the first aspect of the present invention.

[0015] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for constructing a digital twin model of the entire life cycle of automotive parts as described in the first aspect of the present invention.

[0016] The present invention achieves the following beneficial effects: It uses depth-first search to traverse the dynamic causal graph and employs dynamic causal reasoning to quantify and decompose the correlations of key degradation path nodes into principal components, achieving precise modeling of dynamic causal relationships within the lifespan digital twin model and reducing prediction bias. It also uses the analytic hierarchy process to incrementally update the parameters of the initial lifespan digital twin model, enabling the optimized lifespan digital twin model to respond to operating condition changes in real time, significantly enhancing its robustness and reliability under complex operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 Flowchart of the method for constructing a digital twin model for the entire life cycle of automotive parts.

[0019] Figure 2 Flowchart of the process for generating lifespan predictions.

[0020] Figure 3 A flowchart constructed for a dynamic causal graph.

[0021] Figure 4 Flowchart for updating parameters of the initial life digital twin model. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a method for constructing a digital twin model of the entire life cycle of an automobile component, comprising the following steps: S1. Input the life cycle dataset into the initial life digital twin model. The feature extraction layer performs time domain local feature fusion. The time series analysis layer captures the degradation trend to form the life prediction value of the automotive parts.

[0026] The specific operations include the following: S1.1. Collect lifecycle data sets, which include part design parameters, part production quality inspection data, part usage data, and part actual life data. Part design parameters include geometric dimensions, material properties, and mechanical performance data. Geometric dimensions are acquired using CAD software, material properties are measured using material testing instruments, and mechanical performance data are acquired using finite element analysis tools. Parts production quality inspection data includes surface roughness, dimensional accuracy and hardness value data; surface roughness is measured by a three-dimensional coordinate measuring machine, dimensional accuracy is collected using a surface profiler, and hardness value is collected using a hardness tester; Part usage data includes operating temperature, load conditions, and vibration frequency. Operating temperature is collected using a temperature sensor, load conditions are monitored in real time via the on-board diagnostic unit (OBD), and vibration frequency is collected using an acceleration sensor. The actual life data of parts includes failure modes, service life, and failure occurrence time. Failure modes are collected by flaw detectors, service life is collected by timers, and failure occurrence time is obtained by the on-board diagnostic unit. Lifecycle datasets can not only enhance the understanding and control of product lifecycle management, but also optimize designs to extend service life and ultimately improve the market competitiveness of products.

[0027] S1.2. Preprocess the life cycle dataset. In the specific operation, the outlier removal algorithm is used to identify and eliminate abnormal sampling points, and the linear interpolation method is applied to fill the missing values; the time series alignment algorithm is used to unify the life cycle datasets with different sampling frequencies to the same timestamp to ensure strict time series synchronization; then the Savitzky-Golay filter is used to smooth the life cycle dataset in the time domain to suppress noise; Z-score normalization and One-Hot encoding are used for standardization and discretization to eliminate dimensional differences; finally, the principal component analysis is used to decompose the life cycle dataset to complete the compression of high-dimensional features to low-dimensional space and obtain the preprocessed life cycle dataset.

[0028] S1.3. Construct and train the initial life digital twin model. In the specific operation, the neural network is called through the PyTorch framework, and the multi-scale parallel convolution structure of 1D-CNN (one-dimensional convolutional neural network) and 3D-CNN (three-dimensional convolutional neural network) is deployed. At the same time, the ELU activation function is embedded to perform nonlinear transformation and feature fusion on the local features in the time domain to complete the construction of the feature extraction layer; the gated recurrent unit and multi-head attention structure are called through the torch.nn parameter, and the hidden layer dimension is set to 128, the number of attention heads is set to 8, the key value is set to dimension 64, and the softmax function is used for dynamic weight allocation to complete the construction of the timing analysis layer; the feature extraction layer and the timing analysis layer are cross-layered by using the bidirectional gated recurrent unit to obtain the multi-scale spatiotemporal feature weights; according to the spatiotemporal feature weights, the torch.cat function is used to perform multi-scale feature stacking on the feature extraction layer and the timing analysis layer, and normalized by LayerNorm, and the constructed initial life digital twin model is output synchronously; Next, the constructed initial life digital twin model is trained. In the specific operation, the life cycle dataset is divided into a sample set, a training set, and a validation set. On the sample set, the StandardScaler is used to perform feature scaling and data augmentation to obtain enhanced samples. On the training set, the Adam optimizer is used to backpropagate the enhanced samples, and the weighted loss function is used for error calculation. Gradient clipping is used for numerical stabilization to generate updated initial life digital twin model parameters. On the validation set, the root mean square error is used to quantify the error of the updated initial life digital twin model parameters to form a validation error. When the validation error reaches the convergence threshold for five consecutive rounds, the training is terminated, and the torch.save function is used to save the trained initial life digital twin model. It should be noted that the convergence threshold is defined based on the exponential moving average of the validation error and its value range is [0.001, 0.01].

[0029] S1.4. Generate life prediction values ​​for automotive parts through the initial life digital twin model. In the specific operation, the life cycle dataset is input into the feature extraction layer through the data loading interface, and multi-resolution convolution kernels are used to extract local features in the time domain. Small-size convolution kernels capture high-frequency local features, and large-size convolution kernels extract macro trend features to obtain multi-scale time domain features. The ELU activation function is connected after each multi-resolution convolution kernel branch to perform nonlinear transformation and feature enhancement on the multi-scale time domain features, and the activated time domain local features are output. The activated time domain local features are reduced in dimension by maximum pooling and then weightedly fused through the channel attention unit to generate multi-scale fusion features. Subsequently, the dilated convolution is used to expand the receptive field of the multi-scale fusion features, and the residual connection is combined to alleviate the gradient disappearance to form enhanced features. Finally, the enhanced features are time-aligned by the time warping algorithm, and the curve is fitted by the least squares method. After smoothing with the Savitzky-Golay filter, the health index curve is output. The temporal analysis layer uses bidirectional gated recurrent units to capture trends in the lifecycle dataset. The forward GRU captures progressive degradation trends through forward temporal processing, while the backward GRU captures sudden, abnormal degradation trends through reverse temporal processing, obtaining a bidirectional hidden state. A multi-head attention mechanism is then used to take the bidirectional hidden state as the query, key, and value inputs. Attention is calculated using scaled dot products. A causal mask matrix is ​​embedded in the calculation process to ensure temporal dependencies, generating an attention weight matrix. The specific mathematical formula is as follows: ; in, represents the attention weight matrix; represents the query matrix; represents the bond matrix; represents the characteristic dimension of the bond matrix; represents the causal mask matrix; LayerNorm is used to normalize the attention weight matrix to form normalized attention features. The normalized attention features are fused with the bidirectional hidden state through a gated residual connection, and then average pooling is performed in the time dimension to eliminate short-term fluctuations, outputting long-term dependency features with long-term degradation trend perception. The health indicator curve and the long-term dependency feature are concatenated along the channel dimension, where the number of channels of the health indicator curve is 3-dimensional and the number of channels of the long-term dependency feature is 256-dimensional. After the feature channel splicing, a 259-dimensional fusion feature is generated. The fusion feature is temporally downsampled through strided convolution to obtain temporal compression features. The gated linear unit is used to perform feature selection and numerical conversion on the temporal compression features to generate a preliminary life prediction value. Subsequently, the physical degradation parameters of the life cycle dataset are extracted, and the physical degradation parameters are weightedly fused with the preliminary life prediction value. The physical degradation parameters are then dimensional mapped through a fully connected layer and normalized using a sigmoid activation function. The final life prediction value is compressed to the [0,1] interval, and the normalized life prediction value of the automotive parts is output. The life prediction value of automotive parts can not only accurately locate performance changes in key degradation stages, but also dynamically adapt to the needs of product life cycle management, achieving full coverage from fault repair to predictive maintenance, and improving overall operational efficiency and reliability.

[0030] S2. Extract the actual detection values ​​from the life cycle data set, calculate the deviation between the life prediction value and the actual detection value of the automotive parts, map the deviation to a time series relationship, and generate a dynamic causal graph.

[0031] The specific operations include the following: S2.1. Use spline interpolation to perform dynamic multi-scale segmentation on the life cycle dataset and extract the actual detection value. In the specific operation, the timestamp of the life cycle dataset is used as the x-axis, and the multi-source data in the life cycle dataset is used as the y-axis to construct a non-uniform sampling grid in the three-dimensional coordinate system; the cubic B-spline interpolation algorithm is used to density sample the non-uniform sampling grid to obtain discrete degradation feature points; the least squares method is used to perform continuous fitting on the discrete degradation feature points to obtain a smooth degradation trajectory curve; the smooth degradation trajectory curve is dynamically multi-scale segmented through a sliding window, and the smooth degradation trajectory curve is divided into three stages: healthy period, degradation period and failure period; dynamic time warping is used to align the data time within each stage, and the Savitzky-Golay filter is used to eliminate high-frequency noise simultaneously to generate standardized stage features; the standardized stage features are subjected to principal component analysis dimensionality reduction and Z-score-based numerical conversion to finally generate the actual detection value.

[0032] S2.2. Use the mean absolute error method to accumulate and average the absolute deviations of the life prediction value and the actual detection value of the automobile parts to obtain the deviation amount. In the specific operation, the mean absolute error method is used to calculate the absolute difference between the life prediction value and the actual detection value of the automobile parts point by point to obtain the absolute deviation sequence; the time decay factor is used to accumulate the deviation of the absolute deviation sequence to generate the time-weighted deviation sum; the time-weighted deviation sum is averaged to obtain the mean absolute error; the exponentially weighted moving average is used to smooth the mean absolute error, and finally the deviation amount is output. The specific mathematical formula is as follows: ; in, Indicates the deviation, represents the smoothing coefficient; Indicates the total time length; represents the cumulative time step; represents the error calculation time step; represents the time decay factor; Indicates the error calculation time step life expectancy; Indicates the error calculation time step The actual detection value; It should be noted that the smoothing coefficient is defined based on the exponential smoothing of the mean absolute error, and its value range is (0,1); the time decay factor is defined based on the exponential decay of the historical deviation, and its value range is (0,1).

[0033] S2.3. Align the deviations in time series and perform frequency domain energy decomposition through fast Fourier transform to form a causal strength vector. Specifically, the deviations are aligned in time series through linear interpolation to equidistant timestamps, eliminating the differences in device sampling frequencies and obtaining aligned deviations. Fast Fourier transform is then applied to the aligned deviations for frequency domain energy decomposition to generate a power spectrum density distribution. The power spectrum density distribution is subjected to frequency band energy integration to extract the energy distribution of characteristic frequency bands. A Chebyshev II bandpass filter is used to separate the energy distribution of characteristic frequency bands, and the energy components are quantified through short-time Fourier transform to form a time-frequency energy matrix. Granger causality test is used to extract the lag covariance of the time-frequency energy matrix, and causal transfer fitting is performed on the lag covariance to obtain the causal influence strength. The causal influence strength is weighted through Min-Max normalization, and a causal strength vector is generated through tensor splicing.

[0034] S2.4. Use the time-series graph version control method to perform causal chain spatiotemporal correlation analysis on the causal strength vector to obtain contribution quantification indicators. In the specific operation, the causal strength vector is time-sliced ​​and version-tagged according to the time dimension to construct a versioned graph structure, in which each node stores the 8-dimensional causal strength vector of the corresponding moment, and the causal chain time axis of the causal strength vector is aligned through dynamic time warping; the time-series graph version control method is used to extract spatiotemporal features of the versioned graph structure, and graph convolution is used to aggregate node features to generate a spatiotemporal feature matrix, and the spatiotemporal correlation of the causal chain is detected on the spatiotemporal feature matrix to obtain potential causal associations; a threshold mask is applied to the potential causal association to perform noise filtering to form significant causal features; the multi-head attention mechanism is used to perform cross-time step weight allocation on the significant causal features, and the weights are normalized through Softmax to form contribution weights; according to the contribution weights, the key contributions are accumulated through path integrals, and finally the contribution quantification indicators are output.

[0035] S2.5. Based on the contribution quantification indicators, the time series graph database is used to perform time series relationship mapping to generate a dynamic causal graph. In the specific operation, the contribution quantification indicators are imported into the Neo4j time series graph database in batches through CSV, and the APOC time extension function of the Neo4j time series graph database is used to extract the causal effectiveness characteristics of the contribution quantification indicators and map them to attribute nodes with timestamps; the contribution quantification indicators are calculated by sliding the time window to obtain the cross-time step correlation strength matrix; the cross-time step correlation strength matrix is ​​weighted by time decay to obtain the decay correlation matrix, and the decay correlation matrix is ​​fitted nonlinearly by polynomial regression to generate a time series relationship chain; and the time series extension plug-in of the Neo4j time series graph database is used to map the time series relationship chain to weighted space-time edges; the Louvain algorithm is used to perform real-time clustering on the attribute nodes with timestamps and the space-time edges with weights to obtain the causal evolution pattern; the graph snapshot difference method is used to perform topological structure mapping on the causal evolution pattern, and finally a dynamic causal graph is generated.

[0036] S3. Traverse the dynamic causal graph to obtain key degradation path nodes, use dynamic causal reasoning to perform correlation analysis on key degradation path nodes, and generate degradation characteristic parameters; perform multi-dimensional decoupling and weight optimization on the degradation characteristic parameters to form an optimized life digital twin model construction plan.

[0037] The specific operations include the following: S3.1. Use depth-first search to traverse the dynamic causal graph and obtain key degradation path nodes. Specifically, define the initial degradation state node based on the component initial failure mode library. Using a depth-first search algorithm, start from the initial degradation state node and recursively traverse the dynamic causal graph. During the search, record the path node sequence and edge weights of the current branch in real time. Terminate the current branch search when: reaching a node with no out-degree, the total path length exceeds 10 nodes, or the cumulative edge weight product falls below the attenuation threshold. After terminating the search, integrate the reached nodes to obtain the key degradation path nodes.

[0038] It should be noted that a node with no out-degree refers to a terminal node with no outgoing edges in the dynamic causal graph; the decay threshold is defined based on the historical failure statistical characteristics of the key degradation path nodes, and its value range is [0.15, 0.35].

[0039] S3.2. Use dynamic causal reasoning to perform correlation quantification on key degradation path nodes to generate a correlation causal matrix; perform principal component decomposition on the correlation causal matrix to obtain degradation characteristic parameters. In specific operations, use dynamic causal reasoning to perform linear regression decomposition on key degradation path nodes and their adjacent nodes to obtain delay-related data, apply sliding time window to smooth the delay-related data, and generate delay correlation coefficients; use weighted average method to weighted integrate the delay correlation coefficients to obtain causal influence factors; apply L1 regularization constraint to sparsely fuse the delay correlation coefficients and causal influence factors to generate a sparse causal matrix; perform missing value completion and noise filtering on the sparse causal matrix to form a correlation causal matrix; use principal component analysis to perform eigenvalue decomposition on the correlation causal matrix and extract the first 5 largest eigenvalues; apply Z-score normalization to weight and normalize the first 5 largest eigenvalues ​​to generate corresponding eigenvectors; perform Varimax orthogonal rotation on the corresponding eigenvectors to eliminate the correlation between components, and perform importance sampling on the corresponding eigenvectors through linear interpolation to obtain degradation characteristic parameters.

[0040] S3.3. Decouple the degradation characteristic parameters in multiple dimensions through the tensor decomposition method to obtain decoupling features, optimize the weights and perform weighted fusion on the decoupling features to form an optimized life digital twin model construction plan. In the specific operation, the degradation characteristic parameters are organized into a three-dimensional tensor structure of sample, time series and feature dimensions through tensor reconstruction; in the multi-dimensional decoupling stage, the principal component of the three-dimensional tensor structure is extracted by the tensor decomposition method to obtain a low-dimensional approximate tensor, and the principal component extraction is performed on the low-dimensional approximate tensor to obtain the core tensor; the core tensor is modally expanded using matrix slicing, and the dimension is compressed through principal component analysis to generate decoupling features. ; In the weight optimization stage, a two-layer attention mechanism is constructed: the first layer allocates temporal attention to the decoupled features, and uses tanh activation to normalize the weights to obtain the importance weights of each time step; the second layer performs feature dimension aggregation on the importance weights of each time step to obtain preliminary fusion weights, and applies the softmax function to normalize the preliminary fusion weights to generate the final attention weights; in the weighted fusion stage, the decoupled features are weighted summed according to the final attention weights to obtain the fusion feature vector; the fusion feature vector is nonlinearly mapped using the fully connected function to generate an optimized life digital twin model construction plan.

[0041] S4. Based on the optimized life digital twin model construction plan, the hierarchical analysis method is used to sort the feature importance and obtain the priority weight tensor. According to the priority weight tensor, the parameters of the initial life digital twin model are incrementally updated through the Adam optimizer.

[0042] The specific operations include the following: S4.1. Perform feature space mapping on the optimized life digital twin model construction scheme to obtain multidimensional feature vectors, use the hierarchical analysis method to perform recursive hierarchical division and feature importance sorting on the multidimensional feature vectors, and generate a hierarchical weight priority matrix. In the specific operation, use cubic spline interpolation to perform equal interval sampling on the life digital twin model construction scheme to obtain a discrete feature point sequence, use linear discriminant analysis to perform feature space mapping on the discrete feature point sequence, and generate a multidimensional feature vector; use the hierarchical analysis method to perform eigenvalue decomposition on the multidimensional feature vector to obtain the initial weight vector, and simultaneously perform recursive hierarchical division on the initial weight vector to obtain multi-level elements; use the least squares method to perform linear regression fitting on the multi-level elements to obtain the feature weight distribution, perform feature importance sorting on the feature weight distribution, and obtain the importance scale; use the characteristic root method to perform weight synthesis on the importance scale to form a preliminary weight vector; perform normalization on the preliminary weight vector through Z-score normalization to eliminate dimensional differences, and perform matrix conversion through the eigenvector method to obtain a hierarchical weight priority matrix.

[0043] S4.2. Apply the entropy method to perform multi-dimensional feature weight assignment on the hierarchical weight priority matrix to obtain the priority weight tensor. In the specific operation, apply the entropy method to measure the entropy value of the hierarchical weight priority matrix, measure the discrete degree of the criterion feature column by column, and obtain the criterion entropy value vector; perform multi-dimensional feature weight assignment on the criterion entropy value vector to generate the original hierarchical weight; then perform linear transformation and normalization on the criterion entropy value vector through entropy weight conversion to construct the entropy weight correction coefficient matrix; perform Hadamard product on the entropy weight correction coefficient matrix and the original hierarchical weight to realize dynamic weight adjustment, and perform proportional adjustment through weight rebalancing to obtain the balanced weight matrix; perform tensor expansion on the balanced weight matrix through Kronecker product to generate the priority weight tensor.

[0044] S4.3. Weight the loss function according to the priority weight tensor, and perform backpropagation using the weighted loss function to obtain the global gradient tensor. Specifically, the priority weight tensor and the loss function are dimensionally expanded to ensure that the shapes of the priority weight tensor and the loss function match; the dimensionally expanded priority weight tensor is multiplied by the loss function to generate a loss tensor; the loss function is dynamically weighted according to the loss tensor to form a weighted loss function; then, the weighted loss function is backpropagated using the automatic differentiation framework of PyTorch. During the backpropagation process, the gradient flow is injected using the chain rule to generate a local gradient tensor; the local gradient tensors are aggregated using GroupNorm (group normalization) to eliminate the dimensional shift caused by priority differences and obtain the global gradient tensor; It should be noted that the chain rule refers to the derivative rule of composite functions in calculus, which is called by the autograd unit of PyTorch.

[0045] S4.4. Calculate the gradient of the global gradient tensor using an exponentially weighted moving average to obtain the gradient mean and the gradient square mean. Specifically, perform sliding statistics on the global gradient tensor using a sliding window to obtain the current gradient statistics. Initialize the current gradient statistics with a zero tensor to generate the historical gradient mean and the historical gradient square mean. It should be noted that the zero tensor refers to a tensor of the same dimension initialized with all zeros, which is called by the torch.zeros function in the PyTorch framework; According to the historical gradient mean and the historical gradient square mean, two attenuation coefficients (β1 and β2) are set through the Adam optimizer to control the update weights of the historical gradient mean and the historical gradient square mean, respectively. The exponentially weighted moving average is used for gradient calculation: the global gradient tensor is fused with the historical gradient mean according to the attenuation coefficient (β1) to generate the gradient mean; at the same time, the square of the global gradient tensor is fused with the historical gradient square mean according to the attenuation coefficient (β2) to generate the gradient square mean. Among them, the gradient mean records the directional trend of the gradient, and the gradient square mean captures the amplitude change of the gradient.

[0046] S4.5. Use the Adam optimizer to perform bias correction on the gradient mean and the gradient square mean to generate an incremental parameter matrix; based on the incremental parameter matrix, use gradient clipping to perform incremental parameter update, and synchronously output the optimized life digital twin model. In the specific operation, use the Adam optimizer to perform exponential weighted averaging on the gradient mean and the gradient square mean to obtain the gradient moment estimator, perform time step compensation on the gradient moment estimator, and generate the gradient bias; based on the gradient bias, use the bias correction factor to perform bias correction on the gradient mean and the gradient square mean to generate an unbiased estimator; apply learning rate scaling to perform parameter incremental synthesis on the unbiased estimator to form an incremental parameter matrix.

[0047] It should be noted that the bias correction factor is based on the time step decay rate and is defined by the inverse of the decay coefficient, with a value range of (1,∞); Gradient clipping is used to constrain the L2 norm of the incremental parameter matrix to prevent extreme weight updates, generate clipped parameter increments, and perform parameter iteration on the clipped parameter increments to obtain the updated life digital twin model parameters. The weight update of the life digital twin model is simultaneously triggered through an adaptive learning rate mechanism. The relative change rate of the parameter update is obtained using the sliding average statistics method. When the relative change rate of the parameter update is lower than the update threshold for three consecutive rounds, the learning rate decay is automatically triggered until the initial life digital twin model update stops, and the optimized life digital twin model is output. It should be noted that the update threshold is defined based on the statistical distribution of historical life digital twin model updates, and its value range is [1e-5, 1e-3].

[0048] This embodiment also provides a computer device suitable for the method of constructing a digital twin model of the entire life cycle of an automotive part, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the method of constructing a digital twin model of the entire life cycle of an automotive part as proposed in the above embodiment.

[0049] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0050] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a digital twin model of the entire life cycle of automotive parts as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0051] In summary, this invention achieves precise modeling of dynamic causal relationships during digital twin model construction, reducing prediction bias, by traversing the dynamic causal graph using depth-first search and applying dynamic causal reasoning to quantify and decompose the correlations of key degradation path nodes. Using the analytic hierarchy process to incrementally update the parameters of the initial lifespan digital twin model enables the optimized lifespan digital twin model to respond to operating condition changes in real time, significantly enhancing its robustness and reliability under complex operating conditions.

[0052] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for constructing a digital twin model of the entire life cycle of an automotive component, characterized by: include, The life cycle dataset is input into the initial life digital twin model. The feature extraction layer performs time domain local feature fusion, and the time series analysis layer captures degradation trends to form the life prediction value of the automotive parts. Extract the actual test values ​​from the life cycle data set, calculate the deviation between the life prediction value and the actual test value of the automotive parts, map the deviation to a time series relationship, and generate a dynamic cause-effect graph; Traverse the dynamic causal graph to obtain key degradation path nodes, use dynamic causal reasoning to perform correlation analysis on key degradation path nodes, and generate degradation characteristic parameters; Perform multi-dimensional decoupling and weight optimization on degradation characteristic parameters to form an optimized lifespan digital twin model construction plan; Based on the optimized life digital twin model construction plan, the hierarchical analysis method is used to sort the feature importance and obtain the priority weight tensor. According to the priority weight tensor, the Adam optimizer is used to incrementally update the parameters of the initial life digital twin model.

2. The method for constructing a digital twin model of the entire life cycle of an automotive component according to claim 1, wherein: The life cycle data set includes part design parameters, part production quality inspection data, part usage data and part actual life data.

3. The method for constructing a digital twin model of the entire life cycle of an automotive component according to claim 2, characterized in that: The method of forming the life prediction value of automobile parts specifically includes the following steps: The feature extraction layer and the timing analysis layer are stacked at multiple scales through a bidirectional gated recurrent unit to construct a global lifespan digital twin model. The life cycle dataset is input into the global life digital twin model. The feature extraction layer uses multi-resolution convolution kernels and ELU activation functions to fuse local features in the time domain and generate a health index curve. The time series analysis layer uses bidirectional gating and multi-head attention mechanisms to capture degradation trends and generate long-term dependency features; The health index curve and long-term dependency characteristics are spliced ​​in a channel cascade to form the life prediction value of automobile parts.

4. The method for constructing a digital twin model of the entire life cycle of an automotive component according to claim 3, characterized in that: The calculation of the deviation between the life prediction value and the actual detection value of the automobile parts specifically includes the following operations: Spline interpolation is used to dynamically segment the life cycle dataset at multiple scales to extract actual detection values; The mean absolute error method is used to accumulate and average the absolute deviations between the life prediction value and the actual detection value of automobile parts to obtain the deviation amount.

5. The method for constructing a digital twin model of the entire life cycle of an automotive component according to claim 4, characterized in that: The generation of a dynamic causal graph specifically includes the following steps: The deviations are aligned in time series and decomposed into frequency domain energy using fast Fourier transform to form a causal intensity vector. Using the time series graph version control method, we perform spatiotemporal correlation analysis of causal chain on causal strength vectors to obtain contribution quantification indicators. According to the contribution quantification indicators, the timing relationship mapping is performed using the timing graph database to generate a dynamic causal graph.

6. The method for constructing a digital twin model of the entire life cycle of an automotive component according to claim 5, characterized in that: The optimized lifespan digital twin construction scheme specifically includes the following steps: Use depth-first search to traverse the dynamic causal graph and obtain key degradation path nodes; Dynamic causal reasoning is used to perform correlation quantification on key degradation path nodes to generate a correlation causal matrix; principal component decomposition is performed on the correlation causal matrix to obtain degradation characteristic parameters; The degenerate feature parameters are decoupled in multiple dimensions through tensor decomposition to obtain decoupling features; The decoupling features are weighted optimized and weighted fused to form an optimized life digital twin model construction plan.

7. The method for constructing a digital twin model of the entire life cycle of an automotive component according to claim 6, characterized in that: The obtaining of the priority weight tensor specifically includes the following steps: Perform feature space mapping on the optimized lifespan digital twin model construction scheme to obtain multidimensional feature vectors. Use the analytic hierarchy process to perform hierarchical division and feature importance ranking on the multidimensional feature vectors to generate a hierarchical weight priority matrix. The entropy method is applied to assign multi-dimensional feature weights to the hierarchical weight priority matrix to obtain the priority weight tensor.

8. The method for constructing a digital twin model of the entire life cycle of an automotive component according to claim 1, wherein: The method of performing incremental parameter updates on the initial life digital twin model through the Adam optimizer specifically includes the following steps: The loss function is weighted according to the priority weight tensor, and backpropagation is performed using the weighted loss function to obtain the global gradient tensor; The gradient of the global gradient tensor is calculated by exponentially weighted moving average to obtain the gradient mean and gradient square mean; The Adam optimizer is used to perform deviation correction on the gradient mean and the gradient square mean to generate an incremental parameter matrix; based on the incremental parameter matrix, gradient clipping is used to perform incremental parameter updates, and the optimized life digital twin model is output synchronously.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for constructing a digital twin model of the entire life cycle of an automotive component as described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for constructing a digital twin model of the entire life cycle of an automotive component as described in any one of claims 1 to 8 are implemented.

Images