Equipment thermally induced deformation interval parallel reasoning method based on uncertainty quantitative decomposition

Abstract

The invention discloses an equipment thermally-induced deformation interval parallel reasoning method based on uncertainty quantitative decomposition, and relates to the technical field of thermally-induced deformation reasoning, and the method comprises the steps: building a thermal characteristic test data set according to multi-node temperature data and actually measured thermally-induced deformation data under different working conditions; decomposing the thermally-induced deformation data by adopting a Socmholt wavelet basis to remove high-frequency noise, eliminating surface topography interference through band-pass filtering, and analyzing and separating a vibration-induced deformation amount by utilizing a finite element mode to obtain pure thermally-induced deformation data; and constructing a mean-variance parallel computing graph network model, performing parallel reasoning to obtain a mean value and a standard deviation of thermally induced deformation of the equipment component, and generating a thermally induced deformation prediction interval with confidence. According to the method, the data fidelity is improved through multi-step data cleaning, the credibility of the prediction interval is improved through the parallel reasoning model, and support is provided for accurate control of equipment thermally induced deformation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of thermal deformation reasoning, in particular to an equipment thermal deformation interval parallel reasoning method based on uncertainty quantization decomposition. BACKGROUND

[0002] High-end equipment manufacturing is an important embodiment of national strategic competitiveness, and its development level directly affects national economic security. Among them, nuclear power steam isolation valves, gas turbine casings, numerical control machine tool spindles and other key components are the core carriers of equipment performance, and their precision and reliability directly affect the service life and effectiveness of equipment. However, during the service of equipment, due to the influence of inevitable heat sources such as bearing friction and motor heating, thermal deformation of key components occurs, directly affecting the service performance and reliability of equipment. Thermal deformation prediction and compensation technology has become an important means to improve equipment performance due to its strong applicability, high cost-effectiveness and sustainable maintenance precision. Among them, accurate equipment thermal deformation prediction is a prerequisite for accurate thermal deformation compensation, which is of great significance to improve the service performance of high-end equipment in complex thermal environments.

[0003] Currently, the key to accurately predicting equipment thermal deformation is to obtain high-fidelity data and build high-precision models. Among them, high-fidelity sensor data acquisition mainly collects original data through multiple types of sensors, and then obtains thermal deformation by suppressing main noise through preliminary preprocessing methods such as removing outliers and moving average filtering. However, this method does not analyze the internal transformation mechanism from raw sensor data to thermal deformation, limiting the accuracy of data input into the prediction model. In addition, the construction of high-precision models is based on data-driven prediction models. Data-driven models mainly take measurable physical quantities such as temperature and stress near key components as input, and the thermal deformation of key components as output, and realize end-to-end fast calculation through machine learning and other methods, with the characteristics of fast calculation speed and strong adaptability. However, this method generally only outputs point estimates, ignoring the inherent uncertainty of multiple stages in the prediction process, and cannot meet the needs of processing risk assessment. SUMMARY

[0004] The purpose of the present application is to provide an equipment thermal deformation interval parallel reasoning method based on uncertainty quantization decomposition, which can further generate thermal deformation prediction intervals with confidence, and guarantee the precision of high-end equipment thermal deformation compensation control.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] An equipment thermal deformation interval parallel reasoning method based on uncertainty quantization decomposition, comprising the following steps:

[0007] The multi-node temperature data and the measured thermal deformation data of the equipment component under different working conditions are acquired, and a thermal characteristic test data set under different working conditions is established.

[0008] The measured thermal deformation data is decomposed in multiple frequency bands by using a Splinelet wavelet basis, and the boundary threshold of each frequency band is adaptively adjusted through local signal-to-noise ratio calculation results, so as to obtain first thermal deformation data after removing high-frequency noise.

[0009] The surface topography associated with the running characteristic frequency of the equipment component is identified and removed from the first thermal deformation data by using a threshold-adaptive band-pass filtering method, so as to obtain second thermal deformation data after removing surface topography interference.

[0010] The vibration-induced deformation data is solved by using a finite element modal analysis method and an energy spectrum density analysis method, and the second thermal deformation data is subtracted from the vibration-induced deformation data, so as to obtain pure thermal deformation data after removing vibration interference.

[0011] The mean and standard deviation of the thermal deformation of the equipment component are obtained by parallel inference of a mean-variance parallel computation graph network model according to the multi-node temperature data and the corresponding pure thermal deformation data of the equipment component under different working conditions, and a thermal deformation prediction interval with confidence is generated; the mean-variance parallel computation graph network model uses a multi-layer gated recurrent module to mine the time sequence evolution law of temperature and thermal deformation as a time sequence feature, extracts the topological connection information between nodes of the equipment component by using an attention network as a spatial feature, fuses the time sequence feature and the spatial feature to infer the mean and standard deviation of the thermal deformation in parallel, and generates a thermal deformation prediction interval with confidence.

[0012] Optionally, the different working conditions of the equipment component specifically embody combinations of different key process parameters and environmental conditions; the key process parameters include a gas turbine pressure ratio, a nuclear power main steam pressure and a machine tool spindle speed; and the environmental conditions include an environmental temperature and a cooling state.

[0013] Optionally, the multi-node temperature data and the measured thermal deformation data of the equipment component under different working conditions are acquired, and a thermal characteristic test data set under different working conditions is established, specifically including the following steps:

[0014] A thermal characteristic test bench is built around the equipment component, a plurality of temperature sensors are uniformly arranged, and an eddy current displacement sensor is directed to a thermal deformation sensitive direction of the equipment component.

[0015] The working condition of the equipment component is changed, and the temperature data of multiple nodes of the equipment component during operation are collected, so as to obtain the multi-node temperature data and the measured thermal deformation data of the equipment component under different working conditions.

[0016] Bind the multi-node temperature data, the measured thermal deformation data and the corresponding working conditions obtained under the same working condition to establish a thermal characteristic test data set under different working conditions.

[0017] Optionally, the measured thermal deformation data is decomposed in multiple frequency bands using a Symlet wavelet basis, and the boundary threshold of each frequency band is adaptively adjusted through local signal-to-noise ratio calculation results to obtain first thermal deformation data after removing high-frequency noise, including the following steps:

[0018] Before wavelet decomposition, symmetric extension preprocessing is used for the boundary, and the extension length is exponentially positively correlated with the decomposition level to ensure the calculation stability of wavelet decomposition at the endpoints.

[0019] According to the dynamic relationship between the sampling frequency of the measured thermal deformation data and the spindle speed and the feed speed, the logarithmic operation is used to determine the optimal decomposition level of the wavelet, and the measured thermal deformation data is decomposed into high-frequency detail coefficients and low-frequency approximation coefficients.

[0020] Based on the statistical standard deviation of the high-frequency detail coefficients and the low-frequency approximation coefficients, a shrinkage threshold is dynamically generated, the high-frequency detail coefficients dominated by noise are set to zero through soft thresholding, while the low-frequency approximation coefficients are retained, and key thermal deformation features are protected through energy error constraints and frequency band fidelity mechanisms.

[0021] Based on the denoised or retained wavelet coefficients at each level, the wavelet inverse transform is performed based on the overlap reservation method, the truncation effect is eliminated through smooth transition in the overlap area, the continuity of the thermal deformation time domain waveform is ensured, and the multi-scale decomposition signal is reconstructed into the first thermal deformation data after removing high-frequency noise.

[0022] Optionally, through a threshold-adaptive band-pass filtering method, surface topography associated with the operating feature frequency of the equipment component is identified and removed from the first thermal deformation data to obtain second thermal deformation data after removing surface topography interference, including:

[0023] Perform windowed Fourier transform on the spindle thermal deformation data, and dynamically construct a passband based on the spindle speed base frequency and harmonics of the equipment component.

[0024] Extract the frequency spectrum components within the passband through an elliptical filter, and remove the frequency spectrum components that remain stable in amplitude within a continuous machining cycle as surface topography errors caused by clamping eccentricity or measurement position eccentricity.

[0025] Perform zero-phase boundary extension and inverse Fourier transform on the remaining frequency spectrum after separating the surface topography error to reconstruct the second thermal deformation data after removing the surface topography interference.

[0026] Optionally, before the zero-phase boundary continuation and inverse Fourier transform of the remaining spectrum after the separation of the surface topography error, the following steps are included:

[0027] In the offline stage, an eddy current displacement sensor array is arranged equidistantly along the surface of the main shaft to collect the measured surface topography of the main shaft surface.

[0028] The consistency of the separated surface topography and the measured surface topography is verified by the Pearson correlation coefficient, and if the consistency is lower than the consistency threshold, the passband width and order of the elliptical filter are updated, and the process jumps to "extracting the passband spectrum component through the elliptical filter, and the spectrum component with stable amplitude in the continuous machining period is removed as the surface topography error caused by clamping eccentricity".

[0029] Optionally, the vibration-induced deformation data is solved by a finite element modal analysis method and an energy spectrum density analysis method, and the vibration-induced deformation data is subtracted from the second thermal-induced deformation data to obtain pure thermal-induced deformation data free from vibration interference, including the following steps:

[0030] Through finite element modal analysis, the multi-order natural frequency and mode shape of the main shaft of the numerical control machine tool under clamping constraint are solved to obtain the main shaft vibration characteristic data.

[0031] According to the bearing model matching of the numerical control machine tool main shaft, the standard vibration load density spectrum is obtained, and the main shaft modal frequency response function is calculated based on the random vibration theory and the main shaft vibration characteristic data.

[0032] For the main shaft modal frequency response function, the root mean square value and probability density distribution of each order vibration response are solved by statistical energy method, and the vibration response energy spectrum density in physical space is synthesized.

[0033] The vibration response energy spectrum density is integrated in the critical speed frequency band of the main shaft, combined with the modal superposition principle and Hooke's law, and the vibration-induced deformation data is quantitatively calculated.

[0034] The vibration-induced deformation data is separated from the second thermal-induced deformation data by using least square fitting or state space filtering algorithm to obtain pure thermal-induced deformation data free from vibration interference.

[0035] Optionally, the mean-variance parallel computing graph network model includes: a multi-layer gated recurrent unit structure, a graph attention network, a cross-channel attention mechanism, a mean channel, a fully connected layer, a variance channel, and a nonlinear activation function.

[0036] Optionally, based on multi-node temperature data and corresponding pure thermally induced deformation data of equipment components under different operating conditions, parallel inference is performed using a mean-variance parallel computation graph network model to obtain the mean and standard deviation of the thermally induced deformation of the equipment components, generating a thermally induced deformation prediction interval with confidence, specifically including:

[0037] By using a multi-layer gated loop unit structure, the temporal characteristics of the temperature sequence at each measuring point on the main shaft within the sliding time window are extracted; the temporal characteristics characterize the temporal evolution of temperature and thermally induced deformation.

[0038] A graph attention network is constructed based on the heat conduction topology of the spindle components to extract the spatial features of each measuring point on the spindle surface. The node representation dimension in the graph attention network matches the number of temperature measuring points, and the weight of each edge is inversely proportional to the length of the heat flow path.

[0039] Temporal and spatial features are fused using a cross-channel attention mechanism to obtain spatiotemporal fusion features.

[0040] The spatiotemporal fusion features are input into the mean channel and mapped through a fully connected layer to output the trend thermal deformation; the trend thermal deformation represents the steady-state component of the principal axis axial thermal elongation.

[0041] The spatiotemporal fusion features are input into the variance channel and the residual deformation is output through a nonlinear activation function; the residual deformation represents the uncertainty caused by the fluctuation of the material's thermal conductivity and the disturbance of the cooling boundary.

[0042] Using the trend term thermal deformation as the baseline deformation and the residual deformation as the uncertain deformation component, a confidence-based thermal deformation prediction interval is derived through Gaussian probability integral.

[0043] Alternatively, the Gaussian probability integral is as follows:

[0044] P(μ t -z 0.025 σ t ≤δ≤μ t +z 0.025 σ t = 95%.

[0045] Where P() is the probability, μ t σ represents the trend of thermal deformation. t z is the standard deviation of the residual deformation. 0.025 Here, denoted by quantile of the standard normal distribution, and δ represents the actual thermally induced deformation of the equipment; the 95% confidence interval for predicting thermally induced deformation is [μ]. t -z 0.025 σ t ,μ t +z 0.025 σ t ].

[0046] According to the specific embodiments provided in the present application, the present application discloses the following technical effects:

[0047] The present application provides an equipment thermal deformation interval parallel inference method based on uncertainty quantization decomposition. In the method, multi-node temperature data and measured thermal deformation data of equipment components under different working conditions are obtained, the equipment operation scene is covered by the multi-working condition data, comprehensive data support is provided for subsequent quantization decomposition and model inference, and the adaptability of the method to complex service environments is ensured. After that, first, the multiscale decomposition capability of the Sincmelet wavelet basis and the adaptive threshold strategy are used to accurately suppress sensor noise, electromagnetic interference and other high-frequency disturbances, and the time domain fidelity of the thermal deformation data is improved. Second, the threshold adaptive band-pass filtering method is used to identify and remove the surface topography associated with the equipment operation characteristic frequency, eliminate its false contribution to the thermal deformation, and make the second thermal deformation data more focused on the deformation nature driven by temperature. Third, the natural mode analysis of the finite element modal and the vibration energy quantization of the energy spectrum density are combined to accurately separate dynamic deformation interference such as bearing vibration and structural resonance. The pure thermal deformation data obtained can accurately represent the dynamic deformation process driven by the evolution of the temperature field, and the data fidelity is improved. Finally, the temperature-thermal deformation time-varying lag effect and periodic pattern are captured by means of the gating cycle module, the heat conduction topological relationship between the measuring points is analyzed by the attention network, and after the spatio-temporal feature fusion, the mean value (describing the steady-state thermal deformation trend) and the standard deviation (quantifying the uncertainty of parameter fluctuation and boundary disturbance) are derived in parallel, so that the generated prediction interval accurately covers the trend deformation and reasonably quantifies the reliable range, providing a high-reliability decision basis for thermal deformation compensation control. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0049] Figure 1 A flowchart of an equipment thermal deformation interval parallel inference method based on uncertainty quantization decomposition provided by an embodiment of the present application.

[0050] Figure 2 A flowchart of step S1 in an equipment thermal deformation interval parallel inference method based on uncertainty quantization decomposition provided by an embodiment of the present application.

[0051] Figure 3 A flowchart of step S2 in an equipment thermal deformation interval parallel inference method based on uncertainty quantization decomposition provided by an embodiment of the present application.

[0052] Figure 4 A flow chart of step S3 in a parallel inference method for equipment thermal deformation interval based on uncertainty quantization decomposition according to an embodiment of the present application.

[0053] Figure 5 A flow chart of step S4 in a parallel inference method for equipment thermal deformation interval based on uncertainty quantization decomposition according to an embodiment of the present application.

[0054] Figure 6 A structural schematic diagram of a mean-variance parallel calculation graph network model in a parallel inference method for equipment thermal deformation interval based on uncertainty quantization decomposition according to an embodiment of the present application.

[0055] Figure 7 A flow chart of step S5 in a parallel inference method for equipment thermal deformation interval based on uncertainty quantization decomposition according to an embodiment of the present application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0057] In order to make the above objectives, characteristics and advantages of the present application more apparent, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0058] A parallel inference method for equipment thermal deformation interval based on uncertainty quantization decomposition according to an embodiment of the present application includes the following steps in an exemplary embodiment, as shown in Figure 1

[0059] S1, obtaining multi-node temperature data and measured thermal deformation data of equipment components under different working conditions, and establishing thermal characteristic test data sets under different working conditions. In the present embodiment, different working conditions of equipment components specifically represent combinations of different key process parameters and environmental conditions; the key process parameters include gas turbine pressure ratio, nuclear power main steam pressure, and machine tool spindle speed, etc.; the environmental conditions include environmental temperature and cooling state, etc.

[0060] In the present embodiment, as shown in Figure 2

[0061] ​​S11, build a thermal characteristic test bench around the equipment component, uniformly arrange multiple temperature sensors and direct the eddy current displacement sensor to the sensitive direction of thermal deformation of the equipment component.

[0062] S12, change the working condition of the equipment component, collect the temperature data of multiple nodes of the equipment component during operation, and obtain the multi-node temperature data and the measured thermal deformation data of the equipment component under different working conditions.

[0063] S13, bind the multi-node temperature data, the measured thermal deformation data obtained under the same working condition, and the corresponding working condition, and establish a thermal characteristic test data set under different working conditions.

[0064] In an exemplary embodiment, a thermal characteristic test bench of a numerical control machine tool spindle is established, multiple temperature sensors are uniformly arranged on the surface of the spindle shell and near the motor, and the eddy current displacement sensor is directed to the spindle axial direction. Change the key process parameters such as spindle speed and feed speed, and environmental conditions such as environmental temperature and cooling state, collect the thermal deformation data during operation, bind the working condition with the collected temperature-thermal deformation data, and establish a thermal characteristic experimental data set under different working conditions.

[0065] S2, decompose the measured thermal deformation data into multiple frequency bands using the Symlet wavelet basis, and adaptively adjust the boundary threshold of each frequency band through the local signal-to-noise ratio calculation result to obtain the first thermal deformation data after removing high-frequency noise. In this embodiment, as shown in Figure 3 , step S2 includes the following steps:

[0066] S21, before wavelet decomposition, symmetric extension preprocessing is used to process the boundary, and the extension length is exponentially positively correlated with the decomposition layer number, to ensure the calculation stability of the wavelet decomposition at the end point. The end point refers to the end point of the equipment start-up-thermal steady state stage.

[0067] S22, according to the dynamic relationship between the sampling frequency of the measured thermal deformation data and the spindle speed and feed speed, determine the optimal wavelet decomposition level by logarithmic operation, and decompose the measured thermal deformation data into high-frequency detail coefficients and low-frequency approximation coefficients.

[0068] S23, based on the statistical standard deviation of the high-frequency detail coefficients and the low-frequency approximation coefficients, dynamically generate a shrinkage threshold, set the high-frequency detail coefficients dominated by noise to zero through soft threshold, while retaining the low-frequency approximation coefficients, and protect the key thermal deformation features through energy error constraint and frequency band fidelity mechanism.

[0069] S24, based on the denoised or retained wavelet coefficients of each layer, perform wavelet inverse transform reconstruction based on the overlap reservation method, eliminate the truncation effect through smooth transition in the overlap area, ensure the continuity of the thermal deformation time domain waveform, and reconstruct the multi-scale decomposition signal into the first thermal deformation data after removing high-frequency noise.

[0070] S3, identifying and removing the surface topography associated with the operating characteristic frequency of the equipment component from the first thermal-induced deformation data to obtain second thermal-induced deformation data without surface topography interference. In the embodiment, as shown in FIG. 3, step S3 specifically includes the following steps: Figure 4

[0071] S31, performing windowed Fourier transform on the spindle thermal-induced deformation data, and constructing a passband based on the spindle speed fundamental frequency and harmonic dynamics of the equipment component.

[0072] S32, extracting the frequency spectrum components within the passband by an elliptical filter, and removing the frequency spectrum components with stable amplitude within the continuous machining period as surface topography errors caused by clamping eccentricity or measurement position eccentricity.

[0073] S33, performing zero-phase boundary continuation and inverse Fourier transform on the remaining frequency spectrum after separating the surface topography errors to reconstruct the second thermal-induced deformation data without surface topography interference.

[0074] Before step S33, the following steps are further included: arranging an eddy current displacement sensor array equidistantly along the spindle surface in the offline stage, and collecting the measured surface topography of the spindle surface. The consistency of the separated surface topography and the measured surface topography is verified by the Pearson correlation coefficient, and if the consistency is lower than a consistency threshold, the passband width and order of the elliptical filter are updated, and the process jumps to step S2.

[0075] S4, solving the vibration-induced deformation data by the finite element modal analysis method and the energy spectrum density analysis method, and subtracting the vibration-induced deformation data from the second thermal-induced deformation data to obtain pure thermal-induced deformation data without vibration interference. In the embodiment, as shown in FIG. 4, step S4 specifically includes the following steps: Figure 5

[0076] S41, solving the multi-order natural frequency and mode shape of the numerical control machine tool spindle under clamping constraint by finite element modal analysis to obtain spindle vibration characteristic data.

[0077] S42, obtaining the standard vibration load density spectrum according to the bearing model matching of the numerical control machine tool spindle, and calculating the spindle modal frequency response function H i (ω) based on the random vibration theory and the spindle vibration characteristic data.

[0078] S43, for the spindle modal frequency response function, solving the root mean square value σ i and the probability density distribution of each order vibration response by statistical energy method, and synthesizing the vibration response energy spectrum density in physical space.​​

[0079] S44, integrate the vibration response energy spectrum density in the critical speed frequency band of the main shaft, combine the modal superposition principle and Hooke's law, and quantitatively calculate the vibration-induced deformation data.

[0080] S45, separate the vibration-induced deformation data from the second thermal-induced deformation data by using the least square fitting or state space filtering algorithm, and obtain the pure thermal-induced deformation data after removing the vibration interference.

[0081] S5, according to the multi-node temperature data of the equipment component under different working conditions and the corresponding pure thermal-induced deformation data, parallel inference is performed through the mean-variance parallel computational graph network model to obtain the mean and standard deviation of the thermal-induced deformation of the equipment component, and a thermal-induced deformation prediction interval with confidence is generated.

[0082] In the embodiment, the mean-variance parallel computational graph network model uses a multi-layer gated recurrent module to mine the time sequence evolution law of temperature and thermal-induced deformation as a time sequence feature, extracts the topological connection information between nodes of the equipment component through an attention network as a spatial feature, fuses the time sequence feature and the spatial feature to perform parallel inference on the mean and standard deviation of the thermal-induced deformation, and generates a thermal-induced deformation prediction interval with confidence. Specifically, as shown in Figure 6 , the mean-variance parallel computational graph network model includes: a multi-layer gated recurrent unit structure, a graph attention network, a cross-channel attention mechanism, a mean channel, a fully connected layer, a variance channel, and a nonlinear activation function.

[0083] In an exemplary embodiment, as shown in Figure 7 , step S5 specifically includes the following steps:

[0084] S51, extract the time sequence feature of the temperature sequence of each measuring point of the main shaft in the sliding time window through the multi-layer gated recurrent unit structure; the time sequence feature represents the time sequence evolution law of temperature and thermal-induced deformation.

[0085] S52, construct a graph attention network based on the thermal conduction topological relationship of the main shaft component, and extract the spatial feature of each measuring point on the surface of the main shaft; the nodes in the graph attention network match the number of temperature measuring points in dimension, and the edge weights are inversely proportional to the length of the heat flow path. The spatial feature represents the thermal conduction characteristics between the main shaft, bearings and other components.

[0086] S53, fuse the time sequence feature and the spatial feature through the cross-channel attention mechanism to obtain the spatio-temporal fusion feature.

[0087] S54, input the spatio-temporal fusion feature into the mean channel to map the output trend thermal deformation through the fully connected layer; the trend thermal deformation represents the steady-state component of the axial thermal elongation of the main shaft.

[0088] S55, inputting the spatiotemporal fusion feature into a variance channel to output a residual deformation amount via a nonlinear activation function; the residual deformation amount represents uncertainty caused by fluctuations in the thermal conductivity of the material and boundary disturbances during cooling.

[0089] S56, taking the trend item thermal deformation amount as a reference deformation amount μ t , and taking the residual deformation amount as an uncertainty deformation component A thermal deformation prediction interval with a confidence level is derived through Gaussian probability integration. Specifically, based on the assumption that the residual deformation amount obeys a Gaussian distribution , the Gaussian probability integration is as follows:

[0090] P(μ t -z 0.025 σ t ≤δ≤μ t +z 0.025 σ t )=95%.

[0091] wherein P() is a probability, μ t is the trend item thermal deformation amount, σ t is a standard deviation of the residual deformation amount, z 0.025 is a quantile of a standard normal distribution, and δ is an actual thermal deformation amount of the equipment; the thermal deformation prediction interval with a 95% confidence level is [μ t -z 0.025 σ t , μ t +z 0.025 σ t ].

[0092] The equipment thermal deformation interval parallel reasoning method based on uncertainty quantification decomposition provided in this embodiment gradually purifies the thermal deformation data through progressive quantification and decomposition of “noise suppression, morphology separation, and vibration stripping”, ensures that the input model is a pure signal that is strongly associated with temperature, eliminates interference from the source to mislead the model reasoning, improves the accuracy of thermal deformation calculation, and provides high-fidelity data support for thermal deformation reasoning. In step S5, based on the purified data, “spatiotemporal feature double-dimensional extraction + mean-variance parallel reasoning” is used, which accurately learns the deterministic law (mean) of temperature-thermal deformation and quantitatively analyzes the uncertainty source (standard deviation), and finally outputs a prediction interval that has both “trend accuracy” and “risk coverage”. This design forms a closed loop between “quantification and decomposition” and “model reasoning”: the former provides high-fidelity input for the latter, and the latter provides engineering verification (whether the interval reasonably reflects the true deformation range) for the purification effect of the former, and they jointly support the accurate control of the thermal deformation of the equipment.

[0093] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, it should be understood that the application encompasses all possible combinations of the technical features described above.

[0094] The principles and implementation manners of the present application are described herein by using specific examples, and the above embodiments are only used to help understand the method of the present application and its core idea; meanwhile, according to the idea of the present application, the specific implementation manners and application scopes will be changed by those skilled in the art. In conclusion, the content of the present specification should not be understood as a limitation of the present application.

Claims

1. An apparatus thermal induced deformation interval parallel inference method based on uncertainty quantization decomposition, characterized in that, The method comprises the following steps: Obtain multi-node temperature data and measured thermal deformation data of the equipment component under different working conditions, and establish a thermal characteristic test data set under different working conditions; Adopt a Splinelet wavelet basis to perform multi-band decomposition on the measured thermal deformation data, and adaptively adjust the boundary threshold of each frequency band through the local signal-to-noise ratio calculation result to obtain first thermal deformation data after removing high-frequency noise; Through a threshold-adaptive band-pass filtering method, identify and remove the surface topography in the first thermal deformation data associated with the operating characteristic frequency of the equipment component to obtain second thermal deformation data after removing surface topography interference; Solve the vibration-induced deformation data through a finite element modal analysis method and an energy spectral density analysis method, and subtract the vibration-induced deformation data from the second thermal deformation data to obtain pure thermal deformation data after removing vibration interference; According to the multi-node temperature data and the corresponding pure thermal deformation data of the equipment component under different working conditions, perform parallel inference through a mean-variance parallel computation graph network model to obtain the mean and standard deviation of the thermal deformation of the equipment component, and generate a thermal deformation prediction interval with confidence; the mean-variance parallel computation graph network model uses a multi-layer gated recurrent module to mine the time sequence evolution law of temperature and thermal deformation as a time sequence feature, extracts the topological connection information between nodes of the equipment component through an attention network as a spatial feature, fuses the time sequence feature and the spatial feature to perform parallel inference on the mean and standard deviation of the thermal deformation, and generates a thermal deformation prediction interval with confidence.

2. The method of claim 1, wherein, The different working conditions of the equipment component specifically represent different combinations of key process parameters and environmental conditions; the key process parameters include gas turbine pressure ratio, nuclear power main steam pressure and machine tool spindle speed; the environmental conditions include environmental temperature and cooling state.

3. The method of claim 1, wherein the method is performed by a system comprising one or more computers. Obtain multi-node temperature data and measured thermal deformation data of the equipment component under different working conditions, and establish a thermal characteristic test data set under different working conditions, specifically comprising: Build a thermal characteristic test bench around the equipment component, uniformly arrange multiple temperature sensors, and direct the eddy current displacement sensor to the sensitive direction of the thermal deformation of the equipment component; Change the working condition of the equipment component, collect temperature data at multiple nodes of the equipment component during operation, and obtain multi-node temperature data and measured thermal deformation data of the equipment component under different working conditions; Bind the multi-node temperature data and the measured thermal deformation data obtained under the same working condition with the corresponding working condition, and establish a thermal characteristic test data set under different working conditions.

4. The method of Claim 1, wherein, Adopt a Splinelet wavelet basis to perform multi-band decomposition on the measured thermal deformation data, and adaptively adjust the boundary threshold of each frequency band through the local signal-to-noise ratio calculation result to obtain first thermal deformation data after removing high-frequency noise, comprising: Before wavelet decomposition, symmetric extension preprocessing is adopted for the boundary, and the extension length is exponentially positively correlated with the decomposition level to ensure the calculation stability of the wavelet decomposition at the endpoints; According to the dynamic relationship between the sampling frequency of the measured thermal deformation data and the spindle speed and the feed speed, the optimal wavelet decomposition level is determined by logarithmic operation, and the measured thermal deformation data is decomposed into high-frequency detail coefficients and low-frequency approximation coefficients; The shrinkage threshold is dynamically generated based on the statistical standard deviation of the high-frequency detail coefficients and the low-frequency approximation coefficients, the high-frequency detail coefficients dominated by noise are set to zero by soft thresholding, while the low-frequency approximation coefficients are retained, and the key thermal deformation features are protected through energy error constraint and frequency band fidelity mechanism; Based on the denoised or retained wavelet coefficients at each level, the wavelet inverse transform reconstruction is performed based on the overlap reservation method, the transition is smoothed through the overlapping area to eliminate the truncation effect, the continuity of the thermal deformation time domain waveform is guaranteed, and the multi-scale decomposition signal is reconstructed into the first thermal deformation data after removing the high-frequency noise.

5. The method of Claim 1, wherein, Through the threshold adaptive band-pass filtering method, the surface topography associated with the operating frequency of the equipment component is identified and removed from the first thermal deformation data, obtaining the second thermal deformation data after removing the surface topography interference, including: Performing windowed Fourier transform on the spindle thermal deformation data, dynamically constructing the passband based on the spindle speed base frequency and harmonics of the equipment component; Extracting the frequency spectrum components within the passband through an elliptical filter, and removing the frequency spectrum components with stable amplitude within the continuous machining period as the surface topography error caused by clamping eccentricity or measurement position eccentricity; Performing zero-phase boundary continuation and inverse Fourier transform on the remaining frequency spectrum after separating the surface topography error, and reconstructing the second thermal deformation data after removing the surface topography interference.

6. The method of Claim 5, wherein, Before performing zero-phase boundary continuation and inverse Fourier transform on the remaining frequency spectrum after separating the surface topography error, and reconstructing the second thermal deformation data after removing the surface topography interference, it also includes: Arranging an eddy current displacement sensor array equidistantly along the surface of the spindle in the offline stage, and collecting the measured surface topography of the spindle surface; Verify the consistency of the separated surface topography and the measured surface topography through the Pearson correlation coefficient, if the consistency is lower than the consistency threshold, update the passband width and order of the elliptical filter, and jump to "extracting the frequency spectrum components within the passband through an elliptical filter, and removing the frequency spectrum components with stable amplitude within the continuous machining period as the surface topography error caused by clamping eccentricity".

7. The method of Claim 1, wherein, Solving the vibration-induced deformation data by finite element modal analysis method and energy spectrum density analysis method, and subtracting the vibration-induced deformation data from the second thermal deformation data to obtain the pure thermal deformation data after removing the vibration interference, including: Solving the multi-order natural frequency and mode shape of the spindle under clamping constraint through finite element modal analysis to obtain the spindle vibration characteristic data; According to the bearing model of the numerical control machine tool spindle, the standard vibration load density spectrum is matched, and the modal frequency response function of the spindle at each order is calculated based on the random vibration theory and the spindle vibration characteristic data; For the modal frequency response function of the spindle at each order, the root mean square value and the probability density distribution of each order vibration response are solved by statistical energy method, and the vibration response energy spectrum density in physical space is synthesized; The vibration response energy spectrum density is integrated in a critical speed frequency band of a main shaft, and vibration-induced deformation data is quantitatively calculated by combining a modal superposition principle and Hooke's law; The vibration-induced deformation data is separated from the second thermal-induced deformation data by using a least square fitting or a state space filtering algorithm, so that pure thermal-induced deformation data free from vibration interference is obtained.

8. The method of Claim 1, wherein, The mean-variance parallel computational graph network model comprises a multi-layer gated recurrent unit structure, a graph attention network, a cross-channel attention mechanism, a mean channel, a fully connected layer, a variance channel and a nonlinear activation function.

9. The equipment thermal deformation zone interval parallel inference method based on uncertainty quantization decomposition according to claim 8, characterized in that, According to the multi-node temperature data of the equipment component under different working conditions and the corresponding pure thermal-induced deformation data, a mean-variance parallel computational graph network model is used for parallel inference to obtain the mean and standard deviation of the thermal-induced deformation of the equipment component, and a thermal-induced deformation prediction interval with a confidence level is generated, specifically including: Through the multi-layer gated recurrent unit structure, time sequence features of temperature sequences of each measuring point of the main shaft in a sliding time window are extracted; the time sequence features represent the time sequence evolution law of temperature and thermal-induced deformation; A graph attention network is constructed based on the heat conduction topological relationship of the main shaft component to extract spatial features of each measuring point on the surface of the main shaft; the nodes in the graph attention network represent the dimension matching the number of temperature measuring points, and the edge weights are inversely proportional to the heat flow path length; The time sequence features and the spatial features are fused through the cross-channel attention mechanism to obtain spatio-temporal fusion features; The spatio-temporal fusion features are input into the mean channel to map the trend thermal deformation amount through the fully connected layer; the trend thermal deformation amount represents the steady-state component of the axial thermal elongation of the main shaft; The spatio-temporal fusion features are input into the variance channel to output the residual deformation amount through the nonlinear activation function; the residual deformation amount represents the uncertainty caused by the fluctuation of the material thermal conductivity coefficient and the cooling boundary disturbance; The trend thermal deformation amount is taken as the reference deformation amount, and the residual deformation amount is taken as the uncertainty deformation component, and a thermal-induced deformation prediction interval with a confidence level is derived through Gaussian probability integration.

10. The equipment thermal deformation zone interval parallel inference method based on uncertainty quantization decomposition of claim 9, wherein, The Gaussian probability integration is as follows: P(μ t - z 0.025 σ t ≤ δ ≤ μ t + z 0.025 σ t ) = 95% where P() is the probability, μ t is the trended thermal distortion, σ t is the standard deviation of the residual distortion, z 0.025 is the quantile of the standard normal distribution, and δ is the actual equipment thermal induced distortion; then the 95% confidence interval of the thermal induced distortion is [μ t - z 0.025 σ t , μ t + z 0.025 σ t ].

Images