Method and system for monitoring vibration of a radioisotope separation device and evaluating mechanical state thereof

Abstract

The application provides a radioactive separation device vibration monitoring and mechanical state evaluation method and system, relates to the technical field of mechanical vibration fault diagnosis, and the method comprises the following steps: step 1, adaptive noise reduction processing is carried out on original vibration data, noise reduction parameters are dynamically adjusted according to the statistical characteristics of radiation interference and flow-induced turbulence interference, and early fault characteristic vibration signals are extracted; step 2, the early fault characteristic vibration signals are compared with reference vibration characteristics in a healthy state, and deep feature indexes representing the mechanical degradation state are extracted; and step 3, the spatial coordinates of the base and the rack connecting bolt, the bearing locking nut positioning surface and the flange butt ring joint are taken as constraint nodes to construct the structure dynamics envelope surface of the entity part. The application realizes accurate extraction of early faults, multi-fault decoupling identification, accurate positioning and dynamic prediction of the remaining life in a strong interference environment.

Description

Technical Field

[0001] This invention relates to the field of mechanical vibration fault diagnosis technology, and in particular to a method and system for vibration monitoring and mechanical condition assessment of radioactive separation devices. Background Technology

[0002] Vibration monitoring is a crucial means of ensuring the safe operation of radioactive separation devices. Taking a multi-stage centrifugal separation system in a nuclear fuel reprocessing plant as an example, its core equipment includes a high-speed rotor, a multi-stage series frame, and a sealed cavity. This system operates under complex conditions of strong radiation and flow-induced turbulence. When the system is running, the bolts connecting the base and the frame bear alternating loads, the rotor shaft end bearing locking nut positioning surface generates high-frequency vibration due to high-speed rotation, and the circumferential seam where the final centrifugal separation cavity connects to the pipeline flange is affected by fluid impact. The vibration response of these three parts directly reflects the mechanical health status of the entire machine. Existing vibration monitoring methods typically employ a single threshold alarm strategy to identify vibration anomalies with amplitude exceeding limits. This approach cannot effectively separate early, weak fault characteristics submerged in radiation noise and flow-induced interference. Furthermore, because the spatial correlation of the vibration responses of the three key parts is not considered, it is difficult to accurately locate and quantify multi-source coupled faults such as rotor imbalance, bearing failure, seal wear, and flow-induced vibration, resulting in the inability to provide timely warnings in the early stages of fault evolution. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method and system for vibration monitoring and mechanical condition assessment of radioactive separation devices, so as to achieve accurate early fault extraction, multi-fault decoupling identification, precise location and dynamic prediction of remaining life under strong interference environment.

[0004] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0005] Firstly, a method for vibration monitoring and mechanical condition assessment of a radioactive separation device, the method comprising:

[0006] Step 1: Adaptive noise reduction is performed on the original vibration data. The noise reduction parameters are dynamically adjusted according to the statistical characteristics of radiation interference and flow-induced turbulence interference, and vibration signals with early fault characteristics are extracted.

[0007] Step 2: Compare the vibration signals of early fault characteristics with the baseline vibration characteristics under healthy conditions to extract deep feature indicators that characterize the mechanical deterioration state.

[0008] Step 3: Using the spatial coordinates of the bolts connecting the base and the frame, the positioning surface of the bearing locking nut, and the flange butt joint as constraint nodes, construct the structural dynamics envelope of the solid part; perform finite element meshing on the envelope and conduct transient response numerical deduction to obtain the modal contribution distribution on the vibration transmission path under different fault modes; perform discrete point curvature estimation on the mesh nodes of the meshed envelope to obtain the curvature characteristic value at each node, and perform spatial weighted correction on the modal contribution distribution based on the curvature characteristic value to obtain the corrected modal contribution distribution;

[0009] Step 4: Based on the corrected modal contribution distribution, the deep feature indexes are weighted and assigned. The fault features corresponding to rotor imbalance, bearing failure, seal wear and flow-induced vibration are separated by pattern recognition. The severity of each type of fault and the location of the fault are quantified respectively.

[0010] Step 5: Based on the severity of various faults, combined with historical degradation data and current operating stress, dynamically update the prediction parameters through degradation trend modeling to obtain the remaining life and degradation trend prediction results of key components.

[0011] Secondly, the vibration monitoring and mechanical condition assessment system for radioactive separation devices includes:

[0012] The extraction module is used to perform adaptive noise reduction on the raw vibration data. It dynamically adjusts the noise reduction parameters based on the statistical characteristics of radiation interference and flow-induced turbulence interference and extracts vibration signals with early fault characteristics.

[0013] The comparison module is used to compare the vibration signals of early fault characteristics with the baseline vibration characteristics in the healthy state, and extract the deep feature indicators that characterize the mechanical deterioration state.

[0014] The calculation module is used to construct the structural dynamics envelope of the solid parts using the spatial coordinates of the bolts connecting the base and the frame, the positioning surface of the bearing locking nut, and the flange butt joint as constraint nodes; it performs finite element meshing on the envelope and conducts transient response numerical derivation to obtain the modal contribution distribution on the vibration transmission path under different fault modes; it performs discrete point curvature estimation on the mesh nodes of the meshed envelope to obtain the curvature characteristic value at each node, and performs spatial weighted correction on the modal contribution distribution based on the curvature characteristic value to obtain the corrected modal contribution distribution;

[0015] The allocation module is used to assign weights to deep feature indicators based on the corrected modal contribution distribution. It separates the fault features corresponding to rotor imbalance, bearing failure, seal wear and flow-induced vibration through pattern recognition, and quantifies the severity of each type of fault and locates the fault location.

[0016] The update module is used to dynamically update the prediction parameters based on the severity of various faults, combined with historical degradation data and current operating stress, through degradation trend modeling, to obtain the remaining life and degradation trend prediction results of key components.

[0017] Thirdly, a computing device, comprising:

[0018] One or more processors;

[0019] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0020] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0021] The above-described solution of the present invention has at least the following beneficial effects:

[0022] This invention employs an adaptive noise reduction method based on the statistical characteristics of radiation interference and flow-induced turbulence interference in the raw vibration data, extracts early fault characteristic vibration signals, compares these signals with healthy baseline vibration characteristics to extract deep degradation characteristic indicators, constructs a structural dynamic envelope surface using the spatial coordinates of three key locations (such as the bolts connecting the base and the frame) as constraint nodes, and obtains a precise modal contribution distribution through finite element mesh generation, transient response deduction, and discrete point curvature weighting correction. Based on this distribution, weights are assigned to the deep characteristic indicators, and multiple types of faults are separated through pattern recognition, with quantification of severity and fault location. Finally, it combines historical degradation data with current operating stress and dynamically updates prediction parameters through degradation trend modeling. This approach overcomes the technical problems of existing vibration monitoring methods, such as the inability to effectively separate early weak fault characteristics in strong radiation and flow-induced interference, the lack of consideration for the spatial correlation of vibrations in the three key locations leading to difficulty in accurately locating and quantifying multi-source coupled faults, and inaccurate prediction of remaining life. Ultimately, it achieves the technical effect of accurately extracting early faults, decoupling and identifying multiple faults, accurately locating them, and dynamically predicting the remaining life and degradation trend of key components in a strong interference environment for radioactive separation devices, providing reliable support for the safe operation of equipment. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the vibration monitoring and mechanical condition assessment method for a radioactive separation device provided in an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of a vibration monitoring and mechanical condition assessment system for a radioactive separation device provided in an embodiment of the present invention. Detailed Implementation

[0025] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art.

[0026] like Figure 1 As shown, embodiments of the present invention propose a method for vibration monitoring and mechanical condition assessment of a radioactive separation device, the method comprising the following steps:

[0027] Step 1: Adaptive noise reduction is performed on the original vibration data. The noise reduction parameters are dynamically adjusted according to the statistical characteristics of radiation interference and flow-induced turbulence interference, and vibration signals with early fault characteristics are extracted.

[0028] Step 2: Compare the vibration signals of early fault characteristics with the baseline vibration characteristics under healthy conditions to extract deep feature indicators that characterize the mechanical deterioration state.

[0029] Step 3: Using the spatial coordinates of the bolts connecting the base and the frame, the positioning surface of the bearing locking nut, and the flange butt joint as constraint nodes, construct the structural dynamics envelope of the solid part; perform finite element meshing on the envelope and conduct transient response numerical deduction to obtain the modal contribution distribution on the vibration transmission path under different fault modes; perform discrete point curvature estimation on the mesh nodes of the meshed envelope to obtain the curvature characteristic value at each node, and perform spatial weighted correction on the modal contribution distribution based on the curvature characteristic value to obtain the corrected modal contribution distribution;

[0030] Step 4: Based on the corrected modal contribution distribution, the deep feature indexes are weighted and assigned. The fault features corresponding to rotor imbalance, bearing failure, seal wear and flow-induced vibration are separated by pattern recognition. The severity of each type of fault and the location of the fault are quantified respectively.

[0031] Step 5: Based on the severity of various faults, combined with historical degradation data and current operating stress, dynamically update the prediction parameters through degradation trend modeling to obtain the remaining life and degradation trend prediction results of key components.

[0032] In this embodiment of the invention, adaptive noise reduction processing is performed on the original vibration data, and the noise reduction parameters are dynamically adjusted according to the statistical characteristics of radiation interference and flow-induced turbulence interference to extract early fault characteristic vibration signals. These early fault characteristic vibration signals are then compared with the baseline vibration characteristics under healthy conditions to extract deep characteristic indicators representing mechanical degradation. The spatial coordinates of the base-frame connection bolts, bearing locking nut positioning surfaces, and flange butt joint circumferential seams are used as constraint nodes to construct the structural dynamic envelope of the solid parts. After finite element mesh generation, transient response numerical derivation, and discrete point curvature weighting correction, a corrected modal contribution distribution is obtained. Based on this corrected modal contribution distribution, the deep characteristic indicators are... By employing weight allocation and pattern recognition to separate various fault characteristics, quantify fault severity, and locate fault locations, and by combining historical degradation data and current operating stress to dynamically update prediction parameters through degradation trend modeling, this technology overcomes the technical problems of existing technologies, such as difficulty in extracting early weak fault characteristics from strong radiation and flow-induced turbulence interference, inability to accurately decouple multiple types of coupled faults and locate fault locations, and inaccurate prediction of remaining life and degradation trends. This enables the accurate extraction of early fault characteristics of radioactive separation devices under strong interference conditions, the decoupling and precise location of multiple types of faults such as rotor imbalance and bearing failure, the quantification of the severity of various faults, and the dynamic prediction of the remaining life and degradation trends of key components.

[0033] In a preferred embodiment of the present invention, the method further includes the following step before step 1:

[0034] Step 01: Using radiation-resistant, self-powered wireless vibration sensors located at the connection bolts between the base and the multi-stage series frame of the radioactive separation device, the positioning surface of the rotor shaft bearing lock nut, and the circumferential joint between the final centrifugal separation chamber and the pipeline flange, raw vibration data from multiple measurement points is simultaneously collected. Specifically, this includes: spatially calibrating the core vibration monitoring points of the radioactive separation device, using the reference installation plane of the device base as the spatial reference, and calibrating the three points: the connection bolts between the base and the multi-stage series frame, the positioning surface of the rotor shaft bearing lock nut, and the circumferential joint between the final centrifugal separation chamber and the pipeline flange. Three-dimensional position measurements were performed on key components, sequentially determining the horizontal, vertical, and lateral coordinates of each component. The three coordinates of each component were summed to obtain the spatial positioning characteristic value of that monitoring component. Based on this characteristic value, the installation position of each sensor was determined, with the spatial deviation of the sensor installation position controlled within 0.5 mm to ensure complete matching between the sensor placement and the core vibration response components reflecting the overall health status of the equipment. After position calibration, radiation-resistant self-powered wireless vibration sensors were installed at the three calibrated key monitoring locations. The vibration sensing end face of the sensor was tightly fitted to the corresponding solid structure surface, and assembly was completed with a tightening torque of 8 N / m, eliminating installation gaps between the sensor and the equipment structure. This ensured lossless transmission of vibration signals from the equipment to the sensor sensing end face. One radiation-resistant self-powered wireless vibration sensor was deployed at each key monitoring location, forming a multi-point acquisition layout covering the core vibration components of the equipment. The sensor operates independently using its own power supply module, requiring no external power supply or signal wiring, making it suitable for use in harsh, radioactive, and enclosed environments. To achieve synchronous acquisition of data from multiple measurement points, a [further details needed]. A unified reference acquisition clock with an accuracy of 1 microsecond is used to read the local clock value of each radiation-resistant self-powered wireless vibration sensor. The difference between the local clock value of each sensor and the reference acquisition clock value is calculated to obtain the clock deviation value of a single sensor. The clock deviation value is then multiplied by a timing correction coefficient of 1.0 to calculate the timing compensation value of the corresponding sensor. The timing compensation value is then added to the acquisition trigger timing of the sensor to complete the synchronous timing calibration of the three sensors. After calibration, the timing synchronization error between the sensors is no more than 2 microseconds, eliminating the timing error between different sensors.

[0035] After timing calibration, uniform sampling parameters were set according to the actual operating conditions of the radioactive separation device. The vibration characteristic frequency corresponding to the highest operating speed of the equipment was 500 Hz, and the sampling frequency was set to 500 Hz multiplied by 10, i.e., the sampling frequency was directly set to 5000 Hz to ensure complete acquisition of vibration signal information across the entire frequency band. Using the calibrated synchronization timing as a unified trigger signal, three radiation-resistant self-powered wireless vibration sensors simultaneously started acquisition, collecting raw data on vibration acceleration, vibration velocity, and vibration displacement of the corresponding monitoring parts in real time. The continuous acquisition time for a single data segment was set to 1 second, and the power supply to the sensors was kept stable during the acquisition process to continuously acquire uninterrupted data. The vibration data was then collected. Finally, the raw vibration data from the three monitoring locations acquired by the three sensors at the same acquisition time were integrated. Following the fixed sequence of the connection bolts between the base and the multi-stage series frame, the positioning surface of the rotor shaft bearing lock nut, and the circumferential seam between the final centrifugal separation chamber and the pipeline flange, the multi-point data of 1024 sampling points at a single moment were combined into a complete raw vibration data segment. All continuously acquired raw vibration data segments were output sequentially and cached in a 1024-megabyte cache space, forming synchronous, complete, and comprehensive multi-point raw vibration data covering the key parts of the equipment. This provides a real and reliable data foundation for the subsequent adaptive noise reduction processing of the raw vibration data.

[0036] In this embodiment of the invention, by employing a technique of arranging radiation-resistant, self-powered wireless vibration sensors at the connection bolts between the base of the radioactive separation device and the multi-stage series frame, the positioning surface of the rotor shaft end bearing locking nut, and the circumferential seam between the final centrifugal separation chamber and the pipeline flange, and simultaneously collecting raw vibration data from multiple measurement points, the technical problems of difficult wiring and maintenance of wired sensors in a closed environment with strong radioactive radiation, easy failure of ordinary sensors due to radiation interference, and inability of a single measurement point to fully reflect the vibration state of key parts of the equipment are overcome.

[0037] In a preferred embodiment of the present invention, step 1 above may include:

[0038] Step 1.1: Perform time-domain windowing segmentation on the raw vibration data. Perform power spectral density analysis on the signal within each segment to extract the peak pulse frequency band corresponding to radiated interference and the broadband random disturbance frequency band corresponding to flow-induced turbulence interference. Specifically, this includes: performing time-domain windowing segmentation on the acquired multi-point raw vibration data. The sampling frequency of the raw vibration data is fixed at 5000 Hz, and the continuous acquisition time for a single data segment is 1 second. Each second of raw vibration data contains 5000 sampling points. To ensure the continuity of the signal after segmentation and to avoid losing key vibration features, the length of each time-domain window is set to 1024 sampling points, and the corresponding window duration is 1024 sampling points divided by 5000 Hz. The calculated duration of a single window segment is 0.2048 seconds. The window shift between two adjacent windows is set to 512 sampling points, corresponding to a window shift duration of 512 sampling points divided by 5000 Hz, resulting in a window shift duration of 0.1024 seconds. This setting ensures a 50% overlap between adjacent window segments, preventing signal breakage during segmentation and ensuring that each segment retains the complete local characteristics of the original vibration data, while achieving full coverage of the original vibration data. After time-domain windowing segmentation, power spectral density analysis is performed on the original vibration data within each window segment. The periodogram method is used to calculate the power spectral density of each signal segment. The specific calculation process involves... The 1024 sampled points within a single window undergo a Fast Fourier Transform (FFT) to obtain the frequency domain amplitude spectrum. The square of each amplitude point in the frequency domain spectrum is then divided by the frequency resolution of that segment to obtain the power spectral density (PSD) value for each frequency point. The frequency resolution is calculated by dividing the sampling frequency by the window length (e.g., 5000 Hz divided by 1024 sample points), resulting in a resolution of approximately 4.8828 Hz. This ensures that the PSD analysis accurately captures the signal power distribution characteristics at different frequencies. By analyzing the peak characteristics and distribution range of the PSD curve, two types of environmental interference frequency bands are extracted: radiated interference exhibits spike pulse characteristics, and... The power spectral density curve shows obvious sharp peaks in local frequency ranges, and the peak duration frequency range is relatively narrow. Based on this, the peak pulse frequency band corresponding to radiated interference is extracted to be 1500 Hz to 2500 Hz. Flow-induced turbulence interference has broadband random characteristics. On the power spectral density curve, it shows uniform power distribution within the frequency range, without obvious sharp peaks, and covers a wide frequency range. Based on this, the broadband random disturbance frequency band corresponding to flow-induced turbulence interference is extracted to be 0 Hz to 1000 Hz. To ensure the accuracy of interference frequency band extraction, the power spectral density analysis results of all segments are averaged. The power spectral density curve after averaging is used as the final analysis basis to complete the accurate division of the two types of environmental interference frequency bands.

[0039] Step 1.2: Calculate the signal-to-noise ratio (SNR) under the current environment based on the distribution characteristics of the broadband random disturbance frequency band corresponding to the peak pulse frequency band and the flow-induced turbulence interference. Dynamically adjust the number of wavelet packet decomposition levels and threshold coefficients based on the SNR. Perform wavelet packet decomposition on the original vibration data. Use the adjusted threshold coefficients to perform soft threshold quantization on the coefficients of each frequency band. Reconstruct the thresholded coefficients to obtain the denoised vibration signal. Specifically, this includes: based on the extracted power spectral distribution results of the radiated interference peak pulse frequency band and the flow-induced turbulence interference broadband random disturbance frequency band, first calculate the cumulative power value within the corresponding frequency band. The radiated interference power is obtained by cumulatively adding the power spectral density values ​​of all frequency points within the peak pulse frequency band, and the flow-induced turbulence interference power is obtained by cumulatively adding the power spectral density values ​​of all frequency points within the broadband random disturbance frequency band. The total signal power is obtained by accumulating the power spectral density values ​​across the entire frequency band from 0 Hz to 2500 Hz. The effective signal power is the total signal power minus the sum of the radiated interference power and the flow-induced turbulence interference power. Based on this, the signal-to-noise ratio (SNR) under the current monitoring environment is calculated. The SNR is equal to 10 multiplied by the logarithm of the ratio of the effective signal power to the total interference power (base 10). The total interference power is the sum of the radiated interference power and the flow-induced turbulence interference power. The final SNR is expressed in decibels (dB). The number of wavelet packet decomposition layers is dynamically determined based on the calculated SNR value. When the SNR is greater than or equal to 5 dB, the number of wavelet packet decomposition layers is set to three; when the SNR is greater than or equal to 3 dB and less than 5 dB, the number of wavelet packet decomposition layers is set to four; and when the SNR is less than 3 dB, the number of wavelet packet decomposition layers is set to five. Simultaneously, the threshold coefficient is calculated according to a fixed formula. The threshold coefficient is equal to the difference between the basic threshold coefficient of 0.2 plus 5 and the current signal-to-noise ratio value, multiplied by 0.05. This completes the adaptive determination of the wavelet packet decomposition level and the threshold coefficient. The original vibration data is decomposed into wavelet packet coefficients corresponding to different frequency bands layer by layer according to the determined decomposition level. Then, the calculated threshold coefficients are used to perform soft threshold quantization on the wavelet packet coefficients of each frequency band. If the absolute value of the wavelet packet coefficient is greater than the threshold coefficient, the value of the coefficient is subtracted from the threshold coefficient and retained. If the absolute value of the wavelet packet coefficient is less than or equal to the threshold coefficient, the coefficient is directly set to zero. After completing the soft thresholding of all frequency band coefficients, the processed wavelet packet coefficients are subjected to layer-by-layer inverse transformation and signal reconstruction according to the inverse process of wavelet packet decomposition. Finally, the noise-reduced vibration signal after removing radiation interference and flow-induced turbulence interference is obtained.

[0040] Step 1.3: Perform empirical mode decomposition on the denoised vibration signal to obtain several intrinsic mode function (IMF) components. IMF components whose correlation with the early fault characteristic frequency of the equipment exceeds a preset threshold are selected for signal reconstruction to extract the vibration signal with early fault characteristics. Specifically, this includes: performing empirical mode decomposition on the obtained denoised vibration signal, identifying all local maxima and minima in the denoised vibration signal, fitting the envelopes of the maxima and minima using cubic spline interpolation, calculating the local mean curves of the two envelopes, subtracting these local mean curves from the denoised vibration signal to obtain preliminary separated components, repeating the extreme value identification, envelope fitting, and mean removal processes until the preliminary separated components meet the IMF criteria to obtain first-order IMF components, subtracting the first-order IMF components from the original denoised vibration signal to obtain residual components, and repeating the above decomposition process on the residual components to obtain second-order, third-order, and even multi-order IMF components until the residual components become indistinct. The monotonic trend term with fluctuations is identified, and the decomposition process is stopped to obtain all intrinsic mode function (IMF) components. For three typical early faults in the radioactive separation device—rotor imbalance, bearing failure, and seal wear—corresponding early fault characteristic frequencies are pre-determined. The correlation coefficient between each IMF component and each early fault characteristic frequency is calculated. The correlation coefficient is equal to the covariance of the two signals divided by the product of the standard deviations of the first and second signals. The preset threshold for screening IMF components is set to 0.85. The correlation coefficient between each IMF component and the early fault characteristic frequency is judged one by one, retaining only IMF components with a correlation coefficient greater than 0.85 and discarding those with a correlation coefficient below the preset threshold. All the screened effective IMF components are linearly superimposed and reconstructed. The superposition process involves adding the signal amplitudes of each effective IMF component at the same time point by point. Finally, the reconstructed signal is the early fault characteristic vibration signal that retains only the early fault information of the equipment and completely eliminates redundant interference.

[0041] In this embodiment of the invention, by employing time-domain windowing and power spectral density analysis of the original vibration data to extract the peak pulse frequency band of radiation interference and the broadband random disturbance frequency band of flow-induced turbulence interference, dynamically adjusting the wavelet packet decomposition level and threshold coefficient according to the signal-to-noise ratio and performing soft threshold quantization reconstruction to achieve noise reduction, and then performing empirical mode decomposition of the denoised signal and screening highly correlated intrinsic mode function components to reconstruct the signal, the technical problems of strong radiation interference and flow-induced turbulence noise in radioactive environments, easy submersion of early weak fault features, and poor adaptability of traditional fixed parameter noise reduction methods are overcome. Thus, the invention achieves adaptive removal of complex environmental noise and accurate extraction of early fault feature vibration signals.

[0042] In a preferred embodiment of the present invention, step 2 above may include:

[0043] Step 2.1: Establish a reference vibration feature database for the equipment under healthy conditions, including time-domain waveforms, spectral structures, and shaft center trajectory features corresponding to each typical operating condition; perform operating condition identification on the vibration signals of the extracted early fault features, obtain the current speed and flow rate operating parameters, and retrieve the reference vibration features matching the current operating condition from the reference vibration feature database; calculate the waveform similarity between the vibration signals of the early fault features and the reference vibration features, and perform local feature scaling correction on frequency bands with waveform similarity below a preset threshold to eliminate the influence of speed fluctuations on feature comparison; map the vibration signals after operating condition adaptive correction to a high-dimensional phase space to obtain the phase point trajectory in the high-dimensional phase space. Specifically, this includes: establishing a reference vibration feature database for the radioactive separation device under healthy conditions, covering three typical operating conditions of the equipment: low load operating condition with a speed of 1500 rpm and a flow rate of 5 cubic meters per hour; rated load operating condition with a speed of 2000 rpm and a flow rate of 8 cubic meters per hour; and high load operating condition with a speed of 2500 rpm. With a flow rate of 10 cubic meters per hour, vibration data were collected from three key locations under each operating condition: the connection bolts between the base and the frame, the positioning surface of the rotor shaft bearing lock nut, and the circumferential seam between the final-stage centrifugal separation chamber and the pipeline flange. Data was continuously collected for 72 hours, with one set of feature data extracted every hour. Each set of data included time-domain waveforms, spectral structure, and shaft trajectory features. The time-domain waveform features included peak, effective, and peak factor of the vibration acceleration; the spectral structure features included harmonic frequencies and their corresponding amplitudes; and the shaft trajectory features included trajectory radius and ellipticity. All feature data were categorized and stored according to operating conditions to form a complete benchmark vibration feature database, ensuring that the database covered the entire operating range of the equipment's normal operation. The extracted early fault characteristic vibration signals were used for operating condition identification. By analyzing the spectral structure of the signal, the fundamental frequency corresponding to the equipment's rotational speed was extracted from the spectrum. The current equipment rotational speed was calculated based on the fundamental frequency, which is equal to the fundamental frequency multiplied by 60. For example, if the extracted fundamental frequency is 33.3 Hz, then the current rotational speed is 33 Hz.Multiplying 3 Hz by 60 yields a current rotational speed of 2000 rpm. Simultaneously, the flow sensor on the radioactive separation device reads the actual operating flow rate of 8 cubic meters per hour. Combining the current rotational speed and flow rate, the reference vibration characteristics corresponding to the rated load operating conditions of 2000 rpm and 8 cubic meters per hour are retrieved from the reference vibration characteristic database, ensuring a perfect match between the retrieved reference characteristics and the current operating conditions. The waveform similarity between the early fault characteristic vibration signal and the retrieved reference vibration characteristics is calculated using the correlation coefficient method. Specifically, both the early fault characteristic vibration signal and the reference vibration characteristics have a total of 1024 sampling points. The covariance between the two signal sequences is calculated. The variance and covariance are calculated as follows: The covariance is equal to the reciprocal of the total number of sampling points, multiplied by the difference between the fault signal amplitude at each sampling point and the overall mean of the fault signal, then multiplied by the difference between the reference signal amplitude and the overall mean of the reference signal. The results for all sampling points are summed to obtain the covariance. Next, the standard deviation of the fault signal sequence and the standard deviation of the reference signal sequence are calculated separately. The standard deviation of the signal sequence is equal to the reciprocal of the total number of sampling points, multiplied by the square of the difference between the amplitude at each sampling point and the overall mean of the signal. The square root of all these squares is then taken to obtain the standard deviation of the corresponding signal. Finally, the waveform similarity is equal to the covariance divided by the product of the standard deviations of the fault signal and the reference signal. The calculated similarity value ranges from 0 to 1, with values ​​closer to 1 indicating greater similarity between the two waveforms.

[0044] A preset threshold for waveform similarity is set to 0.9. If the calculated waveform similarity is lower than 0.9, it indicates that the current vibration signal is affected by rotational speed fluctuations. Local feature scaling correction is required for frequency bands with similarity below the threshold. The frequency band with similarity below 0.9 is determined to be 100 Hz to 200 Hz. The average amplitude of the reference vibration feature within this frequency band is calculated, along with the average amplitude of the corresponding frequency band of the current early fault characteristic vibration signal. The correction coefficient is equal to the average amplitude of the reference frequency band divided by the average amplitude of the current frequency band. The average amplitude of the reference frequency band is 0.5 mm, and the average amplitude of the current frequency band is 0.4 mm; therefore, the correction coefficient is 0.5 mm divided by 0.4 mm, resulting in a correction coefficient of 1.25. The amplitude of all sampling points within this frequency band in the current early fault characteristic vibration signal is multiplied by the correction coefficient to complete the local feature scaling correction. This process is repeated until all frequency bands with similarity below 0.9 are corrected. All corrections were completed to eliminate the influence of speed fluctuations on feature comparison. After correction, the vibration signal after adaptive correction was mapped to a high-dimensional phase space. The phase space embedding dimension was selected as 5-dimensional, and the delay time was 10 milliseconds, with 10 milliseconds corresponding to 50 sampling points. The specific mapping process was as follows: the amplitudes of the corrected vibration signal were arranged sequentially. Starting from the first amplitude, the current amplitude, the amplitude at intervals of 50 sampling points, 100 sampling points, 150 sampling points, and 200 sampling points were selected as the five-dimensional coordinates of the first phase point. Starting from the second amplitude, five amplitudes were selected as the five-dimensional coordinates of the second phase point according to the same interval rule, and so on, until all amplitude sequences were traversed to form the phase point trajectory in the high-dimensional phase space. Each phase point contained five-dimensional coordinate information, fully preserving the nonlinear characteristics of the vibration signal.

[0045] Step 2.2 involves performing geometric feature analysis on the phase point trajectories. Instantaneous curvature sequences are obtained by calculating the spatial change rate of the direction connecting adjacent phase points. The mean and variance of these instantaneous curvature sequences are used as phase space geometric distortion features. Simultaneously, the cumulative arc length of the phase point trajectories per unit time is calculated as the phase space extension rate feature. These phase space geometric distortion features and phase space extension rate features, along with correlation dimension, Kolmogorov entropy, and Lyapunov exponent, constitute a deep feature index characterizing the mechanical degradation state. Specifically, this includes: performing geometric feature analysis on the phase point trajectories in the obtained high-dimensional phase space; extracting all adjacent phase points from the phase point trajectories; reading the five-dimensional coordinates of any given phase point and the five-dimensional coordinates of the next adjacent phase point; calculating the direction vector connecting two adjacent phase points; and calculating the direction vector for each direction vector. Each component is equal to the coordinates of the next phase point minus the coordinates of the previous phase point, yielding the complete direction vector. Then, the angle between two adjacent direction vectors is calculated. The cosine of the angle is equal to the dot product of the first and second direction vectors divided by the product of the magnitudes of the first and second direction vectors. The dot product is equal to the product of the corresponding dimension components of the two direction vectors and the sum of all the components. The magnitude is equal to the square root of the sum of the squares of all dimension components of the direction vector. The angle is calculated using the inverse cosine function. This angle is divided by the distance between two adjacent phase points to obtain the instantaneous curvature value. The distance between adjacent phase points is equal to the square root of the sum of the squares of the differences in the coordinates of the two phase points. The instantaneous curvature values ​​of all adjacent phase points are calculated sequentially to form an instantaneous curvature sequence.

[0046] Statistical analysis was performed on the instantaneous curvature sequence to calculate its mean and variance, which serve as characteristics of phase space geometric distortion. The mean was calculated by summing all curvature values ​​in the instantaneous curvature sequence and then dividing by the total number of curvature values. Since the instantaneous curvature sequence contains 1000 curvature values, the sum of all curvature values ​​is 500, so the mean is 500 divided by 1000, resulting in a mean of 0.5. The variance was calculated by squaring the difference between each curvature value and the mean, summing all the squares, and then dividing by the total number of curvature values. The sum of the squares of the differences between each curvature value and the mean of 0.5 is 250, so the variance is 250 divided by 1000, resulting in a variance of 0.25. Simultaneously, the cumulative arc length of the phase point trajectory per unit time was calculated. As a characteristic of phase space stretching rate: the distance between every two adjacent phase points is calculated, and the distances of all adjacent phase points are accumulated point by point to obtain the total arc length of the phase point trajectory; then the total arc length is divided by the acquisition time corresponding to the phase point trajectory. The acquisition time is equal to the number of sampling points contained in the phase point trajectory divided by the sampling frequency. If the phase point trajectory contains 1000 phase points and the sampling frequency is 5000 Hz, then the acquisition time is 1000 divided by 5000, which calculates to be 0.2 seconds. If the total arc length is 10 mm, then the cumulative arc length per unit time, i.e., the stretching rate, is 10 mm divided by 0.2 seconds, which calculates to be 50 mm / s; based on this, three nonlinear characteristic indices are calculated: correlation dimension, Kolmogorov entropy, and Lyapunov exponent. The correlation dimension is calculated using the Glasberg-Procacia algorithm, with embedding dimensions increasing sequentially from 1 to 10. The correlation integral is calculated for each embedding dimension. The slope of this linear relationship, where the logarithm of the correlation integral is linearly related to the logarithm of the phase distance, is the correlation dimension. The calculated correlation dimension is 2.3. Kolmogorov entropy is obtained by calculating the probability distribution of phase points in different regions of the high-dimensional phase space. The phase space is divided into 100 equally sized regions. The number of phase points in each region is counted, and the probability of each region is calculated. The probability equals the number of phase points in the region divided by the total number of phase points. The Kolmogorov entropy is equal to the product of the probabilities of all regions and the logarithm of that probability, then the negative number is calculated. The Kolmogorov entropy is 0.8; the Lyapunov index is obtained by calculating the rate of change of the distance between adjacent phase points over time. 100 groups of adjacent phase points are selected, and the trend of the distance between each group of phase points over time is calculated. A linear fit is performed on the trend, and the slope of the fitted line is the Lyapunov index, which is calculated to be 0.05. The instantaneous curvature sequence of phase space geometric distortion features (mean 0.5, variance 0.25, phase space stretching rate feature 50 mm / s), along with the correlation dimension 2.3, Kolmogorov entropy 0.8, and Lyapunov index 0.05, are integrated to form a deep characteristic index characterizing the mechanical deterioration state of a radioactive separation device. This index can comprehensively capture the nonlinear evolution law in the equipment deterioration process.

[0047] In this embodiment of the invention, by employing technical means such as establishing a benchmark vibration feature database for equipment health status, identifying and matching benchmark features for early fault characteristic vibration signals under operating conditions, performing local feature scaling correction through waveform similarity calculation to eliminate the influence of speed fluctuations, mapping the corrected signal to a high-dimensional phase space and analyzing the geometric features of phase point trajectories, and combining multiple nonlinear indicators to construct deep feature indicators, the technical problems of large vibration feature comparison errors caused by operating condition fluctuations, the inability of traditional shallow features to accurately characterize the nonlinear evolution law of mechanical deterioration, and the difficulty in accurately depicting the equipment deterioration state are overcome. Thus, it achieves adaptive elimination of operating condition interference and accurate extraction of deep feature indicators that reflect the essence of mechanical deterioration.

[0048] In a preferred embodiment of the present invention, step 3 above may include:

[0049] Step 3.1: Extract the spatial coordinates of the bolts connecting the base and the frame, the bearing locking nut positioning surface, and the flange butt joint circumferential seam as geometric constraint nodes. By constructing geometric interpolation conditions that satisfy the spatial position and continuity of each constraint node, a three-dimensional envelope surface is fitted and generated. Specifically, this includes: establishing a three-dimensional rectangular coordinate system to provide a unified benchmark for the extraction of geometric constraint nodes and the fitting of the three-dimensional envelope surface; using the geometric center of the base installation reference plane of the radioactive separation device as the origin of the spatial coordinates, ensuring that the three coordinate directions completely correspond to the actual installation posture of the equipment, setting the horizontal direction as the first coordinate direction, the vertical direction as the second coordinate direction, and the vertical upward direction as the third coordinate direction, ensuring that all subsequent spatial coordinate measurements and calculations are based on this coordinate system, avoiding fitting deviations caused by coordinate confusion; and performing precise three-dimensional coordinate measurements on the three key structural parts to extract geometric constraint nodes. The three key parts are the bolts connecting the base and the frame, the bearing locking nut positioning surface, and the flange butt joint circumferential seam. These three parts are the core areas most sensitive to equipment vibration response, and their spatial coordinates directly determine the fit of the three-dimensional envelope surface. A high-precision three-dimensional coordinate measuring instrument was used for measurement, with the measurement accuracy strictly controlled within 0.01 mm. This was adapted to the measurement requirements under strong radiation conditions, reducing the interference of the radiation environment on the measurement data. Three different measuring points were selected for coordinate measurement at each key location to avoid the impact of random errors from a single measuring point on the accuracy of the nodes. The final constraint node coordinates for each location were obtained by averaging the coordinates of the three measuring points. The specific calculation process was as follows: for each coordinate direction (first, second, and third coordinate directions), the measured values ​​of the three measuring points in that direction were added together, and then the sum was divided by 3 to obtain the average coordinate value for that coordinate direction. The average coordinate values ​​of the three coordinate directions were combined to obtain the geometric constraint node coordinates for that location. Based on actual measurements and calculations, the specific coordinates of the three constraint nodes are as follows: the geometric constraint node coordinates at the connection bolts between the base and the frame are 320 mm in the first coordinate direction, 210 mm in the second coordinate direction, and 160 mm in the third coordinate direction; the geometric constraint node coordinates at the bearing locking nut positioning surface are 320 mm in the first coordinate direction, 210 mm in the second coordinate direction, and 820 mm in the third coordinate direction; and the geometric constraint node coordinates at the flange butt joint are 680 mm in the first coordinate direction, 210 mm in the second coordinate direction, and 760 mm in the third coordinate direction. These three nodes precisely correspond to the three core critical parts of the equipment, providing a reliable benchmark for subsequent interpolation and fitting.

[0050] Geometric interpolation conditions were constructed to satisfy the spatial position and continuity requirements of each constraint node, laying the foundation for 3D envelope fitting. The geometric interpolation conditions mainly include two core requirements: first, positional constraints, meaning the fitted 3D envelope must accurately pass through three geometric constraint nodes to ensure a complete spatial correspondence between the envelope and the core part of the equipment; second, continuity constraints, meaning the envelope must satisfy tangential and normal continuity at and between constraint nodes to avoid surface misalignment, discontinuities, and abrupt changes, ensuring the envelope completely and smoothly wraps around the main structure of the equipment, conforming to the actual contour features of the equipment. Considering the structural characteristics and measurement accuracy requirements of the radioactive separation device, the moving least squares method was used to construct the geometric interpolation conditions. This method effectively weakens the 0.01 error during the measurement process. The system employs millimeter-level random errors to accommodate minute fluctuations in measurement data under strong radiation conditions, while ensuring the continuity and fitting accuracy of the surface. Based on the constructed geometric interpolation conditions, a three-dimensional envelope surface is fitted. A local weighting function is defined, and a Gaussian weighting function is selected. The width coefficient of the weighting function is set to 0.8, which balances the fit and smoothness of the surface and adapts to the contour features of the core structure of the equipment. The calculation logic of the weighting function is based on the spatial distance from the point to be fitted to each geometric constraint node as the core variable. The smaller the spatial distance, the larger the value of the weighting function, and the higher the fitting weight of that constraint node; the larger the spatial distance, the smaller the value of the weighting function, and the lower the fitting weight of that constraint node. Through this local weighting method, the envelope surface achieves a higher fit near the constraint nodes, better reflecting the spatial morphology of the core parts of the equipment. The basic polynomial form of the fitted surface is determined, using a bivariate quadratic polynomial as the fitting basis. This polynomial includes a constant term, a first-order term in the first coordinate direction, and a second-order term in the second coordinate direction. The first term, the second term in the first coordinate direction, the intersection term of the first and second coordinate directions, and the second term in the second coordinate direction are used to fully adapt to the spatial distribution characteristics of the three constraint nodes, while satisfying the requirements of tangential and normal continuity of the surface. During the fitting process, the optimization objective is to minimize the sum of squared weighted errors. The coefficients of each term in the bivariate quadratic polynomial are solved. The specific calculation process is as follows: Substitute the values ​​of the first and second coordinate directions of each geometric constraint node into the bivariate quadratic polynomial to calculate the predicted value of the polynomial, which is the calculated value in the third coordinate direction. Subtract the predicted value of the polynomial from the actual measured value of the third coordinate direction of the constraint node to obtain the fitting error of a single node. Multiply each fitting error by the value of the corresponding weight function to obtain the weighted error. Then, square each weighted error and sum the squared values ​​of all weighted errors to obtain the sum of squared weighted errors. By adjusting the coefficients of each term in the bivariate quadratic polynomial, the sum of squared weighted errors is minimized. The polynomial coefficients at this point are the final coefficients of the current point to be fitted.

[0051] Following the above fitting logic, within the spatial range of the main structure of the equipment, with a calculation interval of 0.1 mm, all coordinate combinations in the first and second coordinate directions are traversed, and the coefficients of the corresponding bivariate quadratic polynomials are solved point by point. The value of the third coordinate direction corresponding to each coordinate combination is calculated, and a complete three-dimensional surface is constructed point by point to complete the fitting and generation of the three-dimensional envelope surface. During the fitting process, three geometric constraint nodes are always used as the reference to ensure that the coordinates of each fitting point can fit the actual structural contour of the equipment, while taking into account the continuity and smoothness of the surface to avoid problems such as local protrusions and depressions that do not conform to the actual shape of the equipment. The position deviation of the fitted three-dimensional envelope surface is verified to ensure that it meets the accuracy requirements of structural dynamics numerical calculation. The specific calculation process of deviation verification is as follows: for each geometric constraint node, the difference between the actual coordinates of the node and the corresponding position coordinates on the three-dimensional envelope surface in the three coordinate directions is calculated; the difference in each coordinate direction is squared to obtain the square value of the difference in the three coordinate directions; the three square values ​​are added together to obtain the sum of squares of the differences; the square root of the sum of squares of the differences is taken to obtain the position deviation of the constraint node. Actual calculations show that the positional deviations of the three geometric constraint nodes are all within 0.2 mm, which fully meets the accuracy requirements of subsequent structural dynamics numerical derivation. Finally, a three-dimensional envelope surface is generated that completely encloses the main structural entity of the radioactive separation device.

[0052] Step 3.2 involves tetrahedral meshing of the three-dimensional envelope surface, applying dynamic load boundary conditions derived from deep feature indices to each mesh node, and performing numerical extrapolation of transient responses under different fault modes. Modal strain energy and modal kinetic energy at each mesh node along the vibration transmission path are extracted as the initial distribution of modal contributions. Local surface fitting is performed on all mesh nodes of the meshed envelope surface, and the principal curvature and Gaussian curvature at each node are calculated as discrete-point curvature feature values. These curvature feature values ​​are normalized and used as weighting coefficients for each mesh node, adjusting the initial distribution of modal contributions to obtain the corrected modal contribution distribution. Specifically, this includes tetrahedral meshing of the generated three-dimensional structural dynamics envelope surface. The mesh generation adopted a structured meshing method, with clearly defined meshing parameters: the maximum element size was set to 6 mm, the minimum element size to 2.5 mm, the minimum spacing between mesh nodes was controlled to 1.2 mm, and the aspect ratio of the mesh elements did not exceed 3. This ensured that the mesh quality met the accuracy requirements of transient dynamic numerical extrapolation and avoided distortion of extrapolation results due to poor mesh quality. After mesh generation, a quality check was performed on the mesh, and unqualified mesh elements were removed. Finally, the three-dimensional envelope surface was divided into 4892 tetrahedral mesh elements, generating a total of 1876 mesh nodes. All mesh nodes were evenly distributed on the three-dimensional envelope surface, completely covering the core structural area of ​​the equipment. An applied parameter was then applied to each mesh node after mesh generation. The dynamic load boundary conditions are obtained through inversion of deep characteristic indices. The load inversion process is as follows: Based on the vibration acceleration amplitude, vibration frequency, and other characteristics in the deep characteristic indices, combined with the stiffness parameters of the equipment structure, the deep characteristic indices are converted into corresponding dynamic loads through dynamic inversion calculations. Specifically, the dynamic load is equal to the vibration acceleration amplitude in the deep characteristic indices multiplied by the structural mass of the corresponding part of the equipment. The structural mass is obtained from the equipment design drawings: the structural mass of the base is 50 kg, the structural mass of the bearing is 8 kg, and the structural mass of the flange is 12 kg. The final applied dynamic load range is from 0.9 times the gravitational acceleration to 2.6 times the gravitational acceleration, where the gravitational acceleration is taken as 9.8 m. Every square second, the load application direction is consistent with the actual vibration direction of the equipment, that is, the corresponding loads are applied in the horizontal, longitudinal, and vertical directions to ensure that the load application conforms to the actual operating state of the equipment. For four fault modes, namely rotor imbalance, bearing failure, seal wear, and flow-induced vibration, transient response numerical simulations are carried out. The simulation parameters are set with a time step of 0.001 seconds and a total simulation time of 1 second, for a total of 1000 time steps. Transient response calculations are performed during the simulation to ensure calculation accuracy. Each fault mode is simulated independently to simulate the vibration response law of the equipment under different fault conditions. During the simulation, parameters such as vibration displacement, vibration velocity, and vibration acceleration of each grid node are recorded in real time.After numerical simulation, modal strain energy and modal kinetic energy were extracted from all 1876 grid nodes along the vibration transmission path. The modal strain energy and modal kinetic energy of each grid node were added together to obtain the initial value of the modal contribution of that node. The initial values ​​of the modal contributions of all grid nodes were combined to form the initial distribution of modal contributions. The modal strain energy was calculated as half of the modal strain energy multiplied by the structural stiffness coefficient and then multiplied by the square of the vibration displacement of the grid node. The modal kinetic energy was calculated as half of the modal kinetic energy multiplied by the structural mass and then multiplied by the square of the vibration velocity of the grid node. The initial value of the modal contribution of each grid node was calculated using this method to ensure that the initial distribution reflects the degree of vibration energy contribution of each node. Local surface fitting was performed on all 1876 grid nodes on the partitioned three-dimensional envelope surface. Taking a single grid node as the center, eight adjacent grid nodes around the node were selected to jointly construct a local quadratic fitting surface to ensure that the local surface accurately reflects the surface shape around the node. By fitting the surface twice locally, the principal curvature and Gaussian curvature at the central node are calculated as discrete point curvature feature values. The principal curvature is the maximum and minimum curvature in two orthogonal directions of the local surface at the node. The maximum curvature is calculated by selecting the curvature value corresponding to the maximum degree of curvature at the node on the local surface, and the minimum curvature is the curvature value corresponding to the minimum degree of curvature. The Gaussian curvature is the product of the two principal curvatures, that is, the Gaussian curvature is equal to the maximum principal curvature multiplied by the minimum principal curvature. The maximum principal curvature, minimum principal curvature, and Gaussian curvature are calculated for each node. The average value of the maximum and minimum principal curvatures is taken and used together with the Gaussian curvature as the discrete point curvature feature value of the node. The discrete point curvature feature values ​​of all grid nodes are normalized. The purpose of the normalization is to eliminate the difference in magnitude of the curvature feature values ​​of different nodes, so that the normalized curvature feature values ​​can be directly used as weighting coefficients. The specific calculation process for normalization is as follows: calculate the minimum and maximum values ​​of the curvature eigenvalues ​​of all grid nodes; then subtract the minimum value of the curvature eigenvalues ​​of all nodes from the curvature eigenvalue of the current node to obtain the difference; then subtract the minimum value of the curvature eigenvalues ​​of all nodes from the maximum value of the curvature eigenvalues ​​of all nodes to obtain the difference; finally, divide the difference of the current node by the difference of all nodes to obtain the normalized curvature eigenvalue. The normalized curvature eigenvalue ranges from 0 to 1. The larger the value, the greater the curvature of the surface at that node and the more concentrated the vibration energy transfer.The normalized curvature eigenvalues ​​are used as weighting coefficients in the spatial weighting function to adjust the initial distribution of modal contributions. The corrected modal contribution is calculated as follows: the corrected modal contribution equals the initial modal contribution of the corresponding grid node multiplied by the normalized curvature eigenvalue of that node. Weighting is performed node by node to ensure that the modal contribution of each node reflects the vibration energy transfer weight corresponding to its surface curvature characteristics. After weighting adjustment, a corrected modal contribution distribution covering all 1876 grid nodes is obtained. This distribution accurately reflects the spatial transfer and concentration patterns of vibration energy on the equipment structure.

[0053] In this embodiment of the invention, the technical means of extracting the spatial coordinates of three key parts as geometric constraint nodes and constructing geometric interpolation conditions to fit and generate a three-dimensional envelope surface, performing tetrahedral meshing on the three-dimensional envelope surface and applying dynamic loads to carry out transient response numerical deduction to obtain the initial distribution of modal contribution, and then calculating curvature characteristic values ​​through local surface fitting and constructing a spatial weight function to weight and correct the modal contribution, thus overcoming the technical problems of structural dynamics modeling not fitting the key constraint parts, low accuracy of vibration transmission path analysis, and distortion of analysis results caused by the failure to consider the spatial distribution of surface curvature in modal contribution.

[0054] In a preferred embodiment of the present invention, step 4 above may include:

[0055] Step 4.1: Normalize the contribution values ​​at each grid node in the corrected modal contribution distribution to obtain the spatial weight coefficients corresponding to each node; based on the spatial weight coefficients of the grid nodes where each vibration sensor measurement point is located, perform weighted fusion of the extracted deep feature indicators to obtain a weighted feature vector of fused spatial distribution characteristics; construct a benchmark fault feature library containing four fault modes: rotor imbalance, bearing failure, seal wear, and flow-induced vibration; input the weighted feature vector into the multi-class discrimination model; and output the fault category with the closest distance as the identification result by calculating the distance measure between the weighted feature vector and the benchmark features of each fault mode. Specifically, this includes: normalizing the obtained corrected modal contribution distribution to obtain the spatial weight coefficients corresponding to each grid node. The specific calculation process of normalization is to first traverse the corrected modal contribution values ​​of all grid nodes and find the maximum and minimum values. The numerical values ​​are then calculated by subtracting the minimum value of the modified modal contribution of all nodes from the modified modal contribution of a single grid node, resulting in the first difference. The maximum value of the modified modal contribution of all nodes is then subtracted from the minimum value, resulting in the second difference. Finally, the spatial weight coefficient of the node is equal to the first difference divided by the second difference. Calculations show that the spatial weight coefficients of all grid nodes range from zero to one; a larger value indicates a higher contribution of the node to vibration energy transfer. Among all grid nodes, three grid nodes corresponding to the measurement points of the anti-radiation self-powered wireless vibration sensor are located: the measurement point at the bolt connection between the base and the frame, the measurement point at the positioning surface of the rotor shaft bearing lock nut, and the measurement point at the circumferential seam where the final centrifugal separation chamber and the pipeline flange meet. The spatial weight coefficients corresponding to these three measurement points are directly extracted as the core weight parameters for subsequent feature weighting fusion.

[0056] Based on the spatial weight coefficients of the three measuring points, the extracted deep feature indicators characterizing the mechanical deterioration state are weighted and fused. These deep feature indicators include six components: mean phase space geometric distortion, variance phase space geometric distortion, phase space spread rate, correlation dimension, Kolmogorov entropy, and Lyapunov index. The specific process of weighted fusion involves multiplying each component of the deep feature indicator corresponding to each measuring point by the spatial weight coefficient of that measuring point node, then summing the weighted results of the same feature component from the three measuring points sequentially to obtain the fused value of that feature component. The fused values ​​of the six feature components are combined in a fixed order to ultimately form a weighted feature vector representing the spatial distribution characteristics of the fused equipment structure. A baseline fault feature library containing four typical fault modes—rotor imbalance, bearing failure, seal wear, and flow-induced vibration—is pre-constructed. For each fault mode, data is collected from the equipment at the fault location. Sufficient standard samples under the fault condition are used to generate standard baseline features for the corresponding faults according to the same feature extraction and weighted fusion method. The baseline features of the four faults are stored in the baseline fault feature library as the comparison standard for fault identification. The weighted feature vector is input into the trained multi-class discriminant model. Fault identification is completed by calculating the distance measure between the weighted feature vector and the baseline features of each fault mode in the baseline fault feature library. The specific calculation process of the distance measure is to perform difference operation on the same position feature component of the weighted feature vector and the baseline features of a certain type of fault one by one, square each difference, add them all, and then take the square root of the sum to obtain the distance value corresponding to the fault type. The distance values ​​between the weighted feature vector and the four fault baseline features are calculated in turn, and the fault category with the smallest distance value is selected as the final fault identification result output by the model.

[0057] Step 4.2: Based on the most recent fault category as the identification result, for each identified fault category, extract the sensitive feature components corresponding to the fault type from the weighted feature vector under the fault mode. Calculate the fault severity quantification index by the ratio of the amplitude of the sensitive feature component to the benchmark threshold. Based on the energy-concentrated grid region in the corrected modal contribution distribution, and combined with the corresponding physical structure parts of the grid region, determine the specific physical location of the fault. Specifically, this includes: taking the fault category with the smallest distance value as the final identification result, extracting the corresponding sensitive feature components from the weighted feature vector for this identified fault type. The sensitive feature component for rotor imbalance faults is the phase space expansion rate, and the sensitive feature component for bearing faults is the phase space geometric distortion variance. The sensitive feature component of the fault is the correlation dimension, and the sensitive feature component of the flow-induced vibration fault is the Lyapunov index. Only a single sensitive feature component is extracted for each fault, excluding the interference of irrelevant feature components. The fault severity quantification index is calculated by the ratio of the amplitude of the sensitive feature component to the health status benchmark threshold. The specific calculation process of the fault severity quantification index is that the fault severity quantification index is equal to the amplitude of the extracted sensitive feature component divided by the health status benchmark threshold corresponding to the fault type. The health status benchmark threshold for rotor imbalance fault is set to 50 mm / s, the health status benchmark threshold for bearing fault is set to 0.25, the health status benchmark threshold for seal wear fault is set to 2.3, and the health status benchmark threshold for flow-induced vibration fault is set to 0.05. The larger the calculated fault severity quantification index, the more severe the fault degradation. An index greater than 1.5 is considered a moderate fault, and an index greater than 2.5 is considered a severe fault. In the corrected modal contribution distribution, grid nodes with a normalized spatial weight coefficient greater than 0.8 are selected. The continuous region formed by these nodes is the grid region where vibration energy is concentrated, which is the core region where fault-induced vibration energy accumulation occurs. The grid region where energy is concentrated is matched one-to-one with the corresponding solid structure parts of the three-dimensional envelope surface. The three-dimensional envelope surface is pre-divided into three solid structure regions, corresponding to the bolt connection between the base and the frame, the positioning surface of the rotor shaft end bearing locking nut, and the circumferential seam where the final centrifugal separation chamber and the pipeline flange are connected. The structural region to which the energy-concentrated grid region belongs is determined, and the specific physical location of the fault is directly identified, thereby achieving accurate physical location under multi-source coupled faults.

[0058] In this embodiment of the invention, by normalizing the modified modal contribution distribution to obtain spatial weight coefficients, and then using these coefficients to weight and fuse deep feature indicators to obtain weighted feature vectors, a multi-fault mode benchmark feature library is constructed, fault identification is achieved through a multi-classification discrimination model, and sensitive feature components are extracted to calculate the fault severity quantification index. The specific location of the fault is then located using a combination of energy-concentrated grid regions. Therefore, the technical problems of difficulty in decoupling and separating multi-source coupled faults, inaccurate quantification of fault severity, and difficulty in accurately locating the physical location of faults are overcome. This achieves the technical effect of accurately decoupling and identifying, quantifying and evaluating, and precisely locating multiple types of faults such as rotor imbalance, bearing failure, seal wear, and flow-induced vibration, thereby improving the accuracy and reliability of fault diagnosis.

[0059] In a preferred embodiment of the present invention, step 5 above may include:

[0060] Step 5.1: Establish a historical degradation trajectory database for key components, containing vibration characteristic evolution sequences from healthy state to failure under different fault types. Using the current fault severity quantification index as the current observation value, calculate the morphological similarity between the current observation value and each degradation trajectory in the historical degradation trajectory database. Select several degradation trajectories with similarity exceeding a preset threshold as a reference trajectory set. Specifically, this includes establishing a historical degradation trajectory database for key components of the radioactive separation device. Key components specifically include the base-to-frame connecting bolts, rotor shaft end bearings, and the flange connection between the final-stage centrifugal separation chamber and the pipeline. These three components are the core parts most sensitive to equipment vibration response and have the highest failure rate; their degradation state directly determines the overall operation of the machine. Safety: The database covers four fault types: rotor imbalance, bearing failure, seal wear, and flow-induced vibration. For each fault type, at least 50 complete degradation process samples are collected. Each sample records the vibration characteristic evolution sequence from a healthy state to a complete failure state, with the time interval set to 1 hour. Each time point corresponds to a set of data, including the fault severity quantification index at that moment, deep feature indicators such as the mean of phase space geometric distortion, variance of phase space geometric distortion, phase space spread rate, correlation dimension, Kolmogorov entropy, Lyapunov index, and corresponding operating parameters including speed and flow rate. All samples are stored according to fault type, and the fault development time of each sample is labeled as the total time from healthy to failure. The degradation rates at each stage are used to form a complete historical degradation trajectory database for key components, ensuring that the database covers all scenarios with different fault types and degradation rates. The calculated quantitative index of the current fault severity is obtained and used as the current observation value. Simultaneously, deep feature indicators corresponding to the current moment are extracted to form the current observation sequence, which includes the current observation value and corresponding deep features. The morphological similarity between the current observation sequence and each degradation trajectory in the historical degradation trajectory database is calculated. All trajectories in the historical degradation trajectory database that match the current fault type are selected. Each trajectory contains observation values ​​and deep features at several time points, forming a historical evolution sequence. The current observation sequence is aligned with a single historical evolution sequence to ensure that their feature dimensions are consistent. It includes a fault severity quantification index and six deep features; calculate the covariance between the two, which is equal to the reciprocal of the number of feature dimensions, multiplied by the sum of the products of the current observation value minus the mean of the current observation sequence and the corresponding feature value of the historical evolution sequence minus the mean of the historical evolution sequence for each feature dimension; calculate the standard deviation of the current observation sequence and the standard deviation of the historical evolution sequence respectively, which is equal to the reciprocal of the number of feature dimensions, multiplied by the sum of the squares of the corresponding value minus the mean of the sequence for each feature dimension, and then take the square root; morphological similarity is equal to the covariance divided by the standard deviation of the current observation sequence multiplied by the standard deviation of the historical evolution sequence. The calculated similarity value ranges from 0 to 1. The closer the value is to 1, the more similar the morphology of the two trajectories and the closer their degradation patterns are.A preset threshold for morphological similarity is set at 0.85. The morphological similarity between the current observation sequence and each historical degradation trajectory of the same type of fault is assessed one by one. All historical degradation trajectories with a similarity greater than 0.85 are selected and compiled into a reference trajectory set. The number of trajectories in the reference trajectory set is controlled between 10 and 15. If the number of selected trajectories exceeds 15, the 15 with the highest similarity are selected as the final reference trajectory set; if the number of selected trajectories is less than 10, the threshold is appropriately lowered to 0.8 to ensure the representativeness of the reference trajectory set.

[0061] Step 5.2: Construct a state transition equation based on the reference trajectory set. Use a sequential importance sampling method to recursively update the parameters of the state transition equation, while simultaneously introducing the current operating stress parameters to correct the state transition process. Extrapolate the vibration characteristic evolution trend based on the updated state transition equation, using a preset failure threshold as the boundary condition. Calculate the time interval from the current moment to the first crossing of the failure threshold by the evolution trend curve as the predicted remaining life. Specifically, this includes: constructing a state transition equation for equipment failure degradation based on the obtained reference trajectory set. The state transition equation adopts a linear state transition form to describe the change in the equipment degradation state over time. The core parameters include the state transition coefficient, drift coefficient, and error coefficient. The state transition coefficient reflects the influence of the current degradation state on the next degradation state, the drift coefficient reflects the overall trend of the degradation state, and the error coefficient reflects random disturbances during the degradation process. The parameters of the state transition equation are recursively updated to ensure that the equation parameters adapt to the actual degradation state of the current device. The specific update process is as follows: initial values ​​for the parameters are set: the initial value for the state transition coefficient is set to 0.95, the initial value for the drift coefficient is set to 0.02, and the initial value for the error coefficient is set to 0.01. Several trajectories are randomly selected from the reference trajectory set as sampling samples, with the sampling quantity being the reference... Half the total number of trajectory sets; calculate the parameter weight corresponding to each sample, the weight is equal to the morphological similarity between the sample and the current observation sequence, the higher the similarity, the greater the weight; based on the sample and the corresponding weight, recursively update the parameters of the state transition equation, the updated state transition coefficient is equal to the sum of the state transition coefficients of all sampled samples multiplied by the corresponding weights, and then divided by the sum of all weights; the update method of drift coefficient and error coefficient is the same as that of state transition coefficient, and the parameter recursive update is completed successively, repeated 3 times to ensure that the parameters converge to a stable value and avoid the distortion of prediction results due to initial parameter deviation; during the parameter update process, the current operating conditions of the device are introduced. Force parameters are used to correct the state transition process. The operating stress parameters specifically include the current equipment speed and flow rate. The correction process is as follows: Calculate the deviation between the current operating stress and the operating stress of the corresponding historical samples in the reference trajectory set. The deviation is equal to the current operating stress minus the mean of the operating stress of the historical samples in the reference trajectory set. Then, calculate the stress correction coefficient, which is equal to 1 plus the deviation multiplied by 0.001. The coefficient 0.001 is the stress influence coefficient, calibrated using a large amount of historical data. Finally, multiply the drift coefficient of the state transition equation by the stress correction coefficient to complete the correction of the state transition process, so that the state transition equation can fully consider the impact of the current operating stress on equipment degradation.

[0062] After parameter updates and corrections are completed, the evolution trend of equipment vibration characteristics is extrapolated using the updated state transition equations. The extrapolation process adopts a time-step recursive approach, with a time step set to 1 hour, consistent with the time interval of the historical degradation trajectory database. The fault severity quantification index for the next moment is calculated step by step, forming a complete vibration characteristic evolution trend curve. A failure threshold is preset for each fault type. The failure threshold is determined based on the average fault severity quantification index of all failure samples in the historical degradation trajectory database. The specific calculation process is as follows: the failure threshold equals the sum of the fault severity quantification indices of all failure samples of the same type of fault, divided by the total number of failure samples. The calculated failure thresholds are 5.0 for rotor imbalance faults, 4.8 for bearing faults, 4.5 for seal wear faults, and 5.2 for flow-induced vibration faults. Using the preset failure thresholds as boundary conditions, the time point when the vibration characteristic evolution trend curve first crosses the failure threshold is determined, and the time interval from the current moment to that time point is calculated. This time interval is the predicted remaining life value of the key components of the equipment. The specific calculation process is as follows: determine the time coordinate corresponding to the current moment, find the time coordinate of the evolution trend curve that is first greater than or equal to the failure threshold, and subtract the time coordinate of the current moment from the time coordinate of the first crossing of the failure threshold. The difference is the remaining lifetime prediction value, in hours. For example, if the current moment is the 1000th hour of operation, and the time point when the evolution trend curve first crosses the failure threshold is the 1200th hour, then the remaining lifetime prediction value is 1200 hours minus 1000 hours, which calculates to a remaining lifetime prediction value of 200 hours.

[0063] In this embodiment of the invention, by employing the technical means of establishing a database of historical degradation trajectories of key components, filtering a set of highly similar reference trajectories based on a quantitative index of the current fault severity, constructing a state transition equation based on the reference trajectory set and recursively updating parameters through sequential importance sampling, and simultaneously introducing the current operating stress to correct the state transition process, and then extrapolating the evolution trend based on the updated equation, calculating the remaining life prediction value, and outputting the degradation trend result, the technical problems of static and fixed parameters, distortion of degradation trend extrapolation due to the lack of integration of historical degradation patterns and real-time operating stress, and low accuracy of remaining life estimation in traditional life prediction models are overcome. Thus, the invention achieves dynamic optimization of the prediction model, accurate extrapolation of equipment degradation trends, and accurate prediction of the remaining life of key components.

[0064] like Figure 2 As shown, embodiments of the present invention also provide a vibration monitoring and mechanical condition assessment system for radioactive separation devices, including:

[0065] The extraction module is used to perform adaptive noise reduction on the raw vibration data. It dynamically adjusts the noise reduction parameters based on the statistical characteristics of radiation interference and flow-induced turbulence interference and extracts the vibration signals of early fault characteristics.

[0066] The comparison module is used to compare the vibration signals of early fault characteristics with the baseline vibration characteristics in the healthy state, and extract the deep feature indicators that characterize the mechanical deterioration state.

[0067] The calculation module is used to construct the structural dynamics envelope of the solid parts using the spatial coordinates of the bolts connecting the base and the frame, the positioning surface of the bearing locking nut, and the flange butt joint as constraint nodes; it performs finite element meshing on the envelope and conducts transient response numerical derivation to obtain the modal contribution distribution on the vibration transmission path under different fault modes; it performs discrete point curvature estimation on the mesh nodes of the meshed envelope to obtain the curvature characteristic value at each node, and performs spatial weighted correction on the modal contribution distribution based on the curvature characteristic value to obtain the corrected modal contribution distribution;

[0068] The allocation module is used to assign weights to deep feature indicators based on the corrected modal contribution distribution. It separates the fault features corresponding to rotor imbalance, bearing failure, seal wear and flow-induced vibration through pattern recognition, and quantifies the severity of each type of fault and locates the fault location.

[0069] The update module is used to dynamically update the prediction parameters based on the severity of various faults, combined with historical degradation data and current operating stress, through degradation trend modeling, to obtain the remaining life and degradation trend prediction results of key components.

[0070] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for vibration monitoring and mechanical condition assessment of a radioactive separation device, characterized in that, The method includes: Step 1: Simultaneously collect raw vibration data from multiple measurement points using radiation-resistant, self-powered wireless vibration sensors located at the bolts connecting the base of the radioactive separation device to the multi-stage series frame, the positioning surface of the rotor shaft end bearing lock nut, and the circumferential seam where the final centrifugal separation chamber meets the pipeline flange. Adaptive denoising processing is performed on the raw vibration data. The denoising parameters are dynamically adjusted based on the statistical characteristics of radiation interference and flow-induced turbulence interference, and vibration signals exhibiting early fault characteristics are extracted, including: The original vibration data were processed by time-domain windowing and segmentation. Power spectral density analysis was performed on the signal in each segment to extract the peak pulse frequency band corresponding to radiation interference and the broadband random disturbance frequency band corresponding to flow-induced turbulence interference. The signal-to-noise ratio (SNR) under the current environment is calculated based on the distribution characteristics of the broadband random disturbance frequency band corresponding to the peak pulse frequency band and the flow-induced turbulence interference. The number of wavelet packet decomposition layers and threshold coefficients are dynamically adjusted based on the SNR. Wavelet packet decomposition is performed on the original vibration data. The adjusted threshold coefficients are used to perform soft threshold quantization on the coefficients of each frequency band. The threshold-processed coefficients are then reconstructed to obtain the noise-reduced vibration signal. Empirical mode decomposition is performed on the noise-reduced vibration signal to obtain several intrinsic mode function components. The intrinsic mode function components whose correlation with the frequency characteristics of early equipment failure exceeds a preset threshold are selected for signal reconstruction to extract the vibration signal of early failure characteristics. Step 2: Compare the vibration signals of early fault characteristics with the baseline vibration characteristics under healthy conditions to extract deep feature indicators that characterize the mechanical deterioration state. Step 3: Using the spatial coordinates of the bolts connecting the base and the frame, the positioning surface of the bearing locking nut, and the flange butt joint as constraint nodes, construct the structural dynamics envelope of the solid part; perform finite element meshing on the envelope and conduct transient response numerical deduction to obtain the modal contribution distribution on the vibration transmission path under different fault modes; perform discrete point curvature estimation on the mesh nodes of the meshed envelope to obtain the curvature characteristic value at each node, and perform spatial weighted correction on the modal contribution distribution based on the curvature characteristic value to obtain the corrected modal contribution distribution; Step 4: Based on the corrected modal contribution distribution, the deep feature indexes are weighted and assigned. The fault features corresponding to rotor imbalance, bearing failure, seal wear and flow-induced vibration are separated by pattern recognition. The severity of each type of fault and the location of the fault are quantified respectively. Step 5: Based on the severity of various faults, combined with historical degradation data and current operating stress, dynamically update the prediction parameters through degradation trend modeling to obtain the remaining life and degradation trend prediction results of key components.

2. The method for vibration monitoring and mechanical condition assessment of a radioactive separation device according to claim 1, characterized in that, By comparing vibration signals exhibiting early fault characteristics with baseline vibration characteristics under healthy conditions, deep-seated feature indicators characterizing the mechanical degradation state are extracted, including: A baseline vibration feature database for equipment in a healthy state is established, containing time-domain waveforms, spectral structures, and shaft center trajectory features corresponding to each typical operating condition. Operating condition identification is performed on the vibration signals of extracted early fault features to obtain current speed and flow operating parameters. Baseline vibration features matching the current operating condition are retrieved from the baseline vibration feature database. The waveform similarity between the vibration signals of early fault features and the baseline vibration features is calculated. Local feature scaling correction is performed on frequency bands with waveform similarity below a preset threshold to eliminate the influence of speed fluctuations on feature comparison. The vibration signals after adaptive correction are mapped to a high-dimensional phase space to obtain the phase point trajectories in the high-dimensional phase space. Geometric feature analysis of phase point trajectories is performed. Instantaneous curvature sequences are obtained by calculating the spatial change rate of the direction connecting adjacent phase points. The mean and variance of the instantaneous curvature sequences are statistically analyzed as phase space geometric distortion features. At the same time, the cumulative arc length of the phase point trajectory per unit time is calculated as the phase space extension rate feature. The phase space geometric distortion features and phase space extension rate features, together with correlation dimension, Kolmogorov entropy and Lyapunov exponent, constitute a deep feature index characterizing the mechanical degradation state.

3. The method for vibration monitoring and mechanical condition assessment of a radioactive separation device according to claim 2, characterized in that, Using the spatial coordinates of the bolts connecting the base and the frame, the positioning surface of the bearing locking nut, and the flange butt joint as constraint nodes, a structural dynamic envelope surface for the solid parts is constructed. The envelope surface is then meshed using finite element methods, and transient response numerical derivations are performed to obtain the modal contribution distribution along the vibration transmission path under different fault modes. Discrete point curvature estimation is performed on the mesh nodes of the meshed envelope surface to obtain curvature characteristic values ​​at each node. Based on these curvature characteristic values, the modal contribution distribution is spatially weighted and corrected to obtain the corrected modal contribution distribution, including: The spatial coordinates of the bolts connecting the base and the frame, the positioning surface of the bearing locking nut, and the flange butt joint are extracted as geometric constraint nodes. By constructing geometric interpolation conditions that satisfy the spatial position and continuity of each constraint node, a three-dimensional envelope surface is fitted and generated. The three-dimensional envelope surface is tetrahedral meshed, and dynamic load boundary conditions derived from deep feature indices are applied to each mesh node. Transient response numerical extrapolation is performed under different fault modes. Modal strain energy and modal kinetic energy at each mesh node along the vibration transmission path are extracted as the initial distribution of modal contribution. Local surface fitting is performed on all mesh nodes of the meshed envelope surface, and the principal curvature and Gaussian curvature at each node are calculated as discrete point curvature feature values. The curvature feature values ​​are normalized and used as weighting coefficients for each mesh node. The initial distribution of modal contribution is then weighted and adjusted to obtain the corrected modal contribution distribution.

4. The method for vibration monitoring and mechanical condition assessment of a radioactive separation device according to claim 3, characterized in that, Based on the corrected modal contribution distribution, weights are assigned to deep feature indicators. Pattern recognition is used to separate the fault characteristics corresponding to rotor imbalance, bearing failure, seal wear, and flow-induced vibration. The severity of each type of fault and its location are then quantified, including: The contribution values ​​at each grid node in the corrected modal contribution distribution are normalized to obtain the spatial weight coefficients corresponding to each node. Based on the spatial weight coefficients of the grid nodes where each vibration sensor measurement point is located, the extracted deep feature indicators are weighted and fused to obtain a weighted feature vector of fused spatial distribution characteristics. A benchmark fault feature library containing four fault modes—rotor imbalance, bearing failure, seal wear, and flow-induced vibration—is constructed. The weighted feature vector is input into a multi-class discrimination model. By calculating the distance measure between the weighted feature vector and the benchmark features of each fault mode, the fault category with the closest distance is output as the recognition result. Based on the most recent fault category as the identification result, for the identified fault category, the sensitive feature component corresponding to the fault category is extracted from the weighted feature vector of the fault mode. The fault severity quantification index is calculated by the ratio of the amplitude of the sensitive feature component to the benchmark threshold. Based on the grid region with concentrated energy in the corrected modal contribution distribution, combined with the corresponding physical structure part of the grid region, the specific physical location of the fault is determined.

5. The method for vibration monitoring and mechanical condition assessment of a radioactive separation device according to claim 4, characterized in that, Based on the severity of various failures, combined with historical degradation data and current operating stress, the prediction parameters are dynamically updated through degradation trend modeling to obtain the remaining life and degradation trend prediction results of key components, including: Establish a historical degradation trajectory database for key components, containing vibration characteristic evolution sequences from healthy state to failure under different fault types; use the current fault severity quantification index as the current observation value, and select several degradation trajectories with similarity exceeding a preset threshold as a reference trajectory set by calculating the morphological similarity between the current observation value and each degradation trajectory in the historical degradation trajectory database. A state transition equation is constructed based on a reference trajectory set. The parameters of the state transition equation are recursively updated using a sequential importance sampling method. At the same time, the current operating stress parameters are introduced to correct the state transition process. The vibration characteristic evolution trend is extrapolated based on the updated state transition equation. The preset failure threshold is used as the boundary condition. The time interval from the current moment to the first crossing of the failure threshold by the evolution trend curve is calculated as the remaining life prediction value. The evolution trend curve is output as the degradation trend prediction result.

6. A vibration monitoring and mechanical condition assessment system for a radioactive separation device, wherein the system implements the method as described in any one of claims 1 to 5, characterized in that, include: The extraction module is used to perform adaptive noise reduction on the raw vibration data. It dynamically adjusts the noise reduction parameters based on the statistical characteristics of radiation interference and flow-induced turbulence interference and extracts vibration signals with early fault characteristics. The comparison module is used to compare the vibration signals of early fault characteristics with the baseline vibration characteristics under healthy conditions, and extract deep feature indicators that characterize the mechanical deterioration state. The calculation module is used to construct the structural dynamics envelope of the solid parts using the spatial coordinates of the bolts connecting the base and the frame, the positioning surface of the bearing locking nut, and the flange butt joint as constraint nodes; it performs finite element meshing on the envelope and conducts transient response numerical derivation to obtain the modal contribution distribution on the vibration transmission path under different fault modes; it performs discrete point curvature estimation on the mesh nodes of the meshed envelope to obtain the curvature characteristic value at each node, and performs spatial weighted correction on the modal contribution distribution based on the curvature characteristic value to obtain the corrected modal contribution distribution; The allocation module is used to assign weights to deep feature indicators based on the corrected modal contribution distribution. It separates the fault features corresponding to rotor imbalance, bearing failure, seal wear and flow-induced vibration through pattern recognition, and quantifies the severity of each type of fault and locates the fault location. The update module is used to dynamically update the prediction parameters based on the severity of various faults, combined with historical degradation data and current operating stress, through degradation trend modeling, to obtain the remaining life and degradation trend prediction results of key components.

7. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 5.

Images