Stress response prediction method for calcium carbide operation robot

Abstract

The invention discloses a stress response prediction method for a calcium carbide operation robot, and belongs to the technical field of artificial intelligence, and the method comprises the steps: carrying out the simulation of the impact collision process of the calcium carbide operation robot under different working conditions, collision data, obtained through simulation, of the calcium carbide operation robot in the impact collision process under different working conditions are extracted, and a sample data set is constructed; constructing a stress prediction model; training the constructed stress prediction model by using the sample data set; and utilizing the trained stress prediction model to predict the stress of the calcium carbide impact robot under different working conditions. According to the method, the stress prediction of the calcium carbide robot is realized by fusing finite element simulation and a deep learning technology. And the calculation efficiency and robustness of the model are effectively improved. And the prediction precision of the model in a complex industrial scene is obviously improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a stress response prediction method for a calcium carbide operating robot. Background Art

[0002] Calcium carbide, an important organic chemical raw material, is widely used in steelmaking, foundry, and ferroalloy industries. Its growing demand has driven capacity expansion in the smelting industry (Xiong Moyuan. Calcium Carbide Production and Its Deep Processing [M]. Beijing: Chemical Industry Press, 2008: 147-150). Calcium carbide handling robots, key equipment in calcium carbide smelting, are constantly exposed to complex operating conditions such as high temperatures, high currents, high dust levels, and multi-axis dynamic loads. Their stress state is directly related to the safety, stability, and service life of the equipment. Accurate stress prediction is not only crucial for ensuring the safe operation of robots, improving production efficiency, and reducing maintenance costs, but also provides early warning of potential failure risks in key components, providing data support for preventive maintenance decisions.

[0003] However, traditional stress detection methods usually rely on physical sensors and finite element analysis methods, which have problems such as high installation costs, susceptibility to environmental interference, difficulty in real-time and comprehensive monitoring, complex modeling, high computational costs, and poor real-time performance. These problems make it difficult to meet the needs of rapid prediction under dynamic working conditions. Moreover, traditional finite element analysis relies on high-fidelity modeling and complex calculations, and a single simulation takes too long to meet the real-time prediction needs under dynamic working conditions. Among the existing stress detection methods based on deep learning models, although traditional deep learning models can capture time-series dependencies, they are sensitive to high-frequency noise in simulation data and ignore the nonlinear coupling relationship between multiple variables. At the same time, it is difficult to take into account both local dynamic characteristics and global variable relationships, resulting in limited prediction accuracy.

[0004] In summary, existing stress prediction methods for calcium carbide robots under dynamic impact loads have problems such as poor real-time performance, low accuracy, and poor generalization. Therefore, research on stress prediction methods for calcium carbide working robots has important theoretical significance and application value. Summary of the Invention

[0005] The present invention provides a stress response prediction method for a calcium carbide working robot to solve the technical problems of poor real-time performance, low precision and poor generalization in the existing stress prediction method of calcium carbide robots under dynamic impact loads.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] On the one hand, the present invention provides a stress response prediction method for a calcium carbide working robot

[0008] The impact and collision process of the calcium carbide operation robot under different working conditions is simulated, and the collision data of the calcium carbide operation robot under different working conditions obtained by simulation is extracted to construct a sample data set;

[0009] Construct stress prediction model;

[0010] The constructed stress prediction model is trained using the sample data set;

[0011] The trained stress prediction model is used to predict the stress of the calcium carbide impact robot under different working conditions.

[0012] Furthermore, the impact and collision process of the calcium carbide operating robot under different working conditions is simulated, and the collision data of the calcium carbide operating robot under different working conditions obtained by simulation is extracted to construct a sample data set, including:

[0013] The calcium carbide operating robot is simplified into a three-body structure consisting of a horizontal mechanism, a pitch mechanism, and an impact mechanism, and a hierarchical assembly mode is used to form a simplified three-dimensional model of the calcium carbide operating robot.

[0014] Based on the simplified three-dimensional model, ANSYS finite element simulation software was used to conduct impact collision simulation experiments on the calcium carbide working robot under different plugging materials and impact velocities;

[0015] The collision data obtained from ANSYS simulation under different plugging materials and impact velocities were extracted, and the sample data set was obtained after normalization, missing value processing, and outlier processing.

[0016] The sample data set is divided into a test set and a training set in a ratio of 3:7.

[0017] Furthermore, the collision data is the time series data of the maximum equivalent elastic strain received by the impact rod, the equivalent stress received by the bottom end of the impact platform, the equivalent stress received by the rear end of the impact rod, the equivalent elastic strain of the wall, the equivalent stress received by the wall, the displacement data of the impact platform, the impact velocity, time, impact acceleration and the maximum equivalent stress of the impact rod; wherein, when constructing the sample data set, the maximum equivalent elastic strain received by the impact rod, the equivalent stress received by the bottom end of the impact platform, the equivalent stress received by the rear end of the impact rod, the equivalent elastic strain of the wall, the equivalent stress received by the wall, the displacement data of the impact platform, the impact velocity, time and impact acceleration are used as input variables, and the maximum equivalent stress of the impact rod is used as output.

[0018] Furthermore, the stress prediction model includes an input layer, a BiLSTM layer, an improved iTransformer layer and an output layer; wherein the input layer is used to realize data input; the BiLSTM layer is used to extract local time series features of the data input by the input layer; the improved iTransformer layer is an iTransformer network improved based on multi-head differential gating and relative position encoding attention mechanism, which is used to further extract features and predict stress on the data after feature extraction by the BiLSTM layer; the output layer is used to output stress prediction results.

[0019] Furthermore, the improved iTransformer layer includes an input embedding module, an encoder module, and a projection layer module;

[0020] The input embedding module is used to convert the data after feature extraction by the BiLSTM layer into an embedding vector;

[0021] The encoder module is used to encode the output of the input embedding module through an improved DAG-Attention mechanism and a feedforward neural network to capture long-distance dependencies;

[0022] The projection layer module is used to transform the output of the encoder module into stress prediction results through nonlinear mapping.

[0023] Furthermore, the improved DAG-Attention mechanism combines the multi-head gated differential attention mechanism and relative position encoding;

[0024] The calculation process of the improved DAG-Attention mechanism includes:

[0025] First, the input X is passed through different projection matrices W Q Generate two sets of queries (Q1, Q2) and keys (K1, K2), and calculate the corresponding attention maps A1 and A2 respectively. Then, calculate the attention score matrix through the difference operation. This difference operation offsets the common mode noise in the two sets of attention, making the attention distribution more sparse and focused on key information, thereby effectively improving the performance and robustness of the model. The relevant formula is as follows:

[0026] [Q1;Q2]=XW Q ,[K1;K2]=XW K ,V=XW V

[0027] Among them, W Q represents the projection matrix of the query; W K represents the projection matrix of the key; V represents the value; W V The projection matrix representing the value;

[0028] Then, the relative position encoding is calculated, and the position difference information is added to the attention score through the relative position embedding matrix. The relative distance in time or space is dynamically integrated into the attention mechanism, enabling the model to perceive the association strength of adjacent positions in the sequence:

[0029] POS i,j =ij

[0030]

[0031] A diff =softmax(A1)-λ·softmax(A2)

[0032] Where i is the position index of Q; j is the position index of K; Q and K represent the query and key matrices respectively; pos i,j is the relative distance between positions i and j; N is the maximum length of the sequence; E rel is the relative position embedding matrix; A diff is the attention score matrix; d represents the dimension of the vector; represents the i-th query vector in the first set of query matrices; represents the i-th query vector in the second set of query matrices; λ is a learnable scalar used to balance the weights of the two attention maps. To ensure the consistency of the learning dynamics, λ is reparameterized as: λ = exp(λ q1 ·λ k1 )-exp(λ q2 ·λ k2 )+λ init ; Among them, λ q1 ,λ k1 ,λ q2 , is a learnable vector, λ init Is a constant used to initialize λ; softmax() represents the normalization function; exp() represents the exponential function; represents the d-dimensional real space;

[0033] Then a head-specific learnable bias is introduced to generate a dynamic gating mask G (h) , an independent gating bias is added to each differential attention head, and the sparsity of the attention weight is controlled by the learnable gating bias, thereby filtering out low-correlation variable interactions and achieving dynamic sparsification of the attention weight. The gating mask formula is as follows:

[0034]

[0035] [σ(A diff +b g )]⊙A diff

[0036]

[0037] Where σ is the Sigmoid function; represents the attention score matrix of the h-th attention head; is the learnable bias of the h-th attention head, initialized to a negative value to promote sparsity; b g represents a learnable bias; represents the attention weight matrix after sparseness; represents the attention weight matrix of the hth attention head after sparsification; V (h) Represents the value matrix of the h-th attention head;

[0038] Then, the gated mask is element-wise multiplied with the differential attention weight to achieve dynamic screening of attention connections:

[0039]

[0040] Finally, the weights after sparse Perform group normalization, aggregate the value vectors and concatenate the multiple results to obtain the output projection matrix. The relevant formula is as follows:

[0041]

[0042] Among them, W O is the output projection matrix after multi-head splicing, MGDA(X) is the output projection matrix; Head (h) Represents the output of the h-th attention head; GroupNorm() represents group normalization; Concat() represents concatenation operation; Head (1) Represents the output of the first attention head; Head (H) Represents the output of the last attention head.

[0043] Furthermore, the calculation process of the stress prediction model includes:

[0044] The time series data of each input variable is independently input into the BiLSTM layer through the input layer. The local time series features of the input variables are extracted through the BiLSTM network to avoid mutual interference of cross-variable information. Subsequently, the output dimension is mapped to the original variable space through linear projection to obtain an enhanced time series feature matrix, which is passed as input to the subsequent iTransformer layer. Next, the enhanced time series feature matrix is ​​processed through the improved DAG-Attention mechanism to eliminate common mode noise and filter out low-correlation interactions between variables, thereby reducing redundant calculations. The processing results of the improved DAG-Attention mechanism are passed through the LayerNorm layer to eliminate distribution differences between variables, and then the FNN layer is used to further learn the global nonlinear representation of the time series, thereby realizing multi-scale fusion of time series patterns. Finally, the final stress prediction results are output through the linear layer.

[0045] Furthermore, after the constructed stress prediction model is trained using the sample data set, the method further includes:

[0046] The trained stress prediction model is evaluated using preset evaluation indicators to verify its prediction performance.

[0047] On the other hand, the present invention further provides an electronic device, comprising a processor and a memory; wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the above method.

[0048] In yet another aspect, the present invention further provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, and the instruction is loaded and executed by a processor to implement the above method.

[0049] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0050] By integrating finite element simulation and deep learning technology, the present invention proposes a stress prediction method for calcium carbide robots based on the BiLSTM-iTransformer architecture. First, in response to the limitations of the iTransformer model in terms of sensitivity to variable differences, an iTransformer stress prediction model based on multi-head differential gating and relative position encoding attention mechanism is proposed. Through the difference-driven sparse attention mechanism, effective suppression of interference noise is achieved, effectively improving the computational efficiency and robustness of the model. Secondly, in order to enhance the model's ability to capture local temporal dependencies, the present invention adopts the BiLSTM-iTransformer architecture to construct a variable-level temporal feature extractor, effectively capturing the dynamic forward and backward dependencies of stress changes under impact loads, and realizing the collaborative modeling of local and global features. While improving the local feature capture capability of the iTransformer model, the model's prediction accuracy in complex industrial scenarios is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in 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 creative work.

[0052] Figure 1 This is a flow chart of a stress response prediction method for a calcium carbide working robot provided by an embodiment of the present invention;

[0053] Figure 2 This is a BiLSTM-DAG-iTransformer framework diagram of the stress prediction model provided by an embodiment of the present invention;

[0054] Figure 3 This is a schematic diagram of a simplified model of a multi-body structure after meshing provided by an embodiment of the present invention;

[0055] Figure 4 This is a simulated stress cloud map provided by an embodiment of the present invention;

[0056] Figure 5 Schematic diagram of the model prediction results provided by an embodiment of the present invention;

[0057] Figure 6 This is a system block diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0059] First, it should be noted that in the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "exemplarily" is intended to present concepts in a concrete manner. In addition, in the embodiments of the present invention, the meaning of "and / or" can be both or either of the two.

[0060] First embodiment

[0061] This embodiment provides a stress response prediction method for calcium carbide operating robots. First, the impact and collision process of the calcium carbide operating robot under different working conditions is simulated using ANSYS finite element simulation software to extract multi-dimensional time series data. Secondly, the extracted multi-dimensional time series data is preprocessed and divided into a training set and a test set in a ratio of 7:3. Finally, the data is input into the stress prediction model proposed in the present invention for stress prediction. The stress prediction model proposed in the present invention realizes the collaborative modeling of the relationship between local dynamic features and global variables through bidirectional time series feature extraction and difference-driven sparse attention mechanism, which significantly improves the stress prediction accuracy and computational efficiency. Specifically, the execution process of this method is as follows: Figure 1 As shown, the following steps are included:

[0062] S1, simulating the impact and collision process of the calcium carbide operating robot under different working conditions, extracting the collision data of the calcium carbide operating robot under different working conditions obtained by simulation, and constructing a sample data set;

[0063] Specifically, this embodiment uses ANSYS finite element simulation software to extract multidimensional time series data, and performs normalization processing and data set division to complete data set construction. The implementation process is as follows:

[0064] First, considering the complex structure of the calcium carbide impact robot, the calcium carbide impact robot is simplified into a three-body structure consisting of a horizontal mechanism, a pitch mechanism, and an impact mechanism according to its main structure, and a hierarchical assembly mode is used to form the final simplified three-dimensional model to improve simulation efficiency and reduce analysis time. The simplified multi-body structure model after meshing is as follows: Figure 3 Then, in order to better fit the collision behavior of the high-temperature strong impact operation robot in the smelting furnace, the ANSYS finite element simulation software was used to conduct impact collision simulation experiments on the calcium carbide impact operation robot under different plugging materials and impact speeds. The material parameters are shown in Table 1, and the obtained stress cloud diagram is shown in Figure 4 Finally, the collision data of three different materials at different speeds obtained by ANSYS simulation were selected to establish a prediction model. The specific working conditions are shown in Table 2.

[0065] Table 1 Material parameters

[0066]

[0067] Table 2 Working condition parameters

[0068]

[0069] Among them, the maximum equivalent elastic strain of the impact rod, the equivalent stress of the bottom end of the impact platform, the equivalent stress of the rear end of the impact rod, the equivalent elastic strain of the wall, the equivalent stress of the wall, the displacement data of the impact platform, the impact velocity, time, and the impact acceleration are extracted as input parameters, which have a significant impact on the stress of the impact rod, and the maximum equivalent stress of the impact rod is used as the output to construct a data set.

[0070] After normalizing, processing missing values, and processing outliers, the dataset is obtained and divided into a test set and a training set in a ratio of 3:7. The normalization formula is as follows:

[0071]

[0072] Among them, x norm represents the normalized data, x represents the original time series data, min(X) and max(X) are the minimum and maximum values ​​of the data column. Through minimum-maximum normalization, the data can be linearly mapped to the [0,1] interval.

[0073] S2, construct stress prediction model;

[0074] S3, using the sample data set to train the constructed stress prediction model;

[0075] It's important to note that the Transformer model is a deep learning architecture based on an attention mechanism. Its core concept is to use self-attention to calculate the correlation weights between all positions in the input sequence, dynamically capturing long-range dependencies and significantly improving computational efficiency. Since its introduction in 2017 by Vaswani et al. (Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017), the Transformer model has made significant progress in the field of forecasting, particularly in time series forecasting, spatiotemporal forecasting, and prediction tasks in specific fields. Research results are constantly emerging both domestically and internationally. For example, Xie et al. (Xie Y, Niu J, Zhang Y, et al. Multisize patched spatial-temporal transformer network for short-and long-term crowd flow prediction [J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(11): 21548-21568.) proposed a spatiotemporal Transformer network based on a self-attention mechanism. This network achieves rich context modeling and effectively captures cross-temporal and cross-scale dependencies between people through a cross-attention mechanism, avoiding the common problem of cumulative error in long-term prediction. However, the model has a complex structure and requires high computing resources, which limits its deployment in practical applications. To address the problem of limited computing resources, Zhou et al. (Zhou H, Zhang S, Peng J, et al. Informer: Beyond efficient transformer for long sequence time-series forecasting [C] / / Proceedings of the AAAI conference on artificial intelligence. 2021, 35 (12): 11106-11115.) proposed a new Transformer model, Informer. This model significantly solves the computational complexity, memory bottleneck, and inference speed problems of traditional models in long sequence time series forecasting through self-attention distillation technology and generative decoder. However, Informer still faces the challenge of capturing complex long-term dependencies when processing extremely long sequences.To address the shortcomings of traditional models in capturing long-term dependencies and modeling complex time series, Zhang et al. (Zhang Q, Ding J, Wang K. PR-Transformer: Long-term Prediction of Train Wheel Diameters Using Progressive Refinement Transformer[J]. IEEE Transactions on Instrumentation and Measurement, 2024.) proposed a PR-Transformer model based on time embedding and data decomposition, encoding temporal information into the model to better handle complex long-term dependencies. However, PR-Transformer still has shortcomings in single-variable prediction, data dependency, and multi-scenario generalization capabilities. Furthermore, Liu et al. (Liu Y, Hu T, Zhang H, et al. itransformer: Inverted transformers are effective for time series forecasting [J]. arXiv preprint arXiv: 2310.06625, 2023.) found that the iTransformer model, constructed by introducing methods such as dimensionality inversion, performed well in tasks such as power load forecasting, demonstrating enhanced applicability and generalization. However, while the iTransformer model performs well in multivariate time series forecasting, it lacks the ability to capture local dependencies in temporal pattern modeling. Furthermore, due to its sensitivity to noise under complex working conditions, it cannot meet the requirements for efficient and accurate stress prediction of calcium carbide operating robots under dynamic impact loads.

[0076] Based on the above, this embodiment proposes an improved iTransformer and combines it with BiLSTM to construct a Figure 2 The BiLSTM-DAG-iTransformer stress prediction model shown in the figure consists of an input layer, a BiLSTM layer, an improved iTransformer layer, and an output layer. A variable-level time series feature extractor is constructed using the BiLSTM network to extract local time series features from the input time series data. The improved iTransformer module performs further feature extraction and stress prediction. The encoder output is converted into a stress prediction result through nonlinear mapping. The following details the computational process of each model.

[0077] 1. Build a variable-level time series feature extractor through the BiLSTM network and perform local time series feature extraction on the input time series data.

[0078] LSTM uses the forget gate, input gate, and output gate to control the discarding or adding of information, thereby achieving forgetting and remembering functions. For a single LSTM unit, its calculation process at time step t can be expressed as:

[0079] f t =σ(W f ·[h t-1 ,x t ]+b f ) (2)

[0080] i t =σ(W i ·[h t-1 ,x t ]+b i ) (3)

[0081]

[0082] o t =σ(W o ·[h t-1 ,x t ]+b o ) (5)

[0083] h t =o t *tanh(C t ) (6)

[0084] Where, f t is the output of the forget gate, h t-1 is the output of the previous neuron, x t is the current input, b f is the bias, W f is the forget gate weight matrix, σ is the sigmoid activation function, i t is the input gate output, is the candidate state, o t is the output of the output gate.

[0085] By combining forward and backward LSTM, BiLSTM can capture the local dependencies of input time series data in the forward and backward time series, thereby extracting rich local time series features and further using the obtained features in the next step of stress prediction. For BiLSTM, the outputs of forward LSTM and backward LSTM are respectively expressed as and And concatenate the forward and reverse hidden states, the final output is expressed as:

[0086]

[0087] 2. Methods for further feature extraction and stress prediction based on the improved iTransformer module.

[0088] The improved iTransformer mainly includes the following sub-modules:

[0089] Input embedding module: converts the data after BiLSTM network feature extraction into an embedding vector input model;

[0090] Encoder module: Encodes the input sequence through an improved DAG-Attention mechanism and a feedforward neural network to capture long-range dependencies;

[0091] Projection layer module: Converts the encoder output into stress prediction results through nonlinear mapping.

[0092] The improved DAG-Attention mechanism combines the multi-head gated differential attention mechanism and relative position encoding to effectively enhance the model's ability to model long sequences. The calculation process is as follows:

[0093] First, the input X is passed through different projection matrices W Q Generate two sets of queries (Q1, Q2) and keys (K1, K2), and calculate the corresponding attention maps A1 and A2 respectively. Then, calculate the attention score matrix through the difference operation. This difference operation offsets the common mode noise in the two sets of attention, making the attention distribution more sparse and focused on key information, thereby effectively improving the performance and robustness of the model. The relevant formula is as follows:

[0094] [Q1;Q2]=XW Q ,[K1;K2]=XW K ,V=XW V (8)

[0095] Among them, W Q represents the projection matrix of the query; W K represents the projection matrix of the key; V represents the value; W V The projection matrix representing the value;

[0096] Then, the relative position encoding is calculated, and the position difference information is added to the attention score through the relative position embedding matrix. The relative distance in time or space is dynamically integrated into the attention mechanism, enabling the model to perceive the association strength of adjacent positions in the sequence:

[0097] POS i,j =ij (9)

[0098]

[0099] A diff =softmax(A1)-λ·softmax(A2) (12)

[0100] Where i is the position index of Q; j is the position index of K; Q and K represent the query and key matrices respectively; pos i,j is the relative distance between positions i and j; N is the maximum length of the sequence; E rel is the relative position embedding matrix; A diff is the attention score matrix; d represents the dimension of the vector; represents the i-th query vector in the first set of query matrices; represents the i-th query vector in the second set of query matrices; λ is a learnable scalar used to balance the weights of the two attention maps. To ensure the consistency of the learning dynamics, λ is reparameterized as:

[0101]

[0102] Among them, λ q1 ,λ k1 ,λ q2 , is a learnable vector, λ init Is a constant used to initialize λ; softmax() represents the normalization function; exp() represents the exponential function; represents the d-dimensional real space;

[0103] Then a head-specific learnable bias is introduced to generate a dynamic gating mask G (h) , an independent gating bias is added to each differential attention head, and the sparsity of the attention weight is controlled by the learnable gating bias, thereby filtering out low-correlation variable interactions and achieving dynamic sparsification of the attention weight. The gating mask formula is as follows:

[0104]

[0105] Where σ is the Sigmoid function; represents the attention score matrix of the h-th attention head; is the learnable bias of the h-th attention head, initialized to a negative value to promote sparsity; b g represents a learnable bias; represents the attention weight matrix after sparseness; represents the attention weight matrix of the hth attention head after sparsification; V (h) Represents the value matrix of the h-th attention head;

[0106] Then, the gated mask is element-wise multiplied with the differential attention weight to achieve dynamic screening of attention connections:

[0107]

[0108] Finally, the weights after sparse Perform group normalization, aggregate the value vectors and concatenate the multiple results to obtain the output projection matrix. The relevant formula is as follows:

[0109]

[0110] Among them, W O is the output projection matrix after multi-head splicing, MGDA(X) is the output projection matrix; Head (h) Represents the output of the h-th attention head; GroupNorm() represents group normalization; Concat() represents concatenation operation; Head (1) Represents the output of the first attention head; Head (H) Represents the output of the last attention head.

[0111] Based on the above, the DAG-Attention mechanism explicitly models local positional relationships through differential denoising, a dynamic sparsifying gating mechanism, the introduction of a relative position embedding matrix, and a multi-head-specific design. This effectively suppresses interfering noise and dynamically selects high-value variable relationships, effectively improving the model's computational efficiency and robustness. Compared to traditional methods, this method, based on a multi-head differential attention mechanism, offsets common-mode noise by calculating the difference between two sets of attention maps. It also introduces a dynamic sparsifying gating mask and relative position encoding to suppress interactions between low-correlation variables, enhancing the model's ability to model long sequences and improve its anti-interference capabilities. Furthermore, in actual working conditions, stress data generated by finite element simulation software such as ANSYS accurately reflects the stress variations of the robot under different working conditions, with clear temporal order and correlation. Therefore, utilizing the context-sensitive time series data generated by ANSYS, combined with the improved iTransformer model, can fully leverage the high precision of finite element simulation data and the regularity of the time series to achieve efficient prediction of the stress response of the calcium carbide processing robot. This innovative prediction method, which combines finite element simulation with an improved iTransformer model, addresses the problem of traditional time series prediction models' inadequate capture of local time series dependencies through the deep collaboration of BiLSTM and the improved iTransformer. It achieves collaborative modeling of local and global features, effectively improving the model's prediction accuracy. This overcomes the shortcomings of existing technologies and meets the high-precision, rapid-response requirements of calcium carbide operating robots for stress prediction under complex working conditions.

[0112] S4, using the trained stress prediction model to predict the stress of the calcium carbide impact robot under different working conditions;

[0113] Specifically, the process of achieving stress prediction using this stress prediction model is as follows:

[0114] First, the time series data for each variable is independently fed into a BiLSTM network to extract local time series features and avoid cross-variable information interference. Subsequently, a linear projection is used to map the output dimensions back to the original variable space, resulting in an enhanced time series feature matrix. This matrix is ​​then passed as input to the subsequent iTransformer module. The enhanced features are then fed into the DAG-Attention module to eliminate common-mode noise and filter out low-correlation interactions between variables, thereby reducing redundant computation. After the attention layer output, the results are sequentially passed through a LayerNorm layer to eliminate inter-variable distribution differences. The FNN layer then further learns a global nonlinear representation of the time series, enabling multi-scale fusion of time series patterns. Finally, a linear layer outputs the final stress prediction results.

[0115] S5, evaluate the stress prediction model to verify its prediction performance;

[0116] Specifically, this embodiment uses stress data under three different working conditions to evaluate the model and verify its prediction performance. The evaluation indicators include the coefficient of determination (R 2 ), MAPE (mean absolute percentage error).

[0117] The indicator calculation formula is as follows.

[0118]

[0119] Where Y predict,i is the predicted value of the i-th sample, X real,i is the true value of the i-th sample, is the mean of the true value, and n represents the length of the time series. The more accurate the model prediction, the smaller the MAPE and RMSE values, and the 2 The closer the value of is to 1.

[0120] The prediction results are as follows Figure 5 The evaluation results of the indicators are shown in Table 3.

[0121] Table 3 Index evaluation results

[0122]

[0123] In summary, this embodiment provides a stress response prediction method for calcium carbide operating robots. First, by integrating ANSYS finite element simulation data and deep learning models, an end-to-end prediction framework is constructed to replace the traditional simulation process, significantly reducing the computational time; secondly, a multi-head differential gated attention mechanism is designed, and dynamic sparse gating and relative position encoding are introduced on the basis of the multi-head differential attention mechanism to suppress high-frequency interference in the simulation data, eliminate common-mode noise, and improve the robustness of the model; finally, the bidirectional temporal feature extraction capability of BiLSTM and the global variable interaction modeling of the improved iTransformer are combined to suppress noise interference in the simulation data, and realize the coordinated capture of the local impact response and the global multi-physical field coupling relationship; thereby effectively improving the speed and accuracy of stress prediction of calcium carbide operating robots, and providing reliable technical support for equipment life assessment and safe operation in high-temperature and harsh environments.

[0124] Second embodiment

[0125] This embodiment provides an electronic device, such as Figure 6 As shown, the electronic device includes: a processor and a memory; wherein the processor and the memory can be connected via a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment described above. In addition, the electronic device may also include a transceiver; the processor and the transceiver can be connected via a communication bus, and the transceiver is used to communicate with other devices.

[0126] Next, combine Figure 6 A detailed introduction to the various components of the electronic device is given below:

[0127] Among them, the processor is the control center of the electronic device, and the electronic device may include multiple processors, each of which may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here can be a processor or a general term for multiple processing elements. For example, the processor is one or more central processing units (CPUs), or other general-purpose processors, application specific integrated circuits (ASICs), or one or more integrated circuits configured to implement an embodiment of the present invention, such as one or more microprocessors (digital signal processors, DSPs), or one or more 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, etc. The processor can perform various functions of the electronic device by running or executing software programs stored in the memory and calling data stored in the memory.

[0128] In a specific implementation, as an embodiment, the processor may include one or more CPUs, such as Figure 6 The CPU0 and CPU1 shown in FIG are, of course, only exemplary.

[0129] The memory is used to store the software program for executing the solution of the present invention, and the execution is controlled by the processor. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0130] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may be integrated with the processor or exist independently and accessed through the interface circuit ( Figure 6 (not shown) is coupled to the processor, which is not specifically limited in this embodiment of the present invention.

[0131] The transceiver may include a receiver and a transmitter ( Figure 6 The receiver is used to implement the receiving function, and the transmitter is used to implement the sending function. The transceiver can be integrated with the processor or exist independently and communicate with the electronic device through the interface circuit ( Figure 6 (not shown) is coupled to the processor, which is not specifically limited in this embodiment of the present invention.

[0132] In addition, it should be noted that Figure 6 The structure of the electronic device shown in the figure does not constitute a limitation on the device. The actual device may include more or fewer components than shown, or may combine certain components, or arrange the components differently. In addition, the technical effects achieved by the electronic device when executing the method of the first embodiment can refer to the technical effects described in the first embodiment above, and therefore will not be repeated here.

[0133] Third embodiment

[0134] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device. The instructions stored therein can be loaded by a processor in a terminal to execute the method described above.

[0135] Furthermore, it should be noted that the present invention may be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention may take the form of a fully or partially hardware embodiment, a fully or partially software embodiment, or an embodiment combining software and hardware aspects. Furthermore, when implemented using software, embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The computer program product comprises one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are fully or partially generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired connection (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a DVD), or a semiconductor medium. The semiconductor medium may be a solid state drive.

[0136] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for implementing the process in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0138] It should also be noted that, in this document, relational terms such as first and second are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between these entities or operations. The terms "include," "comprises," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or terminal device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. Without further limitation, an element defined by the phrase "comprising a..." does not preclude the presence of other identical elements in the process, method, article, or terminal device comprising the element. In addition, the term "and / or" is merely a description of an associative relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: the presence of A alone, the presence of A and B simultaneously, or the presence of B alone, where A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding. "At least one" means one or more, and "more" means two or more. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0139] In addition, it can be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0140] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0141] In the several embodiments provided herein, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of functional modules / units is merely a logical functional division. In actual implementation, other division methods may be used, such as multiple units or components being combined or integrated into another device, or some features being ignored or not implemented. Furthermore, the coupling or direct coupling or communication connection shown or discussed between each other may be through some interface, indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of the present embodiment according to actual needs. In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit.

[0142] If the method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0143] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention. It should be noted that, although preferred embodiments of the present invention have been described, those skilled in the art, once understanding the basic inventive concepts of the present invention, may make various improvements and modifications without departing from the principles of the present invention. Such improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as covering the preferred embodiments and all variations and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. A stress response prediction method for a calcium carbide working robot, characterized in that: include: The impact and collision process of the calcium carbide operation robot under different working conditions is simulated, and the collision data of the calcium carbide operation robot under different working conditions obtained by simulation is extracted to construct a sample data set; Construct stress prediction model; The constructed stress prediction model is trained using the sample data set; The trained stress prediction model is used to predict the stress of the calcium carbide impact robot under different working conditions.

2. The stress response prediction method for a calcium carbide working robot according to claim 1, characterized in that: The impact and collision process of the calcium carbide operation robot under different working conditions is simulated, and the collision data of the calcium carbide operation robot under different working conditions obtained by simulation is extracted to construct a sample data set, including: The calcium carbide operating robot is simplified into a three-body structure consisting of a horizontal mechanism, a pitch mechanism, and an impact mechanism, and a hierarchical assembly mode is used to form a simplified three-dimensional model of the calcium carbide operating robot. Based on the simplified three-dimensional model, ANSYS finite element simulation software was used to conduct impact collision simulation experiments on the calcium carbide working robot under different plugging materials and impact velocities; The collision data obtained from ANSYS simulation under different plugging materials and impact velocities were extracted, and the sample data set was obtained after normalization, missing value processing, and outlier processing. The sample data set is divided into a test set and a training set in a ratio of 3:

7.

3. The stress response prediction method for a calcium carbide working robot according to claim 1, wherein: The collision data are the time series data of the maximum equivalent elastic strain received by the impact rod, the equivalent stress received by the bottom end of the impact platform, the equivalent stress received by the rear end of the impact rod, the equivalent elastic strain of the wall, the equivalent stress received by the wall, the displacement data of the impact platform, the impact velocity, time, impact acceleration and the maximum equivalent stress of the impact rod; wherein, when constructing the sample data set, the maximum equivalent elastic strain received by the impact rod, the equivalent stress received by the bottom end of the impact platform, the equivalent stress received by the rear end of the impact rod, the equivalent elastic strain of the wall, the equivalent stress received by the wall, the displacement data of the impact platform, the impact velocity, time and impact acceleration are used as input variables, and the maximum equivalent stress of the impact rod is used as output.

4. The stress response prediction method for a calcium carbide working robot according to claim 1, wherein: The stress prediction model includes an input layer, a BiLSTM layer, an improved iTransformer layer, and an output layer; wherein the input layer is used to realize data input; the BiLSTM layer is used to extract local temporal features of the data input by the input layer; the improved iTransformer layer is an iTransformer network improved based on multi-head differential gating and relative position encoding attention mechanism, which is used to further extract features and predict stress on the data after feature extraction by the BiLSTM layer; the output layer is used to output stress prediction results.

5. The stress response prediction method for a calcium carbide working robot according to claim 4, characterized in that: The improved iTransformer layer includes an input embedding module, an encoder module, and a projection layer module; The input embedding module is used to convert the data after feature extraction by the BiLSTM layer into an embedding vector; The encoder module is used to encode the output of the input embedding module through an improved DAG-Attention mechanism and a feedforward neural network to capture long-distance dependencies; The projection layer module is used to transform the output of the encoder module into stress prediction results through nonlinear mapping.

6. The stress response prediction method for a calcium carbide working robot according to claim 5, characterized in that: The improved DAG-Attention mechanism combines the multi-head gated differential attention mechanism and relative position encoding; The calculation process of the improved DAG-Attention mechanism includes: First, the input X is projected through different matrices to generate two sets of queries Q1 and Q2 and corresponding keys K1 and K2, and the corresponding attention maps A1 and A2 are calculated respectively. Then, the attention score matrix is ​​calculated through a difference operation. This difference operation offsets the common mode noise in the two sets of attention, making the attention distribution more sparse and focused on key information, thereby effectively improving the performance and robustness of the model. The relevant formula is as follows: [Q1;Q2]=XW Q ,[K1;K2]=XW K ,V=XW V Among them, W Q represents the projection matrix of the query; W K represents the projection matrix of the key; V represents the value; W V The projection matrix representing the value; Then, the relative position encoding is calculated, and the position difference information is added to the attention score through the relative position embedding matrix. The relative distance in time or space is dynamically integrated into the attention mechanism, enabling the model to perceive the association strength of adjacent positions in the sequence: pos i,j =i-j A diff =softmax(A1)-λ·softmax(A2) Where i is the position index of Q; j is the position index of K, Q and K represent the query and key matrices respectively; pos i,j is the relative distance between the position with index i and the position with index j; N is the maximum length of the sequence; E rel is the relative position embedding matrix; A diff is the attention score matrix; d represents the dimension of the vector; represents the i-th query vector in the first set of query matrices; represents the i-th query vector in the second set of query matrices; λ is a learnable scalar used to balance the weights of the two attention maps. To ensure the consistency of the learning dynamics, λ is reparameterized as: λ = exp(λ q1 ·λ k1 )-exp(λ q2 ·λ k2 )+λ init ; Among them, λ q1 ,λ k1 ,λ q2 , is a learnable vector, λ init Is a constant used to initialize λ; softmax() represents the normalization function; exp() represents the exponential function; represents the d-dimensional real space; Then a head-specific learnable bias is introduced to generate a dynamic gating mask G (h) , an independent gating bias is added to each differential attention head, and the sparsity of the attention weight is controlled by the learnable gating bias, thereby filtering out low-correlation variable interactions and achieving dynamic sparsification of the attention weight. The gating mask formula is as follows: [σ(A diff +b g )]⊙A diff Where σ is the Sigmoid function; represents the attention score matrix of the h-th attention head; is the learnable bias of the h-th attention head, initialized to a negative value to promote sparsity; b g represents a learnable bias; represents the attention weight matrix after sparseness; represents the attention weight matrix of the hth attention head after sparsification; V (h) Represents the value matrix of the h-th attention head; Then, the gated mask is element-wise multiplied with the differential attention weight to achieve dynamic screening of attention connections: Finally, the weights after sparse Perform group normalization, aggregate the value vectors and concatenate the multiple results to obtain the output projection matrix. The relevant formula is as follows: Among them, W O is the output projection matrix after multi-head splicing, MGDA(X) is the output projection matrix; Head (h) Represents the output of the h-th attention head; GroupNorm() represents group normalization; Concat() represents concatenation operation; Head (1) Represents the output of the first attention head; Head (H) Represents the output of the last attention head.

7. The stress response prediction method for a calcium carbide working robot according to claim 5, characterized in that: The calculation process of the stress prediction model includes: The time series data of each input variable is independently input into the BiLSTM layer through the input layer. The local time series features of the input variables are extracted through BiLSTM to avoid mutual interference of cross-variable information. Subsequently, the output dimension is mapped to the original variable space through linear projection to obtain an enhanced time series feature matrix, which is passed as input to the subsequent improved iTransformer layer. Next, the enhanced time series feature matrix is ​​processed by the improved DAG-Attention mechanism to eliminate common mode noise and filter out low-correlation interactions between variables, thereby reducing redundant calculations. The processing results of the improved DAG-Attention mechanism are passed through the LayerNorm layer to eliminate distribution differences between variables, and then the global nonlinear representation of the time series is further learned through the FNN layer, thereby realizing multi-scale fusion of time series patterns. Finally, the final stress prediction results are output through the linear layer.

8. The stress response prediction method for a calcium carbide working robot according to claim 1, wherein: After training the constructed stress prediction model using the sample data set, the method further includes: The trained stress prediction model is evaluated using preset evaluation indicators to verify its prediction performance.

Images