A method for predicting the remaining life of a space mechanism

Abstract

The present invention relates to the technical field of on-orbit space mechanisms, and discloses a method for predicting the remaining life of a space mechanism, specifically as follows: 1) Determine the life index of the space mechanism and the key components affecting the life index; 2) Establish a relationship equation between the performance loss rate of the key components and the service time; 3) Substitute the performance degradation data obtained from the key component relationship equation into the digital twin model of the space mechanism to obtain a data set X of the life index data of the space mechanism at different service times; 4) Calculate the health index HI of the space mechanism at different service times according to the data set X to obtain a data set of the health index HI; 5) Train the bidirectional LSTM network model using the data set of HI; 6) Collect the real-time life index data of the space mechanism in the service state, process the real-time life index data according to step S4 to obtain the current real-time HI; input the real-time HI into the trained bidirectional LSTM network model for prediction, construct an HI prediction curve according to the prediction result, and calculate the remaining useful life RUL of the space mechanism according to the HI prediction curve.

Description

Technical Field

[0001] The present invention relates to the technical field of in-orbit space mechanisms, and particularly to a method for predicting the remaining life of a space mechanism. Background Art

[0002] The safety and reliability of in-orbit space mechanisms are necessary conditions to ensure the successful completion of their missions. In-orbit space mechanisms serve in complex and harsh space environments, and are affected by various factors during operation, resulting in changes in their performance and functions, and prone to failures. Therefore, predicting the remaining service life of in-orbit space mechanisms, reducing the failure of space missions caused by space mechanism failures, and avoiding major accidents. As products with long service life and high reliability, the failure data of space mechanisms are extremely scarce, so predictions are mainly made through their performance degradation data.

[0003] With the development of multi-field modeling and new information technologies, digital twin has become a research hotspot in the field of operation and maintenance of complex equipment systems. Digital twin can construct a model in the digital space that is in one-to-one real-time mapping with the physical entity, and can realize real-time monitoring of complex systems, detecting performance degradation, and discovering fault symptoms, thereby effectively predicting the faults and remaining life of the system. Due to the high cost and long cycle of space mechanism fault tests, it is necessary to build a digital space twin of the space mechanism by means of digital twin technology, and through the virtual operation of the twin, provide data samples for life prediction. Summary of the Invention

[0004] Aiming at the problems and deficiencies in the prior art, the purpose of the present invention is to provide a method for predicting the remaining life of a space mechanism.

[0005] To achieve the purpose of the invention, the technical solution adopted by the present invention is as follows:

[0006] A method for predicting the remaining life of a space mechanism, comprising the following steps:

[0007] S1: Determine the life index of the space mechanism, and determine the key components affecting the life index of the space mechanism according to the working principle of the space mechanism;

[0008] S2: Conduct an accelerated performance degradation experiment on the key components, analyze the performance degradation law of the key components during service, and establish a relationship equation between the performance loss rate of the key components and the service time;

[0009] S3: Based on the physical entity of the space mechanism, establish a digital twin model of the space mechanism; according to the relationship equation constructed in step S2, calculate the performance degradation data of the key components at different service times, and input the performance degradation data into the digital twin model for virtual operation to obtain a life index data set X of the space mechanism at different service times, X = {x 1,x 2 ,x 3 ,…x i}, x i represents the life index data of key components at any service time;

[0010] S4: Normalize the life index data x i and input it into the DBN network (Deep Belief Network) for feature extraction to obtain the life feature data f of the life index i ; At the same time, input the life index threshold of the aerospace mechanism into the DBN network for feature extraction to obtain the life failure feature data f of the life index fault ; According to the life feature data f i and the life failure feature data f fault , calculate the health index HI of the aerospace mechanism at different service times to obtain the health index HI dataset of the aerospace mechanism at different service times;

[0011] The calculation formula of the health index HI is as follows:

[0012]

[0013] S5: Use the health index HI dataset to train the bidirectional LSTM network model, so that after inputting the health index HI at any service time in the health index HI dataset into the trained bidirectional LSTM network model, the trained bidirectional LSTM network model can output the predicted value of the health index HI at the next service time with a preset accuracy;

[0014] S6: Collect the real-time life index data of the aerospace mechanism in the service state, process the real-time life index data according to the operation in step S4 to obtain the real-time health index HI at the current moment; Input the real-time health index HI at the current moment into the trained bidirectional LSTM network model for prediction to obtain the predicted value of the health index HI at the next moment, and then sequentially input the predicted value of the health index HI at the next moment into the trained bidirectional LSTM network model for rolling prediction to obtain the predicted curve of the health index HI of the key components. When the predicted curve of the health index HI reaches the set threshold, stop the prediction and calculate the remaining useful life RUL of the aerospace mechanism; The calculation formula of the remaining useful life RUL is as follows:

[0015] RUL = T ythreshold -T y0

[0016] where, y threshold is the threshold of the health index HI, and y 0 is the value of the real-time health index HI at the current moment; T ythresholdLet \(T\) be the time corresponding to the predicted value of the health index \(HI\) reaching the set threshold. y0 Let \(t\) be the current running time.

[0017] According to the above-mentioned method for predicting the remaining life of the aerospace mechanism, preferably, the accelerated performance degradation experiment in step S2 is a stress relaxation test, and the stress is temperature.

[0018] According to the above-mentioned method for predicting the remaining life of the aerospace mechanism, preferably, the relationship equation between the performance loss rate of the key component and the service time in step S2 is:

[0019] \(Y = A + B\ln t\);

[0020] where \(Y\) is the performance loss rate of the key component, \(A\) is a constant, \(B\) is the stress relaxation rate of the key component, \(\ln\) represents the natural logarithm with the base of the natural constant \(e\), and \(t\) is the service time of the key component.

[0021] According to the above-mentioned method for predicting the remaining life of the aerospace mechanism, preferably, the performance degradation data in step S3 is the difference between the initial state parameters of the key component and the performance degradation amount, where the performance degradation amount is the product of the performance loss rate of the key component under the service time and the initial state parameters of the key component.

[0022] According to the above-mentioned method for predicting the remaining life of the aerospace mechanism, preferably, the calculation formula for the normalization process in step S4 is:

[0023]

[0024] where \(x\) max is the maximum value in the aerospace life index dataset, \(x\) min is the minimum value in the life index dataset, and \(X\) is the life index data after normalization.

[0025] According to the above-mentioned method for predicting the remaining life of the aerospace mechanism, preferably, the life index in step S1 is a monitorable parameter that affects the normal service of the aerospace mechanism.

[0026] According to the above-mentioned method for predicting the remaining life of the aerospace mechanism, preferably, in step S1, according to the working principle of the aerospace mechanism, a three-level fault transfer relationship of component - part - system is established to determine the key components that affect the life index of the aerospace mechanism.

[0027] According to the above-mentioned method for predicting the remaining life of the aerospace mechanism, preferably, when there are multiple life indexes, the life index that first reaches the life limit is used as the basis for judging the remaining life of the aerospace mechanism.

[0028] According to the above-mentioned method for predicting the remaining life of a space agency, preferably, in order to quantify the uncertainty of the prediction result, the MC dropout method is used to obtain the life prediction interval of the space agency at a 95% confidence level, that is, the value range of the life. Based on the prediction model, the same group of data is predicted 100 times respectively to obtain the prediction interval at a 95% confidence level, depicting the uncertainty of the prediction result, which is more in line with the actual situation.

[0029] Compared with the prior art, the positive and beneficial effects achieved by the present invention are as follows:

[0030] (1) By establishing the relationship between the performance degradation law of key components and the system life index through the device digital twin, the prediction method of the present invention can obtain a large amount of device performance degradation data, reduce the dependence on tests, and solve the problem of insufficient performance degradation data of high-reliability products.

[0031] (2) Regarding the data obtained by the twin as the source domain, based on the deep transfer learning method, the prediction model trained in the source domain is transferred to the actual working process of the device, and the real-time data is regarded as the target domain to realize the life prediction of the on-orbit space agency.

[0032] (3) Compared with the traditional prediction method, the present invention uses MC dropout to quantify the uncertainty of the prediction result, and the prediction result is more in line with the actual situation. Description of the Drawings

[0033] Figure 1 It is a flowchart of the method for predicting the remaining life of the space agency of the present invention. Detailed Embodiments

[0034] The following further describes the present invention in detail through specific embodiments, but it is not intended to limit the protection scope of the present invention.

[0035] Embodiment 1:

[0036] A method for predicting the remaining life of a space agency (as Figure 1 shown) includes the following steps:

[0037] S1: Based on the historical operation situation and working principle of the space agency, clarify the monitorable parameters that affect the normal service of the space agency and determine them as the life indexes of the space agency; further, based on the working mechanism of the space agency, establish the fault transfer relationship of three levels of component - part - system, and determine the key components that affect the life indexes of the mechanism.

[0038] S2: Conduct an accelerated performance degradation experiment on the key component, analyze the performance degradation law of the key component during service, and establish a relationship equation between the performance loss rate of the key component and the service time. Preferably, the accelerated performance degradation experiment is a stress relaxation test, and the stress is temperature; the relationship equation between the performance loss rate of the key component and the service time is:

[0039] Y = A + Blnt;

[0040] Y is the performance loss rate of the key component, A is a constant, B is the stress relaxation rate of the key component, ln represents the natural logarithm with the natural constant e as the base, and t is the service time of the key component.

[0041] S3: Based on the physical entity of the space mechanism, establish a digital twin model of the space mechanism, realize the two-way mapping between the physical and digital twin models, and correct the twin model of the spacecraft through the historical operation data of the spacecraft. According to the relationship equation constructed in step S2, calculate the performance degradation data of the key component at different service times, and input the performance degradation data into the digital twin model for virtual operation to obtain the life index data set X of the space mechanism at different service times, X = {x 1 , x 2 , x 3 , … x i}, where x i represents the life index data of the key component at any service time. Among them, the performance degradation data is the difference between the initial state parameter of the key component at the service time and the performance degradation amount, and the performance degradation amount is the product of the performance loss rate of the key component and the initial state parameter of the key component.

[0042] S4: Normalize the life index data x i and input it into the DBN network (Deep Belief Network) for feature extraction to obtain the life feature data f i of the life index. The calculation formula for the normalization process is:

[0043]

[0044] where x max is the maximum value in the life index data set of the spacecraft, x min is the minimum value in the life index data set, and X is the life index data after normalization.

[0045] At the same time, input the life index threshold of the space mechanism into the DBN network for feature extraction to obtain the life failure feature data f fault of the life index; according to the life feature data f i and the life failure feature data f fault, calculate the health index HI of the space agency at different service times to obtain the health index HI dataset of the space agency at different service times;

[0046] The calculation formula of the health index HI is as follows:

[0047]

[0048] S5: Use the health index HI dataset to train the bidirectional LSTM network model, so that after inputting the health index HI at any service time in the health index HI dataset into the trained bidirectional LSTM network model, the trained bidirectional LSTM network model can output the predicted value of the health index HI at the next service time with a preset accuracy.

[0049] S6: Collect the real-time life index data of the space agency in the service state of the space agency, normalize the real-time life index data according to the operation in step S4, and then input it into the DBN network (deep belief network) for feature extraction to obtain the life feature data f of the life index i , and calculate according to the calculation formula of the health index HI to obtain the real-time health index HI at the current moment.

[0050] Input the real-time health index HI at the current moment into the trained bidirectional LSTM network model for prediction to obtain the predicted value of the health index HI at the next moment, and then sequentially input the predicted value of the health index HI at the next moment into the trained bidirectional LSTM network model for rolling prediction to obtain the predicted values of the health index HI at different moments. According to the relationship between the predicted value of the health index HI and the corresponding time, make the predicted curve of the health index HI of the key component. When the predicted curve of the health index HI reaches the set threshold, stop the prediction and calculate the remaining useful life RUL of the space agency; the calculation formula of the remaining useful life RUL is as follows:

[0051] RUL = T ythreshold -T y0

[0052] where, y threshold is the threshold of the health index HI, y 0 is the value of the real-time health index HI at the current moment; T ythreshold is the time corresponding to when the predicted value of the health index HI reaches the set threshold, and T y0 is the current running time.

[0053] Furthermore, in order to quantify the uncertainty of the prediction results, the MC dropout method is used to obtain the life prediction interval of the space agency at the 95% confidence level, that is, the value range of the life. Based on the prediction model, the same group of data is predicted 100 times respectively to obtain its prediction interval at the 95% confidence level, depicting the uncertainty of the prediction results, which is more in line with the actual situation.

[0054] The above are only the preferred embodiments of the present invention, but are not limited to the above examples. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the remaining life of a space mechanism, characterized in that, it includes the following steps: S1: Determine the life index of the space mechanism, and determine the key components that affect the life index of the space mechanism according to the working principle of the space mechanism; S2: Conduct an accelerated performance degradation experiment on the key components, analyze the performance degradation law of the key components during service, and establish a relationship equation between the performance loss rate of the key components and the service time; S3: Based on the physical entities of the space agency, establish a digital twin model of the space agency; according to the relationship equation constructed in step S2, calculate the performance degradation data of key components at different service times, and input the performance degradation data into the digital twin model for virtual operation to obtain the life index data set X of the space agency at different service times, X = {x 1 , x 2 , x 3 , … x i}, where x i represents the life index data of the space agency at any service time; S4: Normalize the life index data x i and input it into the DBN network for feature extraction to obtain the life feature data f of the life index i ; At the same time, input the life index threshold of the aerospace mechanism into the DBN network for feature extraction to obtain the life failure feature data f of the life index fault ; According to the life feature data f i and the life failure feature data f fault , calculate the health index HI of the aerospace mechanism at different service times to obtain the health index HI data set of the aerospace mechanism at different service times; The calculation formula of the health index HI is as follows: S5: Use the health index HI data set to train the bidirectional LSTM network model, so that after inputting the health index HI at any service time in the health index HI data set into the trained bidirectional LSTM network model, the trained bidirectional LSTM network model can output the predicted value of the health index HI at the next service time with a preset accuracy; S6: Collect the real-time life index data of the space mechanism in the service state, process the real-time life index data according to the operation in step S4 to obtain the real-time health index HI at the current moment; input the real-time health index HI at the current moment into the trained bidirectional LSTM network model for prediction to obtain the predicted value of the health index HI at the next moment, and then sequentially input the predicted value of the health index HI at the next moment into the trained bidirectional LSTM network model for rolling prediction to obtain the predicted curve of the health index HI of the key components. When the predicted curve of the health index HI reaches the set threshold, stop the prediction and calculate the remaining life RUL of the space mechanism; the calculation formula of the remaining life RUL is as follows: RUL = T ythreshold -T y0 where y threshold is the threshold value of the health index HI, and y 0 is the value of the real-time health index HI at the current moment; T ythreshold is the time corresponding to when the predicted value of the health indicator HI reaches the set threshold, T y0 is the current running time.

2. The method for predicting the remaining life of a space mechanism according to claim 1, characterized in that, the accelerated performance degradation experiment in step S2 is a stress relaxation test, and the stress is temperature.

3. The method for predicting the remaining life of a space mechanism according to claim 2, characterized in that, the relationship equation between the performance loss rate of the key components and the service time in step S2 is: Y = A + Blnt; Y is the performance loss rate of the key components, A is a constant, B is the stress relaxation rate of the key components, ln represents the natural logarithm with the natural constant e as the base, and t is the service time of the key components.

4. The method for predicting the remaining life of a space mechanism according to any one of claims 1-3, characterized in that, the performance degradation data in step S3 is the difference between the initial state parameters of the key components and the performance degradation amount at the service time, where the performance degradation amount is the product of the performance loss rate of the key components and the initial state parameters of the key components.

5. The method for predicting the remaining life of a space mechanism according to claim 4, characterized in that, the calculation formula of the normalization process in step S4 is: where x max is the maximum value in the dataset of flight life index data, x min is the minimum value in the dataset of life index data, and X is the life index data after normalization processing.

6. The method for predicting the remaining life of a space mechanism according to claim 1, characterized in that, the life index in step S1 is a measurable parameter that affects the normal service of the space mechanism.

Images