A tool clamping state evaluation method based on self-encoding reconstruction features

Abstract

The application discloses a tool clamping state evaluation method based on self-encoding reconstruction features, comprising the following steps: first, real-time acquisition of vibration, temperature, acoustic emission and strain signals is carried out by using a multi-modal sensor, the signals are pre-processed and feature fusion is carried out, and a unified feature vector is generated. Then, the feature vector is input into a dynamic sparsity adaptive self-encoder for deep feature extraction, and the sparsity parameter is dynamically adjusted. Next, by comparing the reconstructed feature vector with the original feature vector, a residual signal set is constructed, and abnormal signals are identified through multi-scale decomposition and dynamic reconstruction technology. Finally, combined with the biological inspired adaptive signal processing technology and the hybrid attention mechanism, the signal timing is analyzed, the clamping state evaluation result is generated, and the clamping operation is optimized through the fuzzy logic assisted multi-objective optimization strategy. The application effectively reduces the precision loss caused by improper clamping during machining, and improves the accuracy and production efficiency of tool clamping state evaluation.

Description

Technical Field

[0001] This invention relates to the field of tool clamping fault diagnosis technology, and in particular to a tool clamping status evaluation method based on self-encoding reconstruction features. Background Technology

[0002] In modern manufacturing, the clamping condition of cutting tools directly affects the stability of the machining process and the final quality of the workpiece. Proper clamping not only ensures machining accuracy but also prevents machining failures and increases in scrap rates. However, in actual production, monitoring and evaluating the clamping condition of cutting tools still faces numerous technical challenges, making it difficult to meet the increasingly complex machining environments and high-precision requirements.

[0003] Currently, in most machining scenarios, monitoring the clamping status still relies primarily on the operator's experience, ensuring correct clamping through manual inspection and adjustment. While this traditional method can guarantee tool clamping quality to some extent, it also has significant drawbacks. First, manual monitoring is highly dependent on the worker's experience and judgment, making it subjective and easily affected by the operator's skill level and fatigue. Furthermore, manual operation is not only time-consuming and labor-intensive, but also prone to improper clamping due to negligence in high-paced production lines. As machining tasks become more complex, manual inspection methods struggle to handle the changing signal characteristics under different working conditions, especially when real-time, high-precision monitoring is required, where the limitations of manual methods become even more apparent.

[0004] To improve production efficiency, some manufacturing companies have introduced automated monitoring systems. These systems typically rely on a single type of sensor to monitor the tool clamping status. However, existing automated monitoring systems still have several prominent problems in practical applications:

[0005] 1. Limitations of Single Signal Source: Current monitoring systems often rely on a single signal type to determine the tool clamping status. This method easily overlooks other important influencing factors, such as temperature, acoustic emission, and strain signals. Because monitoring methods based on a single signal source cannot comprehensively reflect changes in the clamping status, they may lead to misjudgments or omissions in complex machining environments, thus affecting machining quality.

[0006] 2. Insufficient signal processing capabilities: The signals generated by the cutting tool during machining often contain a large amount of noise and irrelevant information. Existing technologies typically employ simple filtering or thresholding methods to process these complex signals. However, these methods are clearly inadequate when dealing with signal fluctuations and uncertainties, are easily affected by noise interference, and lead to inaccurate monitoring results.

[0007] 3. Lack of dynamic adaptability: Existing automated monitoring systems typically rely on preset fixed rules or models. These rules and models lack dynamic adjustment capabilities and cannot adapt to real-time changes in the processing environment. When processing conditions change, existing systems often fail to adjust their monitoring strategies in a timely manner, resulting in delayed monitoring results and impacting processing efficiency.

[0008] 4. Lack of comprehensive analysis capabilities: Most existing systems only analyze single-type signals and lack the ability to comprehensively process multi-modal signals. This limitation prevents the system from utilizing complementary information from multiple signal sources and from comprehensively and accurately assessing the clamping status. Although multi-modal sensor technology has been developed, existing monitoring systems have not yet fully utilized the comprehensive analysis capabilities of multi-source signals to improve monitoring accuracy.

[0009] Therefore, how to provide a tool clamping status evaluation method based on self-encoding reconstruction features is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0010] One objective of this invention is to propose a tool clamping state evaluation method based on autoencoder reconstructed features. This invention utilizes an autoencoder algorithm to train a model on single-class normal data, learning effective feature representations, discovering latent patterns and structures in the data, improving the model's generalization ability, and effectively filtering noise and redundant information in the input data to reduce interference with the evaluation results. This invention also indirectly monitors sensor information during the machining process to solve the tool clamping accuracy problem in machine tool processing, realizing a shift from manual confirmation to intelligent adaptation, significantly reducing labor costs and machining time.

[0011] A tool clamping state evaluation method based on self-encoding reconstructed features according to an embodiment of the present invention includes the following steps:

[0012] S1. Real-time acquisition of various modal signals, including vibration, temperature, acoustic emission, and strain signals, through multimodal sensors installed near the tool;

[0013] S2. Perform data preprocessing on the acquired multimodal signals and multimodal signal feature fusion to generate a unified feature vector;

[0014] S3. Input the generated feature vector into the dynamic sparsity adaptive autoencoder. The autoencoder performs deep feature extraction on the fused signal and dynamically adjusts the sparsity parameters to adapt to signal changes under different working conditions.

[0015] S4. Compare the reconstructed feature vector output by the autoencoder with the original feature vector to construct a residual signal set. Based on the adaptive multi-scale decomposition and dynamic reconstruction residual detection mechanism, perform multi-scale decomposition and dynamic reconstruction on the residual signal to identify instantaneous anomalies and long-term accumulated anomaly signals.

[0016] S5. Based on bio-inspired adaptive signal processing technology, the signal processing method is adaptively adjusted according to changes in real-time signals to cope with different signal characteristics and noise levels;

[0017] S6. Perform real-time analysis on the preprocessed signal, and combine it with a temporal deep learning model based on a hybrid attention mechanism to analyze the temporal pattern and features of the signal and generate clamping status evaluation results.

[0018] S7. Based on historical data and current evaluation results, use a multi-objective optimization strategy assisted by fuzzy logic to perform fuzzy reasoning and trade-offs among multiple optimization objectives, generate an optimization strategy for the clamping state, and automatically adjust relevant parameters to optimize the clamping operation.

[0019] S8. Through operator-system interaction, model parameters are adjusted in real time, and parameter adjustment behavior is recorded and analyzed. The system automatically optimizes parameter settings through a self-learning mechanism to adapt to different processing tasks and environments.

[0020] Optionally, S3 specifically includes:

[0021] S31. Input the generated feature vector into the dynamic sparse adaptive autoencoder. The feature vector is represented as X = [x1, x2, ..., x...]. n ], where n is the feature dimension;

[0022] S32. Adaptive autoencoders consist of an encoder and a decoder. The encoder compresses the input feature vector into a latent space representation.

[0023]

[0024] Where Z is the latent space representation, f enc For encoder function, θ enc For encoder parameters, W enc Let b be the encoder weight matrix. enc Let W be the bias vector, α, ∈, and δ be small constants, and ∥W enc X∥ 2 The squared norm of the vector after the encoder's linear transformation. The term is the natural logarithm of the square of the eigenvalues;

[0025] S33. Reconstruct the input feature vector from the latent space representation using a decoder:

[0026]

[0027] in, f is the reconstructed feature vector. dec For the decoder function, θ dec For decoder parameters, W dec Let b be the decoder weight matrix. dec Let z be the bias vector, σ be the activation function, and λ, μ, and ν be the adjustment parameters. i For the i-th component in the latent space representation, sin(W dec Z) represents the sinusoidal term after the linear transformation in the latent space. The cosine term is the representation of the latent space;

[0028] S34. Calculate the reconstruction error between the original feature vector and the reconstructed feature vector:

[0029]

[0030] in, For reconstruction error, x i For the i-th component in the original feature vector, To reconstruct the i-th component in the feature vector, β is an adjustment parameter. Let be the natural logarithm of the squared error. This is the fourth power of the error term;

[0031] S35. A dynamic sparsity adjustment mechanism is adopted to adjust the sparsity parameters of the autoencoder based on the reconstruction error:

[0032]

[0033] Where, θ spar Let η be the sparsity parameter, γ be the learning rate, and δ be the regularization coefficients. i Let i be the i-th eigenvalue in the latent space. The partial derivative of the reconstruction error with respect to the sparsity parameters, where k is the specific number of sparsity parameters, log e (1+exp(-||z i ||)) represents the logarithmic term of the latent space representation. For the squared terms of the latent space representation;

[0034] S36. Through multiple iterations of training, the sparsity parameters are adaptively adjusted to optimize the reconstruction capability of the autoencoder and adapt to signal changes under different working conditions.

[0035] Optionally, S4 specifically includes:

[0036] S41. Compare the reconstructed feature vector output by the autoencoder with the original feature vector to construct a residual signal set, where each residual signal represents the difference between the original feature value and the reconstructed feature value.

[0037] S42. Perform adaptive multi-scale decomposition on the constructed residual signal set to generate components of different scales, each component representing the change of the residual signal at different time scales.

[0038] S43. Dynamically reconstruct each component after multi-scale decomposition. The reconstructed signal is adjusted according to the components of multi-scale decomposition to adapt to signal changes.

[0039] S44. Calculate the instantaneous outlier and long-term cumulative outlier for each reconstructed signal. The instantaneous outlier represents the degree of anomaly of the signal at the current moment, and the long-term cumulative outlier represents the degree of anomaly accumulated by the signal over a period of time.

[0040] S45. Identify instantaneous and long-term cumulative abnormal signals, and detect signal abnormalities by setting dynamic thresholds and dynamically adjusting the thresholds according to changes in real-time signals.

[0041] S46. Based on the set dynamic threshold, detect instantaneous and long-term cumulative abnormal signals, generate abnormal detection results, and each detection result indicates whether the corresponding feature is abnormal;

[0042] S47. Output anomaly detection results and update model parameters in real time. Continuously optimize the adaptive multi-scale decomposition and dynamic reconstruction residual detection mechanism based on new signal data and detection results to adapt to different working conditions and signal characteristic changes.

[0043] Optionally, S5 specifically includes:

[0044] S51. A bio-inspired adaptive filter is used to process the real-time signal. The output signal of the filter is represented as S. f (t), the filter processes the signal based on past signal samples and the current weight coefficients, and adapts to changes in the signal by dynamically adjusting the weight coefficients;

[0045] S52. Based on the changes in the real-time signal, the weight coefficients of the filter are adaptively adjusted. The adjustment of the weight coefficients is based on the current filtering error e(t) and the learning rate η.

[0046] S53. Calculate the characteristic values ​​of the filtered signal, including the signal amplitude, frequency, and phase:

[0047]

[0048] Where i∈{1,2,…,n} represents different dimensions of the signal, α k βk γ and δ k To adjust the parameters, N is the number of signal samples, S f (t) represents the filtered signal, s f,i (t) represents the i-th filtered signal, s i (t) represents the i-th signal in the real-time signal data, f i (t) represents the feature value of the i-th signal after filtering. The latent feature representation reconstructed by the autoencoder, where m is the dimension of the latent feature;

[0049] S54. Dynamically adjust the characteristic values ​​of the filtered signal and adaptively adjust the signal processing method according to the changes in the real-time signal.

[0050] S55. Calculate the rate of change of the signal characteristic values. The rate of change is used to evaluate the changes of the signal characteristic values ​​at different times and to capture the dynamic characteristics in the signal.

[0051] S56. Based on the rate of change of signal characteristic values, generate an adaptive strategy for signal processing, and automatically optimize parameter settings through a self-learning mechanism to adapt to different processing tasks and environments.

[0052] Optionally, S6 specifically includes:

[0053] S61. For the preprocessed signal S'(t)=[s'1(t),s'2(t),…,s' n (t)] is analyzed in real time, where s' i (t) represents the i-th signal dimension;

[0054] S62. Input the signal into a temporal deep learning model based on a hybrid attention mechanism. First, the signal features are weighted using a self-attention mechanism to capture key temporal patterns and long-term dependencies:

[0055]

[0056] Among them, H e (t) is the encoder hidden state after self-attention processing, W Q,j and W K,j These are the weights of the query and key matrices, respectively. For the latent feature representation generated by the autoencoder, α j and β j These are weighting coefficients. The result of the convolution is ∈, where ∈ is a constant term to prevent the denominator from being zero;

[0057] S63, The decoder part combines the output of the self-attention mechanism and the reconstructed features generated by the autoencoder. Further analysis of the short-term characteristics of the signal yields the clamping state evaluation result R(t):

[0058]

[0059] in, γ represents the characteristics of the reconstructed signal output by the decoder. j and δ j To adjust the parameters;

[0060] S64. Adjust the model parameters in real time based on the generated clamping status evaluation results;

[0061] S65. The model is updated regularly based on new signal data and evaluation results, and the model parameters are adaptively adjusted through continuous optimization.

[0062] Optionally, S7 specifically includes:

[0063] S71. Based on historical data and current evaluation results, construct a multi-objective optimization function, which combines the latent feature representation reconstructed by the autoencoder. and the filtered signal characteristic value S f The optimization objectives include the accuracy, robustness, and real-time performance of the clamping state. These objectives are weighted and combined using weight coefficients to generate a comprehensive optimization objective function.

[0064] S72. A fuzzy logic-assisted multi-objective optimization strategy is adopted to reason and weigh the optimization objectives. The fuzzy logic reasoning process includes accurately fuzzifying the optimization objectives and generating optimization strategies based on preset fuzzy reasoning rules. The reasoning rules are based on historical data and current evaluation results, and are adjusted for specific key factors in the clamping state to generate specific optimization strategies.

[0065] S73. Automatically adjust the relevant parameters of the model based on the generated optimization strategy:

[0066]

[0067] Where, θ new For the updated parameters, θ old Here, η is the current parameter, and η is the learning rate. To optimize the gradient of the objective function, γ j and δ j These parameters are used to precisely control weight adjustments.

[0068] S74. The iterative optimization process stops when the rate of change of the objective function reaches the convergence condition.

[0069] S75. Apply the optimized parameters to the real-time clamping operation, and continuously adjust the optimization strategy based on new signal data and evaluation results.

[0070] The beneficial effects of this invention are:

[0071] (1) This invention combines vibration, temperature, acoustic emission, and strain signals collected by multimodal sensors, and utilizes the unsupervised learning capability of autoencoders to achieve dimensionality reduction and noise reduction of the data by extracting deep features from the multimodal signals. Compared with traditional single-signal monitoring methods, this invention can comprehensively process multimodal data, reduce the interference of noise and redundant information on the evaluation results, and significantly improve the monitoring accuracy of tool clamping status and the robustness of the system.

[0072] (2) This invention, through a self-encoding reconstruction algorithm, can accurately construct residual signals and use these signals to indirectly detect the tool clamping status. Through adaptive multi-scale decomposition and dynamic reconstruction technology, the system can identify instantaneous anomalies and long-term accumulated anomaly signals in real time. Compared to traditional static signal processing methods, the dynamic adjustment capability of this invention allows it to adapt to real-time changes in machining conditions, promptly capture clamping status anomalies, reduce the problem of large-scale workpiece scrapping due to improper clamping, and lower economic losses.

[0073] (3) This invention, by introducing a fuzzy logic-assisted multi-objective optimization strategy, realizes the automatic adjustment and optimization of parameters during the clamping state evaluation process, reducing reliance on manual intervention and significantly improving production efficiency. Simultaneously, the system's intelligent parameter optimization further enhances processing quality, reduces production costs, and greatly improves overall economic benefits. Attached Figure Description

[0074] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0075] Figure 1 This is a flowchart of a tool clamping state evaluation method based on self-encoding reconstruction features proposed in this invention;

[0076] Figure 2 This is a schematic diagram of the self-encoder proposed in this invention extracting deep features from multimodal signals and generating residual signals;

[0077] Figure 3 This is a schematic diagram illustrating the application of the fuzzy logic-assisted multi-objective optimization strategy proposed in this invention in clamping state evaluation. Detailed Implementation

[0078] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0079] refer to Figure 1-3 A tool clamping state evaluation method based on autoencoder reconstructed features includes the following steps:

[0080] S1. Real-time acquisition of various modal signals, including vibration, temperature, acoustic emission, and strain signals, through multimodal sensors installed near the tool;

[0081] S2. Perform data preprocessing on the acquired multimodal signals and multimodal signal feature fusion to generate a unified feature vector;

[0082] S3. Input the generated feature vector into the dynamic sparsity adaptive autoencoder. The autoencoder performs deep feature extraction on the fused signal and dynamically adjusts the sparsity parameters to adapt to signal changes under different working conditions.

[0083] S4. Compare the reconstructed feature vector output by the autoencoder with the original feature vector to construct a residual signal set. Based on the adaptive multi-scale decomposition and dynamic reconstruction residual detection mechanism, perform multi-scale decomposition and dynamic reconstruction on the residual signal to identify instantaneous anomalies and long-term accumulated anomaly signals.

[0084] S5. Based on bio-inspired adaptive signal processing technology, the signal processing method is adaptively adjusted according to changes in real-time signals to cope with different signal characteristics and noise levels;

[0085] S6. Perform real-time analysis on the preprocessed signal, and combine it with a temporal deep learning model based on a hybrid attention mechanism to analyze the temporal pattern and features of the signal and generate clamping status evaluation results.

[0086] S7. Based on historical data and current evaluation results, use a multi-objective optimization strategy assisted by fuzzy logic to perform fuzzy reasoning and trade-offs among multiple optimization objectives, generate an optimization strategy for the clamping state, and automatically adjust relevant parameters to optimize the clamping operation.

[0087] S8. Through operator-system interaction, model parameters are adjusted in real time, and parameter adjustment behavior is recorded and analyzed. The system automatically optimizes parameter settings through a self-learning mechanism to adapt to different processing tasks and environments.

[0088] In this embodiment, S3 specifically includes:

[0089] S31. Input the generated feature vector into the dynamic sparse adaptive autoencoder. The feature vector is represented as X = [x1, x2, ..., x...]. n], where n is the feature dimension;

[0090] S32. Adaptive autoencoders consist of an encoder and a decoder. The encoder compresses the input feature vector into a latent space representation.

[0091]

[0092] Where Z is the latent space representation, f enc For encoder function, θ enc For encoder parameters, W enc Let b be the encoder weight matrix. enc Let W be the bias vector, α, ∈, and δ be small constants, and ∥W enc X∥ 2 The squared norm of the vector after the encoder's linear transformation. The term is the natural logarithm of the square of the eigenvalues;

[0093] S33. Reconstruct the input feature vector from the latent space representation using a decoder:

[0094]

[0095] in, f is the reconstructed feature vector. dec For the decoder function, θ dec For decoder parameters, W dec Let b be the decoder weight matrix. dec Let z be the bias vector, σ be the activation function, and λ, μ, and ν be the adjustment parameters. i For the i-th component in the latent space representation, sin(W dec Z) represents the sinusoidal term after the linear transformation in the latent space. The cosine term is the representation of the latent space;

[0096] S34. Calculate the reconstruction error between the original feature vector and the reconstructed feature vector:

[0097]

[0098] in, For reconstruction error, x i For the i-th component in the original feature vector, To reconstruct the i-th component in the feature vector, β is an adjustment parameter. Let be the natural logarithm of the squared error. This is the fourth power of the error term;

[0099] S35. A dynamic sparsity adjustment mechanism is adopted to adjust the sparsity parameters of the autoencoder based on the reconstruction error:

[0100]

[0101] Where, θ spar Let η be the sparsity parameter, γ be the learning rate, and δ be the regularization coefficients. i Let i be the i-th eigenvalue in the latent space. The partial derivative of the reconstruction error with respect to the sparsity parameters, where k is the specific number of sparsity parameters, log e (1+exp(-||z i ||)) represents the logarithmic term of the latent space representation. For the squared terms of the latent space representation;

[0102] S36. Through multiple iterations of training, the sparsity parameters are adaptively adjusted to optimize the reconstruction capability of the autoencoder and adapt to signal changes under different working conditions.

[0103] In this embodiment, S4 specifically includes:

[0104] S41. Compare the reconstructed feature vector output by the autoencoder with the original feature vector to construct a residual signal set, where each residual signal represents the difference between the original feature value and the reconstructed feature value.

[0105] S42. Perform adaptive multi-scale decomposition on the constructed residual signal set to generate components of different scales, each component representing the change of the residual signal at different time scales.

[0106] S43. Dynamically reconstruct each component after multi-scale decomposition. The reconstructed signal is adjusted according to the components of multi-scale decomposition to adapt to signal changes.

[0107] S44. Calculate the instantaneous outlier and long-term cumulative outlier for each reconstructed signal. The instantaneous outlier represents the degree of anomaly of the signal at the current moment, and the long-term cumulative outlier represents the degree of anomaly accumulated by the signal over a period of time.

[0108] S45. Identify instantaneous and long-term cumulative abnormal signals, and detect signal abnormalities by setting dynamic thresholds and dynamically adjusting the thresholds according to changes in real-time signals.

[0109] S46. Based on the set dynamic threshold, detect instantaneous and long-term cumulative abnormal signals, generate abnormal detection results, and each detection result indicates whether the corresponding feature is abnormal;

[0110] S47. Output anomaly detection results and update model parameters in real time. Continuously optimize the adaptive multi-scale decomposition and dynamic reconstruction residual detection mechanism based on new signal data and detection results to adapt to different working conditions and signal characteristic changes.

[0111] In this embodiment, S5 specifically includes:

[0112] S51. A bio-inspired adaptive filter is used to process the real-time signal. The output signal of the filter is represented as S. f (t), the filter processes the signal based on past signal samples and the current weight coefficients, and adapts to changes in the signal by dynamically adjusting the weight coefficients;

[0113] S52. Based on the changes in the real-time signal, the weight coefficients of the filter are adaptively adjusted. The adjustment of the weight coefficients is based on the current filtering error e(t) and the learning rate η.

[0114] S53. Calculate the characteristic values ​​of the filtered signal, including the signal amplitude, frequency, and phase:

[0115]

[0116] Where i∈{1,2,…,n} represents different dimensions of the signal, α k β k γ and δ k To adjust the parameters, N is the number of signal samples, S f (t) represents the filtered signal, s f,i (t) represents the i-th filtered signal, s i (t) represents the i-th signal in the real-time signal data, f i (t) represents the feature value of the i-th signal after filtering. The latent feature representation reconstructed by the autoencoder, where m is the dimension of the latent feature;

[0117] S54. Dynamically adjust the characteristic values ​​of the filtered signal and adaptively adjust the signal processing method according to the changes in the real-time signal.

[0118] S55. Calculate the rate of change of the signal characteristic values. The rate of change is used to evaluate the changes of the signal characteristic values ​​at different times and to capture the dynamic characteristics in the signal.

[0119] S56. Based on the rate of change of signal characteristic values, generate an adaptive strategy for signal processing, and automatically optimize parameter settings through a self-learning mechanism to adapt to different processing tasks and environments.

[0120] In this embodiment, S6 specifically includes:

[0121] S61. For the preprocessed signal S'(t)=[s'1(t),s'2(t),…,s' n (t)] is analyzed in real time, where s' i (t) represents the i-th signal dimension;

[0122] S62. Input the signal into a temporal deep learning model based on a hybrid attention mechanism. First, the signal features are weighted using a self-attention mechanism to capture key temporal patterns and long-term dependencies:

[0123]

[0124] Among them, H e (t) is the encoder hidden state after self-attention processing, W Q,j and W K,j These are the weights of the query and key matrices, respectively. For the latent feature representation generated by the autoencoder, α j and β j These are weighting coefficients. The result of the convolution is ∈, where ∈ is a constant term to prevent the denominator from being zero;

[0125] S63, The decoder part combines the output of the self-attention mechanism and the reconstructed features generated by the autoencoder. Further analysis of the short-term characteristics of the signal yields the clamping state evaluation result R(t):

[0126]

[0127] in, γ represents the characteristics of the reconstructed signal output by the decoder. j and δ j To adjust the parameters;

[0128] S64. Adjust the model parameters in real time based on the generated clamping status evaluation results;

[0129] S65. The model is updated regularly based on new signal data and evaluation results, and the model parameters are adaptively adjusted through continuous optimization.

[0130] In this embodiment, S7 specifically includes:

[0131] S71. Based on historical data and current evaluation results, construct a multi-objective optimization function, which combines the latent feature representation reconstructed by the autoencoder. and the filtered signal characteristic value S f The optimization objectives include the accuracy, robustness, and real-time performance of the clamping state. These objectives are weighted and combined using weight coefficients to generate a comprehensive optimization objective function.

[0132] S72. A fuzzy logic-assisted multi-objective optimization strategy is adopted to reason and weigh the optimization objectives. The fuzzy logic reasoning process includes accurately fuzzifying the optimization objectives and generating optimization strategies based on preset fuzzy reasoning rules. The reasoning rules are based on historical data and current evaluation results, and are adjusted for specific key factors in the clamping state to generate specific optimization strategies.

[0133] S73. Automatically adjust the relevant parameters of the model based on the generated optimization strategy:

[0134]

[0135] Where, θ new For the updated parameters, θ old Here, η is the current parameter, and η is the learning rate. To optimize the gradient of the objective function, γ j and δ j These parameters are used to precisely control weight adjustments.

[0136] S74. The iterative optimization process stops when the rate of change of the objective function reaches the convergence condition.

[0137] S75. Apply the optimized parameters to the real-time clamping operation, and continuously adjust the optimization strategy based on new signal data and evaluation results.

[0138] Example 1:

[0139] To verify the feasibility of this invention in practice, it was applied to the precision machining workshop of a Chinese aviation manufacturing company. This workshop is mainly responsible for machining core components of aero-engines. These components have extremely high requirements for machining accuracy; any minute error can lead to the scrapping of the component, thereby affecting the safety of the entire engine. Therefore, the clamping state of the cutting tools is crucial to the stability of the machining process and the final quality of the components.

[0140] In traditional machining processes, workers primarily rely on experience to determine the tool clamping status. This method is significantly subjective and easily affected by the operator's experience level and fatigue, leading to insufficient reliability in clamping status monitoring and consequently impacting machining accuracy and production efficiency. To overcome these problems, the workshop introduced a tool clamping status evaluation method based on self-encoding reconstructed features.

[0141] During machining, multimodal sensors installed near the tool and workpiece collect real-time vibration, temperature, acoustic emission, and strain signals from the tool, reflecting its working state and clamping quality. These signals are transmitted to the central control system, where an autoencoder performs deep feature extraction and noise reduction. By learning the sparse features in the signals, high-dimensional data is compressed into a low-dimensional space, filtering out noise and redundant information to generate residual signals. Subsequently, the residual signals undergo multi-scale decomposition and dynamic reconstruction to identify instantaneous and cumulative anomalies during tool clamping, preventing and correcting potential problems. The reconstructed signals enter a fuzzy logic inference system, which performs multi-objective optimization analysis based on a preset rule base, handling signal uncertainty and automatically adjusting the clamping state according to the current working conditions to optimize the machining process. Based on the results of the fuzzy logic inference, the system provides adjustment suggestions to the operator, such as re-clamping the tool or adjusting the cutting speed. After the operator makes adjustments according to the suggestions, the system collects signals again to verify the adjustment effect and continuously optimizes the clamping state. Finally, the system evaluates the clamping status throughout the entire machining process and further optimizes the machining strategy through feedback loops to ensure that the tool clamping status is always optimal, thereby improving machining accuracy and efficiency. To verify the effectiveness of this method, the workshop selected 100 identical aero-engine turbine blades for machining tests, which were conducted in two phases for comparison.

[0142] Table 1: Comparison of machining effects of tool clamping state evaluation methods based on autoencoder reconstruction features

[0143] Total number of processed blades 100 100 Number of blades with serious machining errors 2 0 Number of blades with slightly substandard surface roughness 3 1 Leaf qualification rate 95％ 99％ Average processing time per blade (hours) 2.0 1.8 Total processing time (hours) 200 180 Extended tool life - +20％ Recommended number of tool adjustments none 5 times <![CDATA[Vibration signal (m / s 2 )]]> 15.3 10.2 Temperature signal (°C) 60.0 50.0 Acoustic emission signal (dB) 85 75 Strain signal (micrometers / meters) 1200 950 <![CDATA[Crack area variation (mm 2 )]]> 200 180 Blade surface roughness (micrometers) 0.8 0.6 Blade size deviation (mm) 0.05 0.02 Tool change frequency (times / 100 blades) 2 times 1 time

[0144] The data in the table shows that the pass rate for blades using the traditional method was 95%, while the pass rate increased to 99% after using the evaluation method based on self-encoded reconstructed features. Furthermore, the new method reduced the average processing time per blade to 1.8 hours and the total processing time to 180 hours, demonstrating a 10% efficiency improvement. More importantly, the new method significantly extended tool life by 20%, which not only reduced production costs but also decreased the frequency of tool replacements.

[0145] This embodiment demonstrates the application of a tool clamping status evaluation method based on autoencoder reconstructed features in the machining of aero-engine components. By combining multimodal signal acquisition, deep feature extraction from an autoencoder, fuzzy logic reasoning, and multi-objective optimization techniques, this method achieves real-time monitoring and intelligent adjustment of the tool clamping status. Experimental data shows that this method can significantly improve machining accuracy, reduce machining errors, lower production costs, and increase production efficiency, providing strong technical support for the field of high-precision manufacturing. The successful application of this method indicates its broad application prospects in the future field of intelligent manufacturing.

[0146] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A tool clamping state evaluation method based on autoencoder reconstructed features, characterized in that, Includes the following steps: S1. By installing a multi-modal sensor near the tool, multiple modal signals, including vibration, temperature, acoustic emission, and strain signals, are collected in real time. S2. Perform data preprocessing on the acquired multimodal signals and multimodal signal feature fusion to generate a unified feature vector; S3. Input the generated feature vector into the dynamic sparsity adaptive autoencoder. The autoencoder performs deep feature extraction on the fused signal and dynamically adjusts the sparsity parameters to adapt to signal changes under different working conditions. S4. Compare the reconstructed feature vector output by the autoencoder with the original feature vector to construct a residual signal set. Based on the adaptive multi-scale decomposition and dynamic reconstruction residual detection mechanism, perform multi-scale decomposition and dynamic reconstruction on the residual signal to identify instantaneous anomalies and long-term accumulated anomaly signals. S5. Bio-inspired adaptive signal processing technology, which adaptively adjusts the signal processing method according to changes in real-time signals to cope with different signal characteristics and noise levels; S6. Perform real-time analysis on the preprocessed signal, and combine it with a temporal deep learning model based on a hybrid attention mechanism to analyze the temporal pattern and characteristics of the signal and generate clamping status evaluation results. S7. Based on historical data and current evaluation results, use a multi-objective optimization strategy assisted by fuzzy logic to perform fuzzy reasoning and trade-offs among multiple optimization objectives, generate an optimization strategy for the clamping state, and automatically adjust relevant parameters to optimize the clamping operation. S8. Through operator-system interaction, model parameters are adjusted in real time, and parameter adjustment behavior is recorded and analyzed. The system automatically optimizes parameter settings through a self-learning mechanism to adapt to different processing tasks and environments.

2. The tool clamping state evaluation method based on self-encoding reconstruction features according to claim 1, characterized in that, S3 specifically includes: S31. Input the generated feature vector into the dynamic sparse adaptive autoencoder. The feature vector is represented as... ,in, For feature dimensions; S32. Adaptive autoencoders consist of an encoder and a decoder. The encoder compresses the input feature vector into a latent space representation. ； in, Representation of the potential space, For encoder functions, For encoder parameters, Here is the encoder weight matrix. For bias vectors, , and It is a small constant. The squared L2 norm of the vector after the encoder's linear transformation. The term is the natural logarithm of the square of the eigenvalues; S33. Reconstruct the input feature vector from the latent space representation using a decoder: ； in, The reconstructed feature vector, For decoder functions, For decoder parameters, This is the decoder weight matrix. For bias vectors, For activation function, , and To adjust the parameters, For the first in the latent space representation One portion, Let the latent space represent the sinusoidal terms after the linear transformation. The cosine term is the representation of the latent space; S34. Calculate the reconstruction error between the original feature vector and the reconstructed feature vector: ； in, For reconstruction error, The first eigenvector in the original feature vector One portion, To reconstruct the 1st eigenvector One portion, To adjust the parameters, Let be the natural logarithm of the squared error. This is the fourth power of the error term; S35. A dynamic sparsity adjustment mechanism is adopted to adjust the sparsity parameters of the autoencoder based on the reconstruction error: ； in, For sparsity parameters, For learning rate, and The regularization coefficient is . For the first in the potential space 1 eigenvalue, The partial derivative of the reconstruction error with respect to the sparsity parameter, For a specific number of sparsity parameters, Let the logarithmic terms be represented by the latent space. For the squared terms of the latent space representation; S36. Through multiple iterations of training, the sparsity parameters are adaptively adjusted to optimize the reconstruction capability of the autoencoder and adapt to signal changes under different working conditions.

3. The tool clamping state evaluation method based on self-encoding reconstruction features according to claim 1, characterized in that, S4 specifically includes: S41. Compare the reconstructed feature vector output by the autoencoder with the original feature vector to construct a residual signal set, where each residual signal represents the difference between the original feature value and the reconstructed feature value. S42. Perform adaptive multi-scale decomposition on the constructed residual signal set to generate components of different scales, each component representing the change of the residual signal at different time scales. S43. Dynamically reconstruct each component after multi-scale decomposition. The reconstructed signal is adjusted according to the components of multi-scale decomposition to adapt to signal changes. S44. Calculate the instantaneous outlier and long-term cumulative outlier for each reconstructed signal. The instantaneous outlier represents the degree of anomaly of the signal at the current moment, and the long-term cumulative outlier represents the degree of anomaly accumulated by the signal over a period of time. S45. Identify instantaneous and long-term cumulative abnormal signals, and detect signal abnormalities by setting dynamic thresholds and dynamically adjusting the thresholds according to changes in real-time signals. S46. Based on the set dynamic threshold, detect instantaneous and long-term cumulative abnormal signals, generate abnormal detection results, and each detection result indicates whether the corresponding feature is abnormal; S47. Output anomaly detection results and update model parameters in real time. Continuously optimize the adaptive multi-scale decomposition and dynamic reconstruction residual detection mechanism based on new signal data and detection results to adapt to different working conditions and signal characteristic changes.

4. The tool clamping state evaluation method based on self-encoding reconstruction features according to claim 1, characterized in that, S5 specifically includes: S51. A bio-inspired adaptive filter is used to process the real-time signal. The output signal of the filter is represented as follows: The filter processes signals based on past signal samples and current weight coefficients, and adapts to signal changes by dynamically adjusting the weight coefficients. S52. Based on the changes in the real-time signal, adaptively adjust the weighting coefficients of the filter. The adjustment of the weighting coefficients is based on the filtering error at the current moment. and learning rate ; S53. Calculate the characteristic values ​​of the filtered signal, including the signal amplitude, frequency, and phase: ； in, Representing different dimensions of the signal, , , and To adjust the parameters, The number of signal samples. The filtered signal The filtered first One signal, The first in real-time signal data One signal, The filtered first The characteristic values ​​of a signal For the latent feature representation reconstructed by the autoencoder, Dimensions of latent features; S54. Dynamically adjust the characteristic values ​​of the filtered signal and adaptively adjust the signal processing method according to the changes in the real-time signal. S55. Calculate the rate of change of the signal characteristic values. The rate of change is used to evaluate the changes of the signal characteristic values ​​at different times and to capture the dynamic characteristics in the signal. S56. Based on the rate of change of signal characteristic values, generate an adaptive strategy for signal processing, and automatically optimize parameter settings through a self-learning mechanism to adapt to different processing tasks and environments.

5. The tool clamping state evaluation method based on autoencoder reconstructed features according to claim 1, characterized in that, S6 specifically includes: S61. The preprocessed signal Perform real-time analysis, among which For the first One signal dimension; S62. Input the signal into a temporal deep learning model based on a hybrid attention mechanism. First, the signal features are weighted using a self-attention mechanism to capture key temporal patterns and long-term dependencies: ； in, This is the encoder's hidden state after self-attention processing. and These are the weights of the query and key matrices, respectively. The latent feature representation generated by the autoencoder and These are weighting coefficients. The result of convolution processing. To prevent constant terms with a denominator of zero; S63, The decoder part combines the output of the self-attention mechanism and the reconstructed features generated by the autoencoder. Further analysis of the short-term characteristics of the signal generates clamping status evaluation results. : ； in, The characteristics of the reconstructed signal output by the decoder. and To adjust the parameters; S64. Adjust the model parameters in real time based on the generated clamping status evaluation results; S65. The model is updated regularly based on new signal data and evaluation results, and the model parameters are adaptively adjusted through continuous optimization.

6. The tool clamping state evaluation method based on autoencoder reconstructed features according to claim 1, characterized in that, Specifically, S7 includes: S71. Based on historical data and current evaluation results, construct a multi-objective optimization function, which combines the latent feature representation reconstructed by the autoencoder. and the characteristic values ​​of the filtered signal The optimization objectives include the accuracy, robustness, and real-time performance of the clamping state. These objectives are weighted and combined using weight coefficients to generate a comprehensive optimization objective function. S72. A fuzzy logic-assisted multi-objective optimization strategy is adopted to reason and weigh the optimization objectives. The fuzzy logic reasoning process includes accurately fuzzifying the optimization objectives and generating optimization strategies based on preset fuzzy reasoning rules. The reasoning rules are based on historical data and current evaluation results, and are adjusted for specific key factors in the clamping state to generate specific optimization strategies. S73. Automatically adjust the relevant parameters of the model based on the generated optimization strategy: ； in, For the updated parameters, For the current parameter, For learning rate, To optimize the gradient of the objective function, and These parameters are used to precisely control weight adjustments. S74. The iterative optimization process stops when the rate of change of the objective function reaches the convergence condition. S75. Apply the optimized parameters to the real-time clamping operation, and continuously adjust the optimization strategy based on new signal data and evaluation results.

Images