Spacecraft thermal control pump degradation perception and residual life prediction method and system

Abstract

The application discloses a spacecraft thermal control pump degradation perception and residual life prediction method and system based on multi-source data fusion. The method collects vibration spectrum data, infrared thermal image data and fluid pressure drop data of the spacecraft thermal control pump, extracts degradation characteristics after preprocessing and time alignment, and constructs a composite health index; a three-parameter Weibull degradation model including shape parameters, scale parameters and position parameters is constructed according to the composite health index, the composite health index is used as a particle filtering observation to adaptively update the model parameters; based on the updated parameter posterior distribution and the preset failure threshold, the residual service life prediction result is determined, and the warning information is output to the thermal control system autonomous management platform. The application can realize unified representation of multi-source degradation information of the spacecraft thermal control pump, improve the adaptability of life prediction, and provide a basis for the autonomous management of the thermal control system.

Description

Technical Field

[0001] This invention relates to the field of spacecraft thermal control system health management technology, and in particular to a method and system for spacecraft thermal control pump degradation sensing and remaining life prediction based on multi-source data fusion. Background Technology

[0002] The thermal control pump is a core power component in the thermal control fluid loop of a spacecraft, and its operating status directly affects the flow distribution, heat exchange efficiency, and on-orbit reliability of the thermal control system. During long-term operation, the thermal control pump may experience degradation phenomena such as bearing wear, rotor imbalance, mechanical loosening, seal wear, local overheating, leakage, cavitation, blockage, impeller wear, or decreased hydraulic efficiency. These degradation phenomena may manifest as changes in vibration spectrum, abnormal pump body temperature field, and changes in inlet and outlet pressure difference, respectively.

[0003] Existing health monitoring technologies for pump equipment typically identify the condition based on a single signal or a single fault type. For example, they may identify mechanical faults based on vibration signals or trigger threshold alarms based on temperature and pressure signals. While these methods can reflect some local anomalies, they struggle to unify mechanical degradation, thermal degradation, sealing degradation, and hydraulic performance degradation into a comprehensive health characterization that can be used for degradation modeling. This makes it difficult to collaboratively utilize degradation information from different data sources.

[0004] In predicting remaining lifetime, existing methods typically employ empirical degradation models, fixed-parameter statistical models, or offline-trained models to estimate equipment degradation trends. For spacecraft thermal pumps, a relatively long stable operating phase is common, the onset of degradation is uncertain, and the degradation rate varies with operating conditions, wear levels, and sealing status. If the degradation model cannot characterize the time shift from a healthy state to a degraded state, and cannot adaptively update model parameters using real-time observation data, it can easily lead to overfitting in the early healthy phase, insufficient tracking of the degradation phase, and unstable remaining lifetime predictions.

[0005] Furthermore, existing health monitoring solutions mostly focus on fault diagnosis, trend display, or single-threshold alarms, lacking effective integration between lifetime prediction results and the autonomous management of the thermal control system. For spacecraft in orbit, the thermal control pump status assessment results need to be further transformed into early warning information that can be used for backup pump switching, thermal control loop reconfiguration, or maintenance decisions. Therefore, there is an urgent need for a technical solution that integrates multi-source degradation sensing of spacecraft thermal control pumps, construction of comprehensive health indicators, online updating of degradation models, remaining lifetime prediction, and early warning decision output. Summary of the Invention

[0006] This invention provides a method and system for spacecraft thermal control pump degradation sensing and remaining lifetime prediction based on multi-source data fusion, in order to solve the problems in the prior art where multi-source degradation information of spacecraft thermal control pumps is difficult to form a unified health characterization, degradation models are difficult to adapt to changes in degradation initiation time and degradation rate, and remaining lifetime prediction results are difficult to support the autonomous management of thermal control systems.

[0007] According to a first aspect of the present invention, a method for spacecraft thermal control pump degradation sensing and remaining lifetime prediction based on multi-source data fusion is provided, comprising: Multi-source operational data of the spacecraft's thermal control pump are collected. The multi-source operational data includes vibration spectrum data, infrared thermographic data, and fluid pressure drop data. The vibration spectrum data is used to characterize the mechanical degradation of the spacecraft's thermal control pump, the infrared thermographic data is used to characterize the thermal degradation or sealing degradation of the spacecraft's thermal control pump, and the fluid pressure drop data is used to characterize the hydraulic performance degradation of the spacecraft's thermal control pump. The vibration spectrum data, infrared thermal image data, and fluid pressure drop data are preprocessed respectively to obtain time-aligned vibration preprocessed data, thermal image preprocessed data, and pressure drop preprocessed data; Vibration degradation features are extracted from the vibration preprocessing data, thermal image degradation features are extracted from the thermal image preprocessing data, and pressure drop degradation features are extracted from the pressure drop preprocessing data. A composite health index characterizing the overall degradation degree of the spacecraft's thermal control pump is constructed based on the vibration degradation features, thermal image degradation features, and pressure drop degradation features. A three-parameter Weibull degradation model is constructed based on the composite health index. The three-parameter Weibull degradation model includes shape parameters, scale parameters, and position parameters. The position parameters are used to characterize the degradation start time or fault-free operation time of the spacecraft's thermal control pump from a healthy state to a degraded state. The shape parameter, scale parameter, and position parameter are used to form the state vector of the particle filter. The composite health index is used as the observation of the particle filter. A nonlinear observation relationship is established between the composite health index and the output degradation of the three-parameter Weibull degradation model. Based on the nonlinear observation relationship, the parameters of the three-parameter Weibull degradation model are adaptively updated to obtain the updated parameter posterior distribution. Based on the updated parameter posterior distribution and the preset failure threshold, the time distribution of the composite health index reaching the preset failure threshold for the first time is determined, and the remaining service life prediction result of the spacecraft heat control pump is determined based on the time distribution. Early warning information is generated based on the predicted remaining service life and / or predicted pump efficiency decline, and the early warning information is output to the thermal control system autonomous management platform.

[0008] According to a second aspect of the present invention, a spacecraft thermal control pump degradation sensing and remaining lifetime prediction system based on multi-source data fusion is provided, comprising: A multi-source data acquisition module is used to acquire multi-source operating data of the spacecraft's thermal control pump. The multi-source operating data includes vibration spectrum data, infrared thermographic data, and fluid pressure drop data. The vibration spectrum data is used to characterize the mechanical degradation of the spacecraft's thermal control pump, the infrared thermographic data is used to characterize the thermal degradation or sealing degradation of the spacecraft's thermal control pump, and the fluid pressure drop data is used to characterize the hydraulic performance degradation of the spacecraft's thermal control pump. The data preprocessing module is used to preprocess the vibration spectrum data, infrared thermal image data and fluid pressure drop data respectively to obtain time-aligned vibration preprocessed data, thermal image preprocessed data and pressure drop preprocessed data; The degradation feature extraction module is used to extract vibration degradation features based on the vibration preprocessing data, extract thermal image degradation features based on the thermal image preprocessing data, and extract pressure drop degradation features based on the pressure drop preprocessing data. The feature fusion module is used to construct a composite health index characterizing the overall degradation degree of the spacecraft's thermal control pump based on the vibration degradation features, thermal imaging degradation features, and pressure drop degradation features. The degradation model construction module is used to construct a three-parameter Weibull degradation model based on the composite health index. The three-parameter Weibull degradation model includes shape parameters, scale parameters, and position parameters. The position parameters are used to characterize the degradation start time or fault-free operation time of the spacecraft's heat control pump from a healthy state to a degraded state. The model update module is used to form the state vector of the particle filter by the shape parameter, scale parameter and position parameter, use the composite health index as the observation of the particle filter, establish a nonlinear observation relationship between the composite health index and the output degradation of the three-parameter Weibull degradation model, and adaptively update the parameters of the three-parameter Weibull degradation model based on the nonlinear observation relationship to obtain the updated parameter posterior distribution. The lifespan prediction module is used to determine the time distribution of the first time the composite health index reaches the preset failure threshold based on the updated parameter posterior distribution and the preset failure threshold, and to determine the remaining lifespan prediction result of the spacecraft heat control pump based on the time distribution. The early warning decision module is used to generate early warning information based on the remaining service life prediction results and / or pump efficiency decay prediction results, and output the early warning information to the thermal control system autonomous management platform.

[0009] This invention collects vibration spectrum data, infrared thermographic data, and fluid pressure drop data of spacecraft thermal control pumps, and uses these three types of data to characterize mechanical degradation, thermal degradation or sealing degradation, and hydraulic performance degradation, respectively. This allows for the synchronous acquisition and coordinated utilization of degradation information of thermal control pumps in multiple dimensions of thermo-mechanical-fluid, thereby improving the completeness of the characterization of the degradation state of spacecraft thermal control pumps.

[0010] This invention extracts degradation features from vibration preprocessing data, thermal imaging preprocessing data, and pressure drop preprocessing data, and constructs a composite health index based on these features. This enables monitoring information corresponding to different data formats, dimensions, and degradation mechanisms to be transformed into a unified comprehensive degradation quantity, providing consistent input for subsequent degradation model construction and life prediction.

[0011] This invention constructs a three-parameter Weibull degradation model based on composite health indicators, including shape parameters, scale parameters, and position parameters. The shape parameters, scale parameters, and position parameters are combined into a particle filter state vector, and the composite health indicators are used as observations to adaptively update the model parameters. Among them, the position parameters are used to characterize the degradation start time or fault-free operation time of the spacecraft's thermal control pump from a healthy state to a degradation state, which is helpful in distinguishing between the stable operation stage and the degradation stage. The particle filter update mechanism is beneficial for tracking the degradation rate change based on real-time observation data, thereby improving the adaptability and stability of the remaining service life prediction.

[0012] This invention determines the time distribution of the composite health index reaching the preset failure threshold for the first time based on the updated parameter posterior distribution and the preset failure threshold, and obtains the remaining service life prediction result accordingly. Then, based on the remaining service life prediction result and / or pump efficiency degradation prediction result, it generates early warning information and outputs it to the thermal control system autonomous management platform, so that the degradation perception result and service life prediction result can be transformed into the decision information required for the autonomous management of the thermal control system, providing a basis for backup pump switching, thermal control loop reconstruction or maintenance arrangements. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the architecture of a spacecraft thermal control pump degradation sensing and remaining lifetime prediction system based on multi-source data fusion, provided in an embodiment of the present invention.

[0014] Figure 2 This is a schematic diagram of the overall process of the spacecraft thermal control pump degradation sensing and remaining lifetime prediction method based on multi-source data fusion provided in an embodiment of the present invention.

[0015] Figure 3 This is another overall flowchart illustrating the spacecraft thermal control pump degradation sensing and remaining lifetime prediction method based on multi-source data fusion provided in this embodiment of the invention.

[0016] Figure 4 This is a schematic diagram of the multi-source operation data acquisition process provided in an embodiment of the present invention.

[0017] Figure 5 This is a schematic diagram of the preprocessing flow for multi-source operational data provided in an embodiment of the present invention.

[0018] Figure 6 This is another schematic diagram of the preprocessing of multi-source operational data provided in an embodiment of the present invention.

[0019] Figure 7 This is a schematic diagram of the degradation feature extraction process provided in an embodiment of the present invention.

[0020] Figure 8 A schematic diagram illustrating the principle of constructing composite health indicators and feature fusion provided in this embodiment of the invention.

[0021] Figure 9 This is a schematic diagram of the composite health indicator construction process provided in an embodiment of the present invention.

[0022] Figure 10 This is a schematic diagram illustrating the construction process of the three-parameter Weibull degradation model provided in an embodiment of the present invention.

[0023] Figure 11 This is a schematic diagram of the adaptive update process for particle filter parameters provided in an embodiment of the present invention.

[0024] Figure 12 This is a schematic diagram of the remaining useful life prediction process provided in an embodiment of the present invention.

[0025] Figure 13 This is a schematic diagram of the early warning information generation and output process provided in an embodiment of the present invention.

[0026] Figure 14 This is a schematic diagram comparing the predicted remaining useful life with the actual useful life provided in the experimental embodiments of the present invention.

[0027] Figure 15 This is a schematic diagram of the pump efficiency attenuation prediction curve and early warning point provided for the experimental embodiment of the present invention. Detailed Implementation

[0028] The technical solution of the present invention will be described below with reference to the accompanying drawings and embodiments. The following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Unless otherwise specified, the embodiments and technical features described in this specification can be combined with each other.

[0029] This embodiment relates to a degradation sensing and remaining life prediction scheme for spacecraft thermal control pumps. The spacecraft thermal control pump is located in the spacecraft's thermal control fluid loop and is used to drive the thermal control medium to circulate between heat exchangers, cold plates, pipelines, and storage components. During long-term operation, the thermal control pump may experience degradation phenomena such as bearing wear, rotor imbalance, mechanical loosening, impeller abnormalities, mechanical seal wear, localized overheating, leakage, cavitation, blockage, impeller corrosion, or decreased hydraulic efficiency. These degradation phenomena may be coupled between mechanical, thermal, and fluid hydraulic performance. Therefore, this embodiment uses vibration spectrum data, infrared thermography data, and fluid pressure drop data to sense mechanical degradation, thermal degradation or seal degradation, and hydraulic performance degradation, respectively. The multi-source sensing results are converted into composite health indicators, and then the remaining life is predicted based on a three-parameter Weibull degradation model and particle filtering.

[0030] refer to Figure 1 In one embodiment, the system includes a multi-source data acquisition module 1, a data preprocessing module 2, a degradation feature extraction module 3, a feature fusion and modeling module 4, a model update and prediction module 5, an early warning and decision-making module 6, and a thermal control system autonomous management platform 7. The multi-source data acquisition module 1 includes a vibration sensor 11, an infrared thermal imager 12, and a pressure sensor 13. The vibration sensor 11 acquires vibration spectrum data of the spacecraft's thermal control pump, the infrared thermal imager 12 acquires infrared thermal image data of the spacecraft's thermal control pump, and the pressure sensor 13 acquires inlet and outlet pressure data of the spacecraft's thermal control pump and forms fluid pressure drop data. The data preprocessing module 2 performs noise reduction, correction, compensation, normalization, and time synchronization processing on the three-source operating data; the degradation feature extraction module 3 extracts vibration degradation features, thermal image degradation features, and pressure drop degradation features; the feature fusion and modeling module 4 includes a feature fusion function for constructing composite health indicators and a degradation model construction function for constructing a three-parameter Weibull degradation model; the model update and prediction module 5 includes a model update function for adaptively updating model parameters and a life prediction function for determining the remaining service life prediction results; the early warning and decision-making module 6 generates early warning information based on the prediction results and outputs it to the thermal control system autonomous management platform 7.

[0031] refer to Figure 2 In one embodiment, the spacecraft thermal control pump degradation sensing and remaining lifetime prediction method based on multi-source data fusion sequentially includes S1-S6, specifically including multi-source operational data acquisition, three-source operational data preprocessing, degradation feature extraction, composite health index construction, three-parameter Weibull degradation model construction, particle filter parameter adaptive update, remaining lifetime prediction, and early warning information generation and output. In one embodiment, referencing Figure 3 The spacecraft thermal control pump degradation sensing and remaining lifetime prediction method based on multi-source data fusion provided in this embodiment includes steps S10 to S70 in sequence.

[0032] Step S10: Collect multi-source operating data of the spacecraft's thermal control pump. The multi-source operating data includes vibration spectrum data, infrared thermographic data, and fluid pressure drop data. Vibration spectrum data is used to characterize the mechanical degradation of the spacecraft's thermal control pump, infrared thermographic data is used to characterize the thermal degradation or sealing degradation of the spacecraft's thermal control pump, and fluid pressure drop data is used to characterize the hydraulic performance degradation of the spacecraft's thermal control pump.

[0033] In step S10, vibration sensor 11 collects vibration spectrum data, infrared thermal imager 12 collects infrared thermal image data, and pressure sensor 13 collects inlet pressure data and outlet pressure data to form fluid pressure drop data. The vibration spectrum data, infrared thermal image data, and fluid pressure drop data are respectively input into data preprocessing module 2, which generates vibration preprocessed data, thermal image preprocessed data, and pressure drop preprocessed data, respectively.

[0034] In one specific embodiment of step S10, the vibration sensor 11 is disposed at the bearing housing of the spacecraft's thermal control pump, the motor end, or the bearing housing of the drive motor; the infrared thermal imager 12 faces the pump body surface, the area where the pump head connects to the motor, or the mechanical seal area; and the pressure sensors 13 are respectively disposed at the inlet and outlet pipes of the spacecraft's thermal control pump. Vibration spectrum data corresponds to mechanical degradation, infrared thermal image data corresponds to thermal degradation or seal degradation, and fluid pressure drop data corresponds to hydraulic performance degradation. (Reference) Figure 4 Step S10 may include steps S110 to S130.

[0035] Step S110: Vibration spectrum data is collected by vibration sensors installed at the bearing housing of the spacecraft's thermal control pump, the motor end, or the bearing housing of the drive motor. The vibration spectrum data is used to characterize mechanical degradation caused by bearing wear, rotor imbalance, mechanical loosening, or impeller abnormalities.

[0036] In one specific embodiment, the vibration sensor 11 is a high-sensitivity piezoelectric accelerometer with a sensitivity of 100 mV / g and a frequency response range of 0.1 Hz to 10 kHz. The vibration sensor 11 can be mounted on the front and rear bearing housings of the heat pump drive motor and can be arranged horizontally and vertically to capture vibration signals caused by bearing wear, rotor imbalance, mechanical loosening, and impeller abnormalities. In another specific embodiment, the sampling frequency of the vibration sensor 11 can be set to 25.6 kHz, the measurement range can be ±50 g, the resolution can be 0.001 g, and the analysis frequency band can be 0.1 kHz to 10 kHz.

[0037] In one example, the analysis frequency band from 0.1 kHz to 10 kHz can cover the characteristic frequencies of common rotating mechanical failures in thermal pumps, including BPFO (Ball Pass Frequency Outer Race), BPFI (Ball Pass Frequency Inner Race), BSF (Ball Spin Frequency), FTF (Fundamental Train Frequency), and impeller pass frequency. By acquiring vibration signals within the above frequency band, the ability to detect mechanical degradation of bearings, impellers, and rotor components can be improved.

[0038] Step S120: Infrared thermal image data is collected by an infrared thermal imager facing the pump body surface, the pump head and motor connection area or the mechanical seal area. The infrared thermal image data is used to characterize thermal degradation or seal degradation caused by frictional heat generation, local overheating, seal wear or leakage.

[0039] In one specific embodiment, the infrared thermal imager 12 can be an uncooled focal plane infrared thermal imager, installed on the side or directly in front of the pump body at an unobstructed field of view, at a distance of 0.5m to 2m from the pump body surface, and focused on the mechanical seal at the connection between the pump head and the motor or the main heat-generating area of ​​the pump body. The NETD (Noise Equivalent Temperature Difference) of the infrared thermal imager 12 can be less than 50mK, and the NETD can be calibrated at 25°C; the spatial resolution can be 640×512 pixels, and the frame rate can be 30Hz. The infrared thermal imager 12 can acquire temperature distribution images of the pump body's ROI (Region of Interest).

[0040] Infrared thermography data is used to reflect the thermal stability of the pump body and the health status of the sealing system. Wear of the mechanical seal may lead to increased frictional heat generation, fluid leakage may cause abnormal cold or hot zones in the local temperature field, and abnormal temperature rise in the connection area between the motor and the pump body may correspond to mechanical coupling degradation. Therefore, infrared thermography data can supplement the information on thermal degradation or seal degradation that vibration data cannot cover.

[0041] In step S130, pressure sensors installed on the inlet and outlet pipes of the spacecraft thermal control pump are used to collect inlet pressure data and outlet pressure data, respectively. Fluid pressure drop data is obtained based on the inlet pressure data and outlet pressure data. The fluid pressure drop data is used to characterize hydraulic performance degradation caused by cavitation, blockage, impeller wear, or decreased hydraulic efficiency.

[0042] In one specific embodiment, the pressure sensor 13 can be a high-frequency response pressure transmitter, installed on straight sections of the pump inlet and outlet pipelines, respectively. The installation position can be at least 5D away from elbows, where D is the inner diameter of the pipeline, to reduce the influence of elbows, valves, or local turbulence on pressure measurement. The measurement accuracy of the pressure sensor 13 can be 0.1%FS, the measurement range can be 0MPa to 1.0MPa, and the sampling frequency can be 10Hz. The pressure sensor 13 is used to capture pressure fluctuations caused by pump inlet and outlet pressure differences, fluid pulsation, and cavitation.

[0043] In one example, the inlet pressure data is denoted as P_in(t), the outlet pressure data is denoted as P_out(t), and the fluid pressure drop data can be represented as: DeltaP(t) = P_out(t) - P_in(t) Where t represents the sampling time, and DeltaP(t) represents the difference between the pump outlet pressure and the inlet pressure. Fluid pressure drop data can be used to characterize the hydraulic output capacity and hydraulic performance degradation of thermally controlled pumps.

[0044] Step S20: The vibration spectrum data, infrared thermal image data and fluid pressure drop data are preprocessed respectively to obtain time-aligned vibration preprocessed data, thermal image preprocessed data and pressure drop preprocessed data.

[0045] In step S20, the data preprocessing module 2 receives the vibration spectrum data, infrared thermographic data, and fluid pressure drop data acquired in step S10. The data preprocessing module 2 performs corresponding preprocessing on the vibration spectrum data, infrared thermographic data, and fluid pressure drop data respectively, and performs time synchronization or time alignment on the three-source preprocessed data to obtain vibration preprocessed data, thermographic preprocessed data, and pressure drop preprocessed data.

[0046] refer to Figure 5 In one specific embodiment of step S20, the vibration spectrum data undergoes denoising, resampling, and normalization to obtain vibration preprocessed data; the infrared thermal image data undergoes image correction, spatial denoising, and pump body region of interest extraction to obtain thermal image preprocessed data; the fluid pressure drop data undergoes denoising, operating condition compensation, and normalization to obtain pressure drop preprocessed data; the vibration preprocessed data, thermal image preprocessed data, and pressure drop preprocessed data are time-synchronized or time-aligned. Step S20 may include steps S210 to S240.

[0047] Step S210: Denoise, resample, and normalize the vibration spectrum data to obtain vibration preprocessed data.

[0048] In one specific implementation, bandpass filtering can be applied to the vibration spectrum data to remove low-frequency drift and non-target frequency noise. The analysis band for the bandpass filter can be set from 0.1 kHz to 10 kHz. The bandpass-filtered vibration data is then resampled, for example, to 10.24 kHz, and normalized to zero mean and unit variance. If the original vibration signal is denoted as x_vib(t), with mean_vib and standard deviation std_vib, then the normalized vibration signal can be expressed as:

[0049] x_vib_norm(t)=(x_vib(t)-mean_vib) / std_vib Where x_vib_norm(t) represents the normalized vibration signal in the vibration preprocessing data.

[0050] Step S220: Image correction, spatial denoising, and region of interest extraction of the pump body are performed on the infrared thermal image data to obtain thermal image preprocessing data.

[0051] In a specific implementation, refer to Figure 6 This process involves performing NUC (Non-Uniformity Correction) on infrared thermal image data to reduce fixed-pattern noise caused by differences in the response of different pixels in the infrared detector. Subsequently, bilateral filtering, median filtering, or other spatial denoising methods can be used to remove random noise while preserving the pump body temperature edge information. Then, temperature distribution data within the pump body's Region of Interest (ROI) is extracted based on the pump body contour, installation location, field-of-view calibration results, or a preset mask. If the infrared thermal image frame is denoted as I(x,y,t), and the pump body's region of interest is denoted as the ROI, the thermal image preprocessing data can include the temperature value T(x,y,t) of each pixel within the ROI at time t.

[0052] Step S230: Denoise, operating condition compensation and normalization are performed on the fluid pressure drop data to obtain pressure drop preprocessed data.

[0053] In a specific implementation, refer to Figure 6 High-frequency noise in pressure signals can be removed using moving average filtering, low-pass filtering, or median filtering. For example, the moving average filter window can be set to 100 points, corresponding to a data window of approximately 10 seconds at a sampling frequency of 10Hz. Since fluid pressure drop is affected by medium temperature, viscosity, flow rate, and operating point, pressure drop data can be compensated for operating conditions. In one example, the pressure drop data can be compensated or normalized to a standard operating condition of 20°C. If the pressure drop data after operating condition compensation and normalization is denoted as DeltaP_norm(t), then DeltaP_norm(t) is used as pressure drop preprocessing data in subsequent feature extraction steps.

[0054] Step S240: Time synchronization or time alignment is performed on the vibration preprocessing data, thermal image preprocessing data, and voltage drop preprocessing data to obtain time-aligned vibration preprocessing data, thermal image preprocessing data, and voltage drop preprocessing data.

[0055] In a specific implementation, refer to Figure 6 Time alignment of three-source data can be performed based on a unified sampling time base, acquisition timestamp, or interpolation method. For data with significant differences in sampling frequencies, alignment can be performed using fixed time windows, such as mapping vibration statistical features, thermal imaging statistical features, and pressure drop statistical features within the same time window to the same feature time point. Time synchronization or time alignment can avoid mismatches in degradation states caused by sampling delays or frequency differences between different data sources.

[0056] Step S30: Extract vibration degradation features from vibration preprocessing data, extract thermal image degradation features from thermal image preprocessing data, extract pressure drop degradation features from pressure drop preprocessing data, and construct a composite health index characterizing the overall degradation degree of the spacecraft's thermal control pump based on the vibration degradation features, thermal image degradation features, and pressure drop degradation features.

[0057] refer to Figure 7 and Figure 8 In one specific embodiment of step S30, vibration preprocessing data enters the vibration feature extraction branch to obtain the root mean square value, peak factor, and spectral entropy of vibration; thermal image preprocessing data enters the thermal image feature extraction branch to obtain the mean temperature, temperature variance, and hot spot temperature of the region of interest in the pump body; pressure drop preprocessing data enters the pressure drop feature extraction branch to obtain the inlet and outlet pressure difference and pressure drop change rate. The process of extracting vibration degradation features, thermal image degradation features, and pressure drop degradation features in step S30 may include steps S310 to S330.

[0058] Step S310: Extract the root mean square value, peak factor and spectral entropy of vibration from the vibration preprocessing data to obtain the vibration degradation characteristics.

[0059] In a specific implementation, for a vibration signal x_i within a time window, i=1,2,...,N, the root mean square (RMS) value of the vibration can be expressed as: RMS=sqrt(sum(x_ix_i) / N) RMS is used to characterize changes in vibration energy. Bearing wear, rotor imbalance, or impeller abnormalities often lead to an increase in vibration energy levels, so RMS can serve as a sensitive feature for mechanical degradation.

[0060] The peak factor CF can be expressed as: CF = max(abs(x_i)) / RMS Among them, CF is used to characterize the changes in the impact component in the vibration signal. Local damage to the bearing, cavitation, or mechanical loosening may lead to an enhancement of the impact component, so the peak factor can supplement the RMS's ability to characterize impact-related faults.

[0061] The spectral entropy SE can be expressed as: SE=-sum(p_kln(p_k)) p_k = E_k / sum(E_k) Where E_k represents the spectral energy of the k-th frequency band, p_k represents the proportion of the k-th frequency band's energy to the total spectral energy, k=1,2,...,K, and K represents the number of frequency bands. Spectral entropy SE is used to characterize the complexity variation of the spectral energy distribution.

[0062] Step S320: Extract the mean temperature, temperature variance, and hot spot temperature of the region of interest in the pump body based on the thermal image preprocessing data to obtain the thermal image degradation characteristics.

[0063] In a specific implementation, for the temperature value T_j within the pump body ROI, j=1,2,...,M, the average temperature T_mean can be expressed as: T_mean=sum(T_j) / M The temperature variance T_var can be expressed as: T_var=sum((T_j-T_mean)(T_j-T_mean)) / M The hotspot temperature T_max can be expressed as: T_max = max(T_j) Among them, the mean temperature T_mean is used to characterize the overall thermal state of the pump body, the temperature variance T_var is used to characterize the non-uniformity of the temperature field, and the hot spot temperature T_max is used to characterize the risk of local overheating.

[0064] Step S330: Extract the inlet and outlet pressure difference and pressure drop change rate from the pressure drop preprocessing data to obtain the pressure drop degradation characteristics.

[0065] In a specific implementation, the inlet and outlet pressure difference can be obtained by DeltaP(t) = P_out(t) - P_in(t). The rate of change of pressure drop can be obtained based on the pressure drop difference between adjacent sampling times. If the sampling interval between adjacent times t and t-1 is Delta_t, then the rate of change of pressure drop can be expressed as:

[0066] dDeltaP_dt(t)=(DeltaP(t)-DeltaP(t-1)) / Delta_t Among them, the inlet and outlet pressure difference is used to characterize the hydraulic output state, and the pressure drop change rate is used to characterize the rate of hydraulic performance degradation.

[0067] refer to Figure 8 In step S30, the vibration degradation features, thermal imaging degradation features, and pressure drop degradation features are standardized and reduced by principal components to obtain the vibration principal component score, thermal imaging principal component score, and pressure drop principal component score. The correlation between each principal component score and the pump efficiency attenuation label is used to determine the vibration weight, thermal imaging weight, and pressure drop weight. The vibration principal component score, thermal imaging principal component score, and pressure drop principal component score are weighted according to their corresponding weights to generate a composite health index.

[0068] refer to Figure 8 and Figure 9 In one specific implementation of step S30, the composite health indicator construction process includes four processing steps: feature standardization, principal component dimensionality reduction, correlation weight determination, and weighted fusion. For example... Figure 9 As shown, the process of constructing the composite health index in step S30 may include steps S340 to S370.

[0069] Step S340: Standardize the vibration degradation features, thermal image degradation features, and pressure drop degradation features respectively to obtain standardized vibration features, standardized thermal image features, and standardized pressure drop features.

[0070] In a specific implementation, each feature class can be standardized with zero mean and unit variance, or it can be normalized by minimax. Standardization is used to eliminate differences in the dimensions and numerical scales of different features, and to prevent a certain type of feature from occupying an unreasonable weight in the fusion process due to its large numerical range.

[0071] Step S350: Perform principal component dimensionality reduction on the standardized vibration features, standardized thermal image features, and standardized pressure drop features respectively to obtain the vibration principal component score, thermal image principal component score, and pressure drop principal component score.

[0072] In a specific implementation, PCA (Principal Component Analysis) can be used to reduce the dimensionality of each standardized feature class. The vibration principal component score is denoted as F_vib, the thermal imaging principal component score as F_therm, and the pressure drop principal component score as F_press. The score of each principal component can be selected from the first principal component score with the highest contribution rate, or multiple principal components can be selected based on their cumulative contribution rates to form a feature score class.

[0073] In step S360, the correlation coefficients between the vibration principal component score, the thermal image principal component score, and the pressure drop principal component score and the pump efficiency attenuation label are calculated respectively, and the absolute values ​​of the correlation coefficients are normalized to obtain the vibration weight, thermal image weight, and pressure drop weight.

[0074] In a specific implementation, pump efficiency degradation labels can be derived from ground test bench calibration data, historical operating data, or pump performance evaluation results. The correlation coefficient can be the Pearson correlation coefficient. If the i-th principal component score sequence is X_i and the pump efficiency degradation label sequence is Y, then the correlation coefficient rho_i can be expressed as:

[0075] rho_i=corr(X_i,Y) The corresponding weight w_i can be expressed as: w_i=abs(rho_i) / sum(abs(rho_i)) Here, i corresponds to the three types of features: vibration, thermal imaging, and pressure drop. By normalizing the absolute value of the correlation coefficient, features that are more strongly correlated with pump efficiency decay can be given higher weights.

[0076] Step S370: The composite health index is constructed by weighting and fusing the vibration principal component score, thermal image principal component score, pressure drop principal component score, vibration weight, thermal image weight, and pressure drop weight.

[0077] In a specific implementation, the Composite Health Index (CHI) can be expressed as: CHI=w_vibF_vib+w_thermF_therm+w_pressF_press Where w_vib + w_therm + w_press = 1, w_vib represents the vibration weight, w_therm represents the thermal imaging weight, and w_press represents the pressure drop weight. In a specific embodiment, w_vib can be 0.4 to 0.5, w_therm can be 0.3 to 0.4, and w_press can be 0.1 to 0.2. For example, w_vib = 0.45, w_therm = 0.35, and w_press = 0.20. Through the above fusion method, CHI can simultaneously reflect the comprehensive degradation degree of three dimensions: mechanical, thermal or sealing, and hydraulic performance.

[0078] Step S40: Construct a three-parameter Weibull degradation model based on composite health indicators. The three-parameter Weibull degradation model includes shape parameters, scale parameters, and position parameters. The position parameters are used to characterize the degradation start time or fault-free operation time of the spacecraft's thermal control pump from a healthy state to a degraded state.

[0079] refer to Figure 5 In step S40, the composite health index is input as the comprehensive degradation amount of the spacecraft's thermal control pump into the three-parameter Weibull degradation model. The three-parameter Weibull degradation model is integrated with the particle filter parameter update process, and the composite health index participates as an observation in the online update of the three-parameter Weibull degradation model.

[0080] refer to Figure 10 In one specific embodiment of step S40, the three-parameter Weibull degradation model construction process includes determining the comprehensive degradation amount, mapping the physical meaning of the three parameters, and constructing the degradation function. Step S40 may include steps S410 to S430.

[0081] Step S410: Use composite health indicators as the overall degradation of the spacecraft's thermal control pump.

[0082] In one specific implementation, CHI can be normalized to the range of 0 to 1, so that CHI increases with the degree of degradation. When CHI approaches a preset failure threshold, the spacecraft's thermal pump can be considered to be close to a failure state or a state of performance degradation.

[0083] Step S420: The shape parameter characterizes the characteristics of the degradation rate as a function of operating time, the scale parameter characterizes the characteristic lifetime of the spacecraft thermal control pump, and the position parameter characterizes the time offset of the spacecraft thermal control pump entering the degradation phase after the stable operation phase.

[0084] In the three-parameter Weibull degradation model, the shape parameter is denoted as beta, the scale parameter as eta, and the location parameter as tau. The shape parameter beta describes the degradation rate trend, the scale parameter eta describes the characteristic lifetime scale, and the location parameter tau describes the degradation onset time, fault-free operation time, or the time offset between the start of the degradation phase and the start of the degradation phase after a stable operation phase. The location parameter tau can be calibrated using historical degradation data or dynamically estimated during online updates using particle filtering.

[0085] Step S430: Based on the comprehensive degradation amount, shape parameters, scale parameters, and position parameters, a three-parameter Weibull degradation model is constructed to describe the degradation process of the spacecraft's heat control pump.

[0086] In a specific implementation, the degradation function of the three-parameter Weibull degradation model can be expressed as: F(t;beta,eta,tau)=1-exp(-((t-tau) / eta)^beta) Where t is the running time; when t <= tau, F(t;beta,eta,tau) can be set to 0 or a degradation amount close to 0; when t > tau, F(t;beta,eta,tau) increases with time. In one example, the initial shape parameter beta_0 can be 2.5, the initial scale parameter eta_0 can be 5000h, and the initial position parameter tau_0 can be set according to experimental data, historical data, or experience.

[0087] Step S50: The shape parameter, scale parameter and position parameter are combined to form the state vector of the particle filter. The composite health index is used as the observation of the particle filter. A nonlinear observation relationship is established between the composite health index and the output degradation of the three-parameter Weibull degradation model. Based on the nonlinear observation relationship, the parameters of the three-parameter Weibull degradation model are adaptively updated to obtain the updated parameter posterior distribution.

[0088] In step S50, particle filtering constructs a state vector from the shape, scale, and position parameters of the three-parameter Weibull degradation model, and uses the real-time composite health index obtained in step S30 as the observation. The particle filtering process includes state prediction, observation likelihood calculation, particle weight update, and resampling.

[0089] refer to Figure 11 In one specific embodiment of step S50, the adaptive update process for particle filter parameters includes particle set initialization, state vector construction, state transition prediction, establishment of nonlinear observation relationships, particle observation likelihood calculation, weight update, and resampling. Step S50 may include steps S500 to S550.

[0090] Step S500: Initialize the particle set.

[0091] In one specific implementation, N particles are generated, each representing a set of three-parameter Weibull model parameters: shape parameter beta, scale parameter eta, and position parameter tau. The i-th particle can be represented as X_0_i = [beta_0_i, eta_0_i, tau_0_i]. The initial particles can be randomly sampled around the initial parameters beta_0, eta_0, and tau_0, or generated based on historical degradation data or a prior distribution. The initial weight of each particle can be set to 1 / N. In one example, the number of particles N can be between 1000 and 5000, for example, N = 2000.

[0092] Step S510: Combine the shape parameters, scale parameters, and position parameters into a state vector.

[0093] The state vector can be represented as: X_k=[beta_k,eta_k,tau_k] Where k represents the k-th update time, and beta_k, eta_k, and tau_k represent the shape parameter, scale parameter, and position parameter at the k-th update time, respectively.

[0094] Step S520: Construct the state transition equation of the state vector based on the random walk law of the parameters, and predict the particle state based on the state transition equation.

[0095] In a specific implementation, the state transition equation can be expressed as: X_k=X_(k-1)+w_k Here, w_k represents system noise, used to describe the slow drift, individual differences, or operating condition disturbances in model parameters caused by equipment aging. Through this state transition equation, each particle is predicted from the previous update time to the current update time.

[0096] Step S530: Using the degradation amount output by the three-parameter Weibull degradation model at the current moment as the predicted observation value, establish a nonlinear observation relationship between the predicted observation value and the composite health index obtained in real time.

[0097] In a specific implementation, the observation equation can be expressed as: Z_k=F(t_k;X_k)+v_k Where Z_k represents the real-time CHI observation obtained at the k-th update time, F(t_k;X_k) represents the predicted degradation output of the three-parameter Weibull degradation model under the state vector X_k, and v_k is the observation noise. This observation equation couples the real-time multi-source fusion degradation index with the three-parameter Weibull degradation model.

[0098] Step S540: Calculate the observation likelihood of each particle based on the nonlinear observation relationship, and update the particle weights based on the observation likelihood.

[0099] In a specific implementation, if the observation noise follows a Gaussian distribution, the observation likelihood L_i of the i-th particle can be expressed as: L_i=exp(-((Z_k-F(t_k;X_k_i))(Z_k-F(t_k;X_k_i))) / (2sigma_vsigma_v)) Where X_k_i represents the state vector of the i-th particle, and sigma_v represents the standard deviation of the observation noise. The weight of the i-th particle is updated according to L_i, and the weights of all particles are normalized.

[0100] Step S550: Resample according to the updated particle weights to obtain the updated posterior distribution of parameters.

[0101] In one specific implementation, resampling can be determined based on the effective particle count. The effective particle count N_eff can be expressed as:

[0102] N_eff=1 / sum(w_iw_i) Here, w_i represents the normalized weight of the i-th particle. When N_eff is less than or equal to the preset resampling threshold, system resampling, residual resampling, or other resampling methods can be performed to remove low-weight particles and replicate high-weight particles. After resampling, the particle set represents the updated posterior distribution of the shape, scale, and position parameters. In one example, the resampling threshold can be set to N / 2.

[0103] In an optional implementation, the model update and prediction module 5 can implement the three-parameter Weibull degradation model and particle filter update process within the PHM (Prognostics and Health Management) tool framework, and define the state transition function, observation function, and likelihood function within this framework. This implementation is not limited to a specific software tool, nor does it affect the use of other equivalent computational frameworks to implement the above model update process.

[0104] In an optional embodiment, the model update and prediction module 5 can also constrain the parameter update process of the three-parameter Weibull degradation model based on the relative changes in vibration degradation characteristics, thermal imaging degradation characteristics, and pressure drop degradation characteristics. Specifically, the rate of change of vibration degradation characteristics, thermal imaging degradation characteristics, and pressure drop degradation characteristics within the current time window can be calculated separately, and the dominant degradation mode can be determined based on the relative magnitudes of the three types of rates of change. When the rate of change of vibration degradation characteristics is greater than the rate of change of thermal imaging degradation characteristics and the rate of change of pressure drop characteristics, the current degradation process is determined to be dominated by mechanical degradation. When the rate of change of thermal imaging degradation characteristics is greater than the rate of change of vibration degradation characteristics and the rate of change of pressure drop characteristics, the current degradation process is determined to be dominated by thermal degradation or sealing degradation. When the rate of change of pressure drop characteristics is greater than the rate of change of vibration degradation characteristics and the rate of change of thermal imaging degradation characteristics, the current degradation process is determined to be dominated by hydraulic performance degradation. The model update and prediction module 5 adjusts the system noise covariance and / or observation noise covariance in the particle filter according to the dominant degradation mode, so that the parameter update magnitude of the three-parameter Weibull degradation model matches the current dominant degradation mechanism. This degradation mechanism constraint reduces the impact of short-term disturbances in multi-source data on model parameter updates and ensures that the model parameter update process is consistent with the actual degradation mechanism of the spacecraft's thermal control pump, thereby improving the stability of subsequent remaining service life prediction results.

[0105] Step S60: Based on the updated parameter posterior distribution and the preset failure threshold, determine the time distribution of the composite health index reaching the preset failure threshold for the first time, and determine the predicted remaining service life of the spacecraft thermal control pump based on the time distribution.

[0106] refer to Figure 12In one specific embodiment of step S60, multiple future degradation trajectories are generated based on the updated parameter posterior distribution; when the composite health index in any degradation trajectory reaches a preset failure threshold for the first time, the first crossing time corresponding to that trajectory is determined; multiple first crossing times form a first crossing time distribution; the difference between the first crossing time distribution and the current time forms a remaining useful life distribution; and the remaining useful life point estimate and confidence interval are then determined from the remaining useful life distribution. Step S60 may include steps S610 to S650.

[0107] Step S610: Based on the updated posterior distribution of parameters, multiple degradation trajectories are predicted to obtain the predicted trajectory distribution of composite health indicators.

[0108] In one specific implementation, multiple sets of parameters can be extracted from the updated posterior distribution of the parameters, with each set corresponding to a future degradation trajectory. By predicting multiple degradation trajectories, the probability distribution of future changes in CHI can be obtained.

[0109] Step S620: Determine the first crossing time when the composite health index first reaches the preset failure threshold in each degradation trajectory based on the predicted trajectory distribution.

[0110] The preset failure threshold is denoted as TH. When the CHI in a certain predicted trajectory first satisfies CHI(t)>=TH, that moment is the first pass time T_FPT corresponding to that predicted trajectory. FPT (First Passage Time) is used to represent the time when the degradation index first reaches the preset failure threshold. In a specific embodiment, TH can be 0.8, which is the example threshold under normalized CHI.

[0111] In a specific implementation, when the three-parameter Weibull degenerate function can be directly inversely solved, the first crossing time can be estimated based on the parameters beta_t, eta_t, and tau_t at the t-th update time: T_FPT_t=tau_t+eta_t(-ln(1-TH))^(1 / beta_t) The corresponding remaining useful life can be expressed as: RUL_t=T_FPT_t-t_now Where t_now represents the current time. The above inverse formula can be used to estimate the first penetration time under a single set of parameters; under the particle filter framework, the first penetration time can be calculated separately for each set of parameters corresponding to each particle, thus forming the first penetration time distribution.

[0112] Step S630: Determine the distribution of first-wearing times based on multiple first-wearing times.

[0113] Since multiple degradation trajectories correspond to multiple sets of parameter samples, multiple first-crossing times constitute a first-crossing time distribution. This distribution can reflect the uncertainty of failure time caused by sensor measurement noise, model parameter uncertainty, and the randomness of the degradation process.

[0114] Step S640: Determine the remaining service life distribution based on the difference between the first penetration time distribution and the current time.

[0115] If the current time is t_now, then the remaining lifetime RUL_i corresponding to a certain predicted trajectory can be expressed as: RUL_i=T_FPT_i-t_now Where T_FPT_i represents the first crossing time corresponding to the i-th degradation trajectory. Multiple RUL_i constitute the remaining lifetime distribution.

[0116] Step S650: Determine the estimated value and confidence interval of the remaining useful life based on the distribution of remaining useful life, and obtain the prediction result of the remaining useful life.

[0117] In a specific implementation, the point estimate of remaining useful life can be the mean, median, or mode of the remaining useful life distribution. The confidence interval can be determined based on quantiles. For example, the 2.5% quantile and the 97.5% quantile can be used as the upper and lower bounds of the 95% confidence interval.

[0118] In one alternative implementation, Monte Carlo simulation can be used to perform uncertainty quantification and propagation analysis. Specifically, the uncertainties of sensor measurement noise, observation noise, and the parameters of the three-parameter Weibull model can be sampled to obtain multiple sets of uncertainty samples; remaining useful life prediction can be performed based on each set of uncertainty samples to obtain multiple RUL prediction samples; and a confidence interval for the remaining useful life prediction result can be formed based on the multiple RUL prediction samples. This uncertainty quantification result can be output together with the RUL point estimate.

[0119] Step S70: Generate early warning information based on the remaining service life prediction results and / or pump efficiency decay prediction results, and output the early warning information to the thermal control system autonomous management platform.

[0120] refer to Figure 13In one specific embodiment of step S70, the early warning and decision-making module 6 determines the early warning level based on the remaining service life prediction result, the pump efficiency decay prediction result, and the composite health index threshold status. When the remaining service life prediction result is less than or equal to the first early warning threshold and the pump efficiency decay prediction result is greater than or equal to the preset decay threshold, a level one early warning message is generated; when the remaining service life prediction result is less than or equal to the second early warning threshold, a level two early warning message is generated; and when the composite health index reaches the preset failure threshold, an emergency alarm message is generated. Early warning messages of different levels are output to the thermal control system autonomous management platform 7 via the spacecraft data bus. Step S70 may include steps S710 to S740.

[0121] Step S710: When the remaining service life prediction result is less than or equal to the first warning threshold and the pump efficiency decay prediction result is greater than or equal to the preset decay threshold, a first-level warning message is generated.

[0122] In one specific embodiment, the first warning threshold can be 120 hours, and the preset attenuation threshold can be 15%. When the predicted RUL is less than or equal to 120 hours and the predicted pump efficiency attenuation is greater than or equal to 15%, a Level 1 warning message can be generated. The Level 1 warning message may include the thermal control pump number, the current CHI, the estimated RUL point value, the RUL confidence interval, the predicted pump efficiency attenuation, and a prompt to start the backup pump and arrange maintenance.

[0123] Step S720: When the remaining service life prediction result is less than or equal to the second warning threshold, generate a secondary warning message.

[0124] In one specific embodiment, the second warning threshold can be 72 hours. When the predicted RUL is less than or equal to 72 hours, a secondary warning message can be generated. The secondary warning message may include prompts suggesting an immediate switch to the backup system, reducing the current pump load, or performing a thermal control loop reconfiguration.

[0125] Step S730: When the composite health indicators reach the preset failure threshold, an emergency alarm message is generated.

[0126] When the CHI reaches or exceeds the TH, it can be determined that the spacecraft's thermal control pump has reached or is close to a preset failure state, generating an emergency alarm message. The emergency alarm message may include the failure risk level, a summary of the current three-source sensor data, a predicted cause of failure, and a prompt to immediately perform a backup switch or autonomous reconfiguration of the thermal control system.

[0127] Step S740: The first-level warning information, second-level warning information, or emergency alarm information is output to the thermal control system autonomous management platform through the spacecraft data bus to generate warning information.

[0128] In one specific implementation, the early warning information can be output via the spacecraft data bus, health management communication interface, or ground telemetry link. The autonomous management platform 7 of the thermal control system can perform backup pump switching, thermal control loop reconfiguration, load adjustment, or maintenance plan generation based on the early warning information.

[0129] Through the above steps S10 to S70, this embodiment integrates vibration spectrum data, infrared thermal imaging data, and fluid pressure drop data into a CHI that can characterize the overall degradation level of the spacecraft's thermal control pump. This allows mechanical degradation, thermal degradation, sealing degradation, and hydraulic performance degradation to all participate in degradation modeling, reducing the risk of missed detection caused by insufficient characterization capability of a single signal under complex operating conditions.

[0130] In the construction and particle filter update of the three-parameter Weibull degradation model, this embodiment uses shape, scale, and position parameters as state variables and CHI as an observation. This approach couples the multi-source degradation sensing results of the heat-controlled pump with the degradation model parameter update, enabling the model parameters to adaptively change based on real-time monitoring data. The position parameter is used to characterize the time offset between the stable operation phase and the degradation phase, avoiding misfitting early stable operation data as a continuous degradation process.

[0131] In the remaining useful life (RUL) prediction process, this embodiment predicts multiple degradation trajectories based on the updated posterior distribution of parameters, and determines the RUL distribution using the first-pass time when the CHI first reaches the preset failure threshold. Compared with methods that only output a single failure time, this method can simultaneously provide point estimates and confidence intervals, providing decision-making basis with uncertain information for the autonomous management platform of the thermal control system.

[0132] In the early warning decision-making process, this embodiment combines the RUL prediction results, pump efficiency degradation prediction results, and CHI failure threshold to form a first-level early warning, a second-level early warning, and an emergency alarm. This graded early warning result can be linked with the backup pump switching, thermal control loop reconstruction, or maintenance decisions of the spacecraft thermal control system, thereby forming a closed-loop processing flow from degradation perception to lifetime prediction and then to early warning output.

[0133] refer to Figure 14 In one experimental embodiment, a spacecraft circulating pump was used as the test object. The pump had a rated flow rate of 10 L / min and a rated head of 20 m. The test conditions simulated on-orbit operation, with a rotation speed of 3000 rpm, a 50% ethylene glycol aqueous solution as the medium, and a temperature range of -10°C to +50°C. Degradation injection methods included progressive bearing wear and impeller corrosion to simulate the natural degradation process. Figure 6The horizontal axis represents the actual lifetime, the vertical axis represents the predicted lifetime, and the diagonal represents the ideal prediction line y=x. The experimental data points are distributed near the ideal prediction line, indicating good consistency between the predicted and actual lifetimes. Under these experimental conditions, at a 95% confidence level, the RUL prediction error based on three-source data fusion is 17.3%, meeting the requirement of no more than 20%. Compared to the prediction method using only vibration data, which has a RUL prediction error of 28.6%, the RUL prediction error after three-source fusion is reduced to 17.3%, improving prediction accuracy by 39.5%.

[0134] refer to Figure 15 In one experimental embodiment, Figure 15 This can represent the relationship between the pump efficiency attenuation prediction curve and the warning point. Figure 15 The horizontal axis represents operating time, the vertical axis represents pump efficiency degradation or composite health indicators, and the curve represents the predicted trajectory of pump efficiency degradation. The preset degradation threshold is 15%. In the experiment, the system was able to provide a lifespan risk warning on average 128 hours before the pump efficiency degradation reached 15%; at the early warning point of RUL=120h, the actual pump efficiency degradation was approximately 12.5%. When the predicted RUL result is less than or equal to the first warning threshold and the predicted pump efficiency degradation result is greater than or equal to the preset degradation threshold, the warning and decision module 6 generates a first-level warning message. The above experimental results show that the warning information can reserve a safety margin for subsequent backup pump switching, maintenance arrangements, or thermal control system reconfiguration.

[0135] In an optional embodiment, the feature fusion and modeling module 4 can construct a CHI using a trained feature fusion model. Specifically, a three-layer fully connected neural network can be constructed, with the input layer including 12 neurons, the hidden layer including 64 neurons, and the output layer including 1 neuron. The input can be vibration degradation features, thermal image degradation features, and pressure drop degradation features, and the supervision label can be the pump efficiency attenuation rate. After training, the feature fusion model outputs a fused degradation representation and constructs a CHI based on the fused degradation representation. In a specific experiment, when the sample size is greater than 1000 groups, the RUL prediction error using the deep learning fusion method can be reduced to 15.1%. This method has high computational complexity and can be accelerated using a GPU (Graphics Processing Unit). In scenarios with limited onboard computing resources, principal component dimensionality reduction and correlation weighted fusion methods can be prioritized; in ground-based health management platforms or scenarios with sufficient computing resources, this deep learning fusion method can be used.

[0136] In another alternative embodiment, the acquisition method of multi-source operational data can be configured according to the type of spacecraft thermal control pump. For small or micro pumps, such as pumps with a flow rate of less than 2 L / min, due to their compact structure and limited installation space, a configuration of single-point vibration monitoring, infrared thermal imaging monitoring, and fluid pressure drop monitoring can be adopted. The vibration sensor 11 can be installed on the motor end cover, the infrared thermal imager 12 can be focused on the connection between the pump body and the motor or the mechanical seal area, and the pressure sensor 13 can collect the inlet and outlet pressures respectively. Under experimental conditions, the RUL prediction error of this configuration of single-point vibration monitoring, infrared thermal imaging monitoring, and fluid pressure drop monitoring is 19.8%, which still meets the requirement of a prediction error of no more than 20%.

[0137] In another optional embodiment, for large circulating pumps, such as pumps with a flow rate greater than 50 L / min, bearing temperature data can be further collected on top of vibration spectrum data, infrared thermographic data, and fluid pressure drop data to form a four-source monitoring system of vibration, thermography, pressure drop, and temperature. Bearing temperature data can be collected by a temperature sensor to supplement local bearing heating information. The feature fusion and modeling module 4 can use bearing temperature features, vibration degradation features, thermographic degradation features, and pressure drop degradation features together for CHI construction. Under experimental conditions, the RUL prediction error of the four-source configuration is 14.5%, which is better than the prediction results of the three-source configuration. The above-mentioned small pump and large circulating pump configurations are optional sensor arrangements based on the pump type, without changing the core technical path of this invention based on multi-source data fusion, CHI construction, three-parameter Weibull degradation modeling, and particle filter updates.

[0138] In one system embodiment, the multi-source data acquisition module 1 is used to acquire multi-source operational data of the spacecraft's thermal control pump. The multi-source operational data includes vibration spectrum data, infrared thermal image data, and fluid pressure drop data. The vibration spectrum data is used to characterize the mechanical degradation of the spacecraft's thermal control pump, the infrared thermal image data is used to characterize the thermal degradation or sealing degradation of the spacecraft's thermal control pump, and the fluid pressure drop data is used to characterize the hydraulic performance degradation of the spacecraft's thermal control pump. The multi-source data acquisition module 1 may include a vibration sensor 11, an infrared thermal imager 12, and a pressure sensor 13.

[0139] Data preprocessing module 2 is used to preprocess vibration spectrum data, infrared thermal image data, and fluid pressure drop data respectively, to obtain time-aligned vibration preprocessed data, thermal image preprocessed data, and pressure drop preprocessed data. Data preprocessing module 2 can perform denoising, resampling, and normalization on vibration spectrum data; image correction, spatial denoising, and pump body region of interest extraction on infrared thermal image data; denoising, operating condition compensation, and normalization on fluid pressure drop data; and time synchronization or time alignment on the three-source preprocessed data.

[0140] The degradation feature extraction module 3 is used to extract vibration degradation features from vibration preprocessing data, thermal image degradation features from thermal image preprocessing data, and pressure drop degradation features from pressure drop preprocessing data. The degradation feature extraction module 3 can output features such as vibration root mean square value, peak factor, spectral entropy, mean temperature, temperature variance, hot spot temperature, inlet and outlet pressure difference, and pressure drop change rate.

[0141] Feature fusion and modeling module 4 includes a feature fusion module and a degradation model construction module. The feature fusion module is used to construct a composite health index characterizing the overall degradation degree of the spacecraft's thermal control pump based on vibration degradation features, thermal imaging degradation features, and pressure drop degradation features. The degradation model construction module is used to construct a three-parameter Weibull degradation model based on the composite health index. The three-parameter Weibull degradation model includes shape parameters, scale parameters, and position parameters. The position parameters are used to characterize the degradation onset time or fault-free operation time of the spacecraft's thermal control pump from a healthy state to a degraded state.

[0142] Model update and prediction module 5 includes a model update module and a lifetime prediction module. The model update module is used to form the state vector of the particle filter by combining the shape parameters, scale parameters, and position parameters, and to use the composite health index as the observation of the particle filter. It establishes a nonlinear observation relationship between the composite health index and the degradation output of the three-parameter Weibull degradation model, and adaptively updates the parameters of the three-parameter Weibull degradation model based on the nonlinear observation relationship to obtain the updated parameter posterior distribution. The lifetime prediction module is used to determine the time distribution of the composite health index first reaching the preset failure threshold based on the updated parameter posterior distribution and the preset failure threshold, and to determine the remaining lifetime prediction result of the spacecraft heat control pump based on the time distribution.

[0143] The early warning and decision-making module 6 includes an early warning and decision-making module. This module generates early warning information based on the predicted remaining service life and / or pump efficiency degradation, and outputs the early warning information to the thermal control system autonomous management platform 7. The early warning and decision-making module can generate first-level early warning information, second-level early warning information, and emergency alarm information.

[0144] The modules in the aforementioned system can be implemented by software programs, hardware circuits, or a combination of both. This system can be integrated into the spacecraft's thermal control system health management unit, spacecraft integrated electronic equipment, ground health management platform, edge computing devices, or a combination of these. Modules can exchange data via the spacecraft data bus, internal high-speed bus, network communication interface, or shared storage area.

[0145] The specific values, equipment models, model parameters, sampling frequency, measurement accuracy, analysis frequency band, installation distance, threshold, number of particles, neural network structure, number of samples, confidence interval, error result, improvement ratio, and early warning time mentioned above are all examples under specific embodiments or experimental conditions, used to illustrate the feasible implementation methods and technical effects of the present invention, and should not be construed as limiting the scope of protection of the present invention.

[0146] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, or combinations made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for spacecraft thermal control pump degradation sensing and remaining lifetime prediction based on multi-source data fusion, characterized in that, include: Multi-source operational data of the spacecraft's thermal control pump are collected. The multi-source operational data includes vibration spectrum data, infrared thermographic data, and fluid pressure drop data. The vibration spectrum data is used to characterize the mechanical degradation of the spacecraft's thermal control pump, the infrared thermographic data is used to characterize the thermal degradation or sealing degradation of the spacecraft's thermal control pump, and the fluid pressure drop data is used to characterize the hydraulic performance degradation of the spacecraft's thermal control pump. The vibration spectrum data, infrared thermal image data, and fluid pressure drop data are preprocessed respectively to obtain time-aligned vibration preprocessed data, thermal image preprocessed data, and pressure drop preprocessed data; Vibration degradation features are extracted from the vibration preprocessing data, thermal image degradation features are extracted from the thermal image preprocessing data, and pressure drop degradation features are extracted from the pressure drop preprocessing data. A composite health index characterizing the overall degradation degree of the spacecraft's thermal control pump is constructed based on the vibration degradation features, thermal image degradation features, and pressure drop degradation features. A three-parameter Weibull degradation model is constructed based on the composite health index. The three-parameter Weibull degradation model includes shape parameters, scale parameters, and position parameters. The position parameters are used to characterize the degradation start time or fault-free operation time of the spacecraft's thermal control pump from a healthy state to a degraded state. The shape parameter, scale parameter, and position parameter are used to form the state vector of the particle filter. The composite health index is used as the observation of the particle filter. A nonlinear observation relationship is established between the composite health index and the output degradation of the three-parameter Weibull degradation model. Based on the nonlinear observation relationship, the parameters of the three-parameter Weibull degradation model are adaptively updated to obtain the updated parameter posterior distribution. Based on the updated parameter posterior distribution and the preset failure threshold, the time distribution of the composite health index reaching the preset failure threshold for the first time is determined, and the remaining service life prediction result of the spacecraft heat control pump is determined based on the time distribution. Early warning information is generated based on the predicted remaining service life and / or predicted pump efficiency decline, and the early warning information is output to the thermal control system autonomous management platform.

2. The method according to claim 1, characterized in that, Collect multi-source operational data of the spacecraft's heat control pump, including: The vibration spectrum data is collected by vibration sensors installed at the bearing housing of the spacecraft's heat control pump, the motor end, or the bearing housing of the drive motor. The vibration spectrum data is used to characterize mechanical degradation caused by bearing wear, rotor imbalance, mechanical loosening, or impeller abnormalities. The infrared thermal image data is acquired by an infrared thermal imager facing the surface of the pump body, the area where the pump head is connected to the motor, or the mechanical seal area. The infrared thermal image data is used to characterize thermal degradation or seal degradation caused by frictional heat generation, local overheating, seal wear, or leakage. Pressure sensors installed on the inlet and outlet pipes of the spacecraft's thermal control pump collect inlet and outlet pressure data, respectively. Based on the inlet and outlet pressure data, fluid pressure drop data is obtained. The fluid pressure drop data is used to characterize hydraulic performance degradation caused by cavitation, blockage, impeller wear, or decreased hydraulic efficiency.

3. The method according to claim 1, characterized in that, Vibration degradation features are extracted from the vibration preprocessing data, thermal image degradation features are extracted from the thermal image preprocessing data, and pressure drop degradation features are extracted from the pressure drop preprocessing data, including: The vibration root mean square value, peak factor and spectral entropy are extracted from the vibration preprocessing data to obtain the vibration degradation characteristics. The vibration root mean square value is used to characterize the vibration energy change, the peak factor is used to characterize the impact component change, and the spectral entropy is used to characterize the spectral complexity change. Based on the thermal image preprocessing data, the mean temperature, temperature variance, and hot spot temperature of the region of interest in the pump body are extracted to obtain the thermal image degradation features. The mean temperature is used to characterize the overall thermal state, the temperature variance is used to characterize the non-uniformity of the temperature field, and the hot spot temperature is used to characterize the risk of local overheating. Based on the pressure drop preprocessing data, the inlet and outlet pressure difference and pressure drop change rate are extracted to obtain the pressure drop degradation characteristics. The inlet and outlet pressure difference is used to characterize the hydraulic output state, and the pressure drop change rate is used to characterize the hydraulic performance degradation rate.

4. The method according to claim 3, characterized in that, Based on the aforementioned vibration degradation characteristics, thermal imaging degradation characteristics, and pressure drop degradation characteristics, a composite health index characterizing the overall degradation degree of spacecraft thermal control pumps is constructed, including: The vibration degradation features, thermal imaging degradation features, and pressure drop degradation features are standardized respectively to obtain standardized vibration features, standardized thermal imaging features, and standardized pressure drop features; Principal component dimensionality reduction was performed on the standardized vibration features, standardized thermal imaging features, and standardized pressure drop features respectively to obtain the vibration principal component score, thermal imaging principal component score, and pressure drop principal component score. The correlation coefficients between the vibration principal component score, thermal image principal component score, and pressure drop principal component score and the pump efficiency attenuation label are calculated respectively, and the absolute values ​​of the correlation coefficients are normalized to obtain the vibration weight, thermal image weight, and pressure drop weight. The composite health index is constructed by weighting and fusing the vibration principal component score, thermal image principal component score, pressure drop principal component score, vibration weight, thermal image weight, and pressure drop weight.

5. The method according to claim 1, characterized in that, A three-parameter Weibull degradation model is constructed based on the aforementioned composite health indicators, including: The aforementioned composite health index is used as the comprehensive degradation amount of the spacecraft's thermal control pump; The shape parameter characterizes the degradation rate as a function of operating time, the scale parameter characterizes the characteristic lifetime of the spacecraft thermal control pump, and the position parameter characterizes the time offset of the spacecraft thermal control pump from entering the degradation phase after the stable operation phase. Based on the comprehensive degradation amount, shape parameter, scale parameter, and position parameter, the three-parameter Weibull degradation model for describing the degradation process of the spacecraft's heat control pump is constructed.

6. The method according to claim 5, characterized in that, The shape parameter, scale parameter, and position parameter are used to form the state vector of the particle filter. The composite health index is used as the observation of the particle filter. A nonlinear observation relationship is established between the composite health index and the degradation output of the three-parameter Weibull degradation model. Based on the nonlinear observation relationship, the parameters of the three-parameter Weibull degradation model are adaptively updated to obtain the updated parameter posterior distribution, including: The shape parameters, scale parameters, and position parameters are combined to form a state vector; The state transition equation of the state vector is constructed based on the random walk law of the parameters, and the particle state is predicted based on the state transition equation. Using the degradation amount output by the three-parameter Weibull degradation model at the current moment as the predicted observation value, a nonlinear observation relationship is established between the predicted observation value and the real-time obtained composite health index. The observation likelihood of each particle is calculated based on the nonlinear observation relationship, and the particle weights are updated based on the observation likelihood. Resampling is performed based on the updated particle weights to obtain the updated posterior distribution of parameters.

7. The method according to claim 6, characterized in that, Based on the updated posterior distribution of the parameters and the preset failure threshold, the time distribution at which the composite health index first reaches the preset failure threshold is determined, and the remaining service life prediction result of the spacecraft heat control pump is determined based on the time distribution, including: Based on the updated parameter posterior distribution, multiple degradation trajectories are predicted to obtain the predicted trajectory distribution of the composite health index. The first-pass time when the composite health index first reaches the preset failure threshold in each degradation trajectory is determined based on the predicted trajectory distribution. The distribution of first-wearing times is determined based on multiple first-wearing times; The remaining service life distribution is determined based on the difference between the first penetration time distribution and the current time. Based on the remaining useful life distribution, the estimated value of the remaining useful life and the confidence interval are determined to obtain the remaining useful life prediction result.

8. The method according to claim 7, characterized in that, Based on the predicted remaining service life and / or predicted pump efficiency decline, an early warning message is generated and output to the thermal control system autonomous management platform, including: When the predicted remaining service life is less than or equal to the first warning threshold and the predicted pump efficiency decay is greater than or equal to the preset decay threshold, a first-level warning message is generated. When the predicted remaining useful life is less than or equal to the second warning threshold, a secondary warning message is generated; When the composite health indicator reaches the preset failure threshold, an emergency alarm message is generated; The first-level warning information, second-level warning information, or emergency alarm information are output to the autonomous management platform of the thermal control system through the spacecraft data bus to generate the warning information.

9. The method according to claim 2, characterized in that, The vibration spectrum data, infrared thermal image data, and fluid pressure drop data are preprocessed respectively to obtain time-aligned preprocessed vibration data, preprocessed thermal image data, and preprocessed pressure drop data, including: The vibration spectrum data is denoised, resampled, and normalized to obtain the vibration preprocessed data; The infrared thermal image data is subjected to image correction, spatial denoising, and pump body region of interest extraction to obtain the thermal image preprocessing data; The fluid pressure drop data is subjected to noise reduction, operating condition compensation and normalization to obtain the pressure drop preprocessed data; The vibration preprocessing data, thermal imaging preprocessing data, and voltage drop preprocessing data are time-synchronized or time-aligned to obtain time-aligned vibration preprocessing data, thermal imaging preprocessing data, and voltage drop preprocessing data.

10. A spacecraft thermal control pump degradation sensing and remaining lifetime prediction system based on multi-source data fusion, characterized in that, include: A multi-source data acquisition module is used to acquire multi-source operating data of the spacecraft's thermal control pump. The multi-source operating data includes vibration spectrum data, infrared thermographic data, and fluid pressure drop data. The vibration spectrum data is used to characterize the mechanical degradation of the spacecraft's thermal control pump, the infrared thermographic data is used to characterize the thermal degradation or sealing degradation of the spacecraft's thermal control pump, and the fluid pressure drop data is used to characterize the hydraulic performance degradation of the spacecraft's thermal control pump. The data preprocessing module is used to preprocess the vibration spectrum data, infrared thermal image data and fluid pressure drop data respectively to obtain time-aligned vibration preprocessed data, thermal image preprocessed data and pressure drop preprocessed data; The degradation feature extraction module is used to extract vibration degradation features based on the vibration preprocessing data, extract thermal image degradation features based on the thermal image preprocessing data, and extract pressure drop degradation features based on the pressure drop preprocessing data. The feature fusion module is used to construct a composite health index characterizing the overall degradation degree of the spacecraft's thermal control pump based on the vibration degradation features, thermal imaging degradation features, and pressure drop degradation features. The degradation model construction module is used to construct a three-parameter Weibull degradation model based on the composite health index. The three-parameter Weibull degradation model includes shape parameters, scale parameters, and position parameters. The position parameters are used to characterize the degradation start time or fault-free operation time of the spacecraft's heat control pump from a healthy state to a degraded state. The model update module is used to form the state vector of the particle filter by the shape parameter, scale parameter and position parameter, use the composite health index as the observation of the particle filter, establish a nonlinear observation relationship between the composite health index and the output degradation of the three-parameter Weibull degradation model, and adaptively update the parameters of the three-parameter Weibull degradation model based on the nonlinear observation relationship to obtain the updated parameter posterior distribution. The lifespan prediction module is used to determine the time distribution of the first time the composite health index reaches the preset failure threshold based on the updated parameter posterior distribution and the preset failure threshold, and to determine the remaining lifespan prediction result of the spacecraft heat control pump based on the time distribution. The early warning decision module is used to generate early warning information based on the remaining service life prediction results and / or pump efficiency decay prediction results, and output the early warning information to the thermal control system autonomous management platform.

Images