CN114925441B - A method for airborne distributed PHM computational modeling - Google Patents

Abstract

The invention discloses an airborne distributed PHM calculation modeling method, which specifically comprises the following steps: modeling an airborne PHM operation platform; PHM calculates task allocation optimization conditions. According to the method, the computing resources are classified and graded, the performance of the computing resources is defined, and a model of the computing resources is constructed; classifying the algorithms by distinguishing online algorithms from offline algorithms and the accuracy and cost of the algorithms, and constructing a PHM algorithm model; the importance of the airborne equipment is divided, the monitoring requirements of the components are determined, and an operation requirement model of the airborne PHM system is constructed corresponding to the corresponding algorithm.

Description

A method for airborne distributed PHM computational modeling

Technical Field

The present invention relates to a modeling method for fault prediction and health management (PHM) calculation, and in particular to an airborne distributed PHM calculation modeling method.

Background technique

With the continuous development of aviation technology, the complexity of aircraft is constantly increasing, and PHM technology has become one of the essential technologies for advanced aircraft. The implementation of PHM technology relies on powerful data processing capabilities. Most of the current PHM algorithms are developed and run based on general-purpose processor platforms. However, the energy and space on the aircraft are limited. The algorithms developed based on the general-purpose processor platform will consume a lot of CPU resources and energy when running, which seriously restricts the implementation of PHM calculations based on general-purpose processors on airborne platforms. With the continuous development of PHM technology, the number of parameters monitored by airborne PHM systems is increasing, the number and type of data that need to be processed by airborne PHM calculations are growing rapidly, and PHM algorithms are becoming more and more complex. Therefore, distributed embedded systems have begun to become the main platform for airborne PHM systems to perform calculations. With the continuous development of distributed system communication technology, the boundaries of data and tasks between airborne devices have been broken, and computing resources and data can be shared between various airborne devices, which provides the possibility for distributed computing. However, the existing distributed embedded system computing tasks for airborne PHM are solidified in the corresponding computing resources, and the running state is fixed, resulting in a serious waste of computing resources.

Optimizing the allocation and scheduling of distributed computing tasks in airborne PHM systems has become a top priority for optimizing the operational capabilities of airborne PHM systems.

Summary of the invention

In view of the problems existing in the prior art, the present invention provides an airborne distributed PHM computational modeling method, which specifically includes the following steps:

Step 1: Modeling the airborne PHM operation platform

STEP 1. Modeling of airborne distributed computing resources

The airborne PHM system consists of many computing resources that can run the PHM algorithm. Assume that a certain airborne PHM system has a total of M computing resources that can be used to run the PHM algorithm.

(1) Computing Resource Classification C _Type

These computing resources are different embedded platforms with different processor architectures. Each computing resource is a homogeneous system or a heterogeneous system. The computing resources are classified according to the embedded platform and system structure used by the computing resources of the airborne PHM system. Assuming that there are n types of computing resources in the whole PHM system, a label is assigned to each type of resource. i = 1, 2, ... n, as shown in Table 1;

Table 1 Classification of airborne PHM computing resources

in, Refers to computing resources of a single ARM architecture, /> Refers to the computing resources of the dual DSP architecture. Refers to the computing resources of DSP+FPGA heterogeneous architecture,/> Refers to the computing resources of ARM+DSP+FPGA heterogeneous architecture; the current trend of embedded platform development is heterogeneous platform. New embedded heterogeneous system architectures often appear under different functional requirements. It is impossible to list all architecture types. The subscript rules here are sorted according to the types of single-core architecture, multi-core homogeneous architecture, simple heterogeneous architecture, and complex heterogeneous architecture. When a new embedded system architecture is applied to the airborne PHM system, the new subscript number can be added to the class code;

(2) Computing Power Classification C _Ability

On the basis of the classification of computing resources, each type of computing resource is graded according to its computing power, and the computing power level of the i-th computing resource is specified as i = 1, 2, ..., M; Assuming that the computing resource with the smallest throughput in the whole PHM system is the computing resource No. 1, then its computing power level is specified/> And stipulate/> is the unit computing power; assuming that the computing resource with the largest throughput in the whole PHM system is the kth computing resource, refer to the definition of unit computing power to define its computing power level/> Then C _Abilitymax is the maximum value of the computing power level of the computing resources in the whole aircraft PHM system; the computing resources are rounded down with reference to the unit computing power to obtain the computing power level corresponding to all PHM algorithm model running platforms of the whole aircraft PHM system/> i = 1, 2, ..., M;

According to Table 1 and the computing power classification, each computing resource in the airborne PHM system can correspond to and i＝1,2,…,n,j＝1,2,…,M, thus we get Table 2;

Table 2 Performance of airborne PHM computing resources

STEP 2. PHM algorithm performance modeling

For a certain airborne component, the operational requirements of the PHM algorithm need to be modeled;

(1) Determine real-time performance: When developing PHM algorithms, PHM engineers divide PHM algorithms into online algorithms and offline algorithms based on the timeliness of the data processed by the algorithm and whether the algorithm needs to use historical data. Algorithm models that process time-sensitive data and do not need to use historical data are online algorithms with real-time performance. Algorithm models that process data that reflects the status parameters of components over a period of time or need to use historical data are offline algorithms that do not have real-time performance.

(2) Determine the computing resource requirements: For the algorithm α that needs to be run on the airborne PHM system, PHM engineers need to test the algorithm on various computing resources and computing resources of various computing power levels according to the algorithm complexity, logical structure, data length, and data cache requirements when developing the PHM algorithm. Based on the test results, determine the algorithm’s performance on various resources. When running on a class resource, at least/> α represents different algorithms, i = 1, 2, ..., n, and computing resources of a certain computing power level can meet the requirements of algorithm operation. Thus, an operation requirement can be obtained for each algorithm, which is shown in Table 3;

Table 3 Algorithm α operation requirements

STEP 3. Component monitoring requirements modeling

Assumption: An aircraft can perform m types of flight missions. The aircraft's onboard PHM system monitors the status of x onboard components. The onboard PHM system obtains its health status parameters to characterize the health status of the components. First, construct parameters to characterize the monitoring requirements of the components.

(1) Component Importance i = 1, 2, ..., x:

The importance of components is divided according to the degree of harm caused by component failure to flight safety. The greater the harm caused by component failure to flight safety, the greater the importance of components. The bigger;

(2) Task importance j = 1, 2, ..., m:

Each flight mission of the aircraft has a number j, j = 1, 2, ..., m, where m is the type of flight mission that the aircraft can perform. The components are divided into task importance according to the harm of component failure to the completion of the flight mission. The greater the harm of component failure to the completion of a certain flight mission, the greater the The larger the size, the more likely it is that the same component will perform different flight missions. Probably the same;

(3) Component health level N _Health (t):

t represents N _Health (t) as a function of time. As time goes by, the performance of the component gradually degrades and N _Health (t) changes;

The failure of airborne components is a gradual degradation process. In the early stage of component degradation, the degradation process is often stable. The components are divided into health levels based on the current assessment of the component health status level. The greater the possibility of component failure, the greater N _Health (t);

(4) Monitoring requirement N _Monitor (t):

The constructed monitoring requirement N _Monitor (t) represents the urgency of a component’s need for monitoring;

STEP 4. Set the PHM algorithm operation mode

For a certain component, engineers need to select several algorithms for it according to its degradation model during the development phase of its PHM algorithm, and use different monitoring sampling rates and fault diagnosis and life prediction algorithms according to different monitoring requirements;

The settings of the algorithm, sampling rate, and monitoring interval follow the following rules:

① When the monitoring demand is low, the selected algorithm has the characteristics of low power consumption and low computing power requirements. The algorithm can be an offline algorithm or an online algorithm. The sampling rate of the sensor is low and the monitoring interval is long. The goal of the algorithm is to monitor and discover the early fault characteristics of the components.

② As the monitoring demand increases, the real-time performance and calculation accuracy of the selected algorithm are improved, with the online algorithm as the main one, the sampling rate of the sensor is increased, the monitoring interval is shortened or even continuous monitoring is adopted; The goal of the algorithm is to accurately identify the accelerated degradation process of the component after the early fault characteristics of the component appear;

③ When the monitoring demand continues to increase, an algorithm with high precision and real-time characteristics should be selected, the sensor sampling rate should be the highest, and the monitoring method should be real-time monitoring; the purpose of the algorithm is to accurately predict the remaining service life of the component during the accelerated degradation stage of the component, and to be able to immediately and accurately identify and locate the fault when the component fails;

Among all x airborne components, the highest monitoring requirement is The PHM developer finds a PHM algorithm for each component under each determined N _Monitor (t). Algorithm (1, 0) represents the algorithm that component 1 runs when N _Monitor (t) = 0, and algorithm (1, 1) represents the algorithm that component 1 runs when N _Monitor (t) = 1. Similarly, algorithm (x, N) represents the algorithm that component x runs when N _Monitor (t) = N. The PHM developer sets the sampling rate of the PHM algorithm runtime data and the time interval between two consecutive runs of the algorithm for each component under each determined N _Monitor (t). Table 4 is obtained.

Table 4 PHM algorithm operation mode

STEP 5. Determine the algorithm performance

According to the sampling rate and monitoring interval in Table 4, PHM engineers need to test the performance of the PHM algorithm under different computing resources when developing the PHM algorithm. The test content is: for algorithm α, when it runs on Class, computing power is/> When the computing resources are sufficient, the computing delay of a single execution can be determined./> and power consumption/> Table 5 is obtained;

Table 5 PHM algorithm α running performance

STEP 6. Determine the PHM system operation requirements

When the airborne PHM system is running, it can determine all algorithms that need to be run in the entire PHM system at the current moment and their sampling rates, running intervals and algorithm running performance according to the monitoring requirements of each airborne component at the current moment, and assume that each PHM algorithm is only run on one computing resource; assume that there are a total of l PHM algorithms that need to be run at time t, then these PHM algorithms that need to be run and their sampling rates, running intervals and algorithm running performance constitute a set; the running requirements and running performance of the airborne PHM system at time t are determined, that is, the running requirements of the airborne PHM system and the space for task allocation;

Step 2: PHM calculation task allocation and optimization of spatial structure

Assume that at a certain moment, the airborne PHM system needs to run K algorithms simultaneously;

Based on the requirements of PHM system operation and the space of task allocation, the PHM computing task allocation is optimized within the task allocation space;

The performance index function J(D) is constructed by taking the delay and power consumption of all algorithms executed in the airborne PHM system as evaluation indicators:

in,

is a K-dimensional column vector, indicating that there are K algorithms that need to be run, which are respectively assigned to n ₁ , n ₂ , …, n _K computing resources for execution;

and/> They represent the delay time and power consumption of the i-th algorithm executed in the currently allocated computing resources respectively;

k ₁ and k ₂ are weight factors, indicating the weights of algorithm execution delay and computing power consumption in performance indicators, satisfying k ₁ +k ₂ =1, k ₁ , k ₂ ≥0;

The constraints are: the algorithm needs to run on computing resources that can meet the minimum computing power level requirements;

On the basis of the above allocation optimization conditions, the optimization algorithm is used to allocate each PHM algorithm that needs to be run to a computing resource, so that the performance indicator function J(D) in the above allocation optimization conditions is minimized, that is, the overall performance of the PHM system algorithm operation is optimal.

The present invention classifies and grades computing resources, clarifies the performance of computing resources, and constructs a computing resource model. The algorithm is classified by distinguishing online and offline algorithms and the accuracy and cost of the algorithm to construct a PHM algorithm model. The monitoring requirements of components are determined by dividing the importance of airborne equipment, and the corresponding algorithms are used to construct an operation requirement model of the airborne PHM system.

Detailed ways

The airborne distributed PHM computational modeling method of the present invention specifically includes the following steps:

Step 1: Modeling the airborne PHM operation platform

First, the computational requirements and computing power for the aircraft PHM system are modeled;

STEP 1. Modeling of airborne distributed computing resources

The airborne PHM system consists of many computing resources that can run the PHM algorithm. Assume that a certain airborne PHM system has a total of M computing resources that can be used to run the PHM algorithm.

(1) Computing Resource Classification C _Type

These computing resources are different embedded platforms with different processor architectures. Each computing resource is a homogeneous system or a heterogeneous system. The computing resources are classified according to the embedded platform and system structure used by the computing resources of the airborne PHM system. Assuming that there are n types of computing resources in the whole PHM system, a label is assigned to each type of resource. As shown in Table 1;

Table 1 Classification of airborne PHM computing resources

in, Refers to computing resources of a single ARM architecture, /> Refers to the computing resources of the dual DSP architecture. Refers to the computing resources of DSP+FPGA heterogeneous architecture,/> Refers to the computing resources of ARM+DSP+FPGA heterogeneous architecture... The current development trend of embedded platforms is heterogeneous platforms. New embedded heterogeneous system architectures often appear under different functional requirements. It is impossible to list all architecture types. The subscript rule here is to sort by single-core architecture, multi-core homogeneous architecture, simple heterogeneous architecture, and complex heterogeneous architecture. When a new embedded system architecture is applied to the airborne PHM system, just continue to add new subscript numbers in the class code.

(2) Computing Power Classification C _Ability

On the basis of the classification of computing resources, each type of computing resource is graded according to its computing power, and the computing power level of the i-th computing resource is specified as Assuming that the computing resource with the smallest throughput in the whole PHM system is the No. 1 computing resource (for example, an ARM), its computing power level is specified/> And stipulate Assume that the computing resource with the largest throughput in the whole PHM system is the kth computing resource (e.g., a certain FPGA). Referring to the definition of unit computing power, its computing power level is specified. Then C _Abilitymax is the maximum value of the computing power level of computing resources in the whole PHM system. For example, assuming that the throughput of the above computing resource No. 1 (a certain ARM) is 2Mb/s, then/> Assuming that the throughput of the computing resource No. k (the computing resource with the highest throughput, a certain FPGA) is 1234.5Mb/s, its computing power level is defined as/> (1234.5/2=617.25 rounded down) is the maximum value of the computing power level of computing resources in the whole PHM system; assuming that the throughput of computing resource a is 666.6Mb/s, its computing power level/> (666.6/2=333.3 rounded down). The remaining computing resources are rounded down with reference to the unit computing power to obtain the computing power level corresponding to all PHM algorithm model operation platforms of the entire aircraft's onboard PHM system/>

According to Table 1 and the computing power classification, each computing resource in the airborne PHM system can correspond to and/> Thus, Table 2 is obtained. (For example, resource No. 1 is a single ARM architecture, which is/> Class computing resources, computing power level 1; resource No. 2 is a dual DSP homogeneous architecture, for/> Class computing resources, computing power level 3; M resources are ARM+DSP+FPGA heterogeneous architecture, for/> Class computing resources, computing power level is/> ).

Table 2 Performance of airborne PHM computing resources

STEP 2. PHM algorithm performance modeling

For a certain airborne component, the operational requirements of the PHM algorithm need to be modeled;

(1) Determine real-time performance: When developing PHM algorithms, PHM engineers divide PHM algorithms into online algorithms and offline algorithms based on the timeliness of the data processed by the algorithm and whether the algorithm needs to use historical data. An online algorithm is one that processes data that is time-sensitive and does not require the use of historical data; an offline algorithm is one that processes data that reflects the state parameters of a component over a period of time or requires the use of historical data.

(2) Determine the computing resource requirements: For the algorithm α (for example, α represents the sliding average filter algorithm) that needs to be run on the airborne PHM system, PHM engineers need to test the algorithm on various computing resources and computing resources of various computing power levels according to the algorithm complexity, logical structure, data length, and data cache requirements when developing the PHM algorithm. Based on the test results, determine the algorithm's performance on various resources. When running on a class resource, at least (α represents different algorithms, i=1, 2, ..., n) Computing resources of a certain computing power level can meet the requirements of algorithm operation, so an operation requirement can be obtained for each algorithm, which is shown in Table 3.

Table 3 Algorithm α operation requirements

For example, Table 3 shows that algorithm α needs to Class, computing power level greater than/> computing resources; or in/> Class, computing power level greater than/> on computing resources; or on/> Class, computing power level greater than on computing resources of ...; or on /> Class, computing power level greater than/> can run on computing resources.

STEP 3. Component monitoring requirements modeling

Assumption: An aircraft can perform m types of flight missions. The aircraft's onboard PHM system monitors the status of x onboard components. The onboard PHM system obtains its health status parameters to characterize the health status of the components. First, construct parameters to characterize the monitoring requirements of the components.

(1) Component Importance

The importance of components is divided according to the degree of harm caused by component failure to flight safety. The greater the harm caused by component failure to flight safety, the greater the importance of components. The bigger;

(2) Task importance

Each flight mission of the aircraft has a number j (j = 1, 2, ..., m), where m is the type of flight mission that the aircraft can perform. The components are divided into task importance according to the harm of component failure to the completion of the flight mission. The greater the harm of component failure to the completion of a certain flight mission, the greater the The larger the size, the more likely it is that the same component will perform different flight missions. Probably the same;

(3) Component health level N _Health (t):

t represents N _Health (t) as a function of time. As time goes by, the performance of the component gradually degrades and N _Health (t) changes.

The failure of airborne components is a gradual degradation process. In the early stage of component degradation, the degradation process is often stable. The components are divided into health levels based on the current assessment of the component health status level. The greater the possibility of component failure, the greater N _Health (t);

(4) Monitoring requirement N _Monitor (t):

The constructed monitoring requirement N _Monitor (t) represents the urgency of a component’s need for monitoring;

STEP 4. Set the PHM algorithm operation mode

For a certain component, engineers need to select several algorithms for it according to its degradation model during the development phase of its PHM algorithm, and use different monitoring sampling rates and fault diagnosis and life prediction algorithms according to different monitoring requirements;

The settings of the algorithm, sampling rate, and monitoring interval follow the following rules:

④ When the monitoring demand is low, the selected algorithm has the characteristics of low power consumption and low computing power requirements. The algorithm can be an offline algorithm or an online algorithm. The sampling rate of the sensor is low and the monitoring interval is long. The goal of the algorithm is to monitor and discover the early fault characteristics of the components.

⑤ With the increase of monitoring demand, the real-time performance and calculation accuracy of the selected algorithm are improved, with the online algorithm as the main one, the sampling rate of the sensor is increased, the monitoring interval is shortened or even continuous monitoring; The goal of the algorithm is to accurately identify the accelerated degradation process of the component after the early fault characteristics of the component appear;

⑥ When the monitoring demand continues to increase, an algorithm with high precision and real-time characteristics should be selected, the sensor sampling rate should be the highest, and the monitoring method should be real-time monitoring; the purpose of the algorithm is to accurately predict the remaining service life of the component during the accelerated degradation stage of the component, and to be able to immediately and accurately identify and locate the fault when the component fails;

Among all x airborne components, the highest monitoring requirement is The PHM developer finds a PHM algorithm for each component under each determined N _Monitor (t). Algorithm (1, 0) represents the algorithm that component 1 runs when N _Monitor (t) = 0 (/> Among the three indicators, only NMissionj can be 0. When N _Monitor (t) is 0, N Monitor (t) is 0). Algorithm (1, 1) represents the algorithm that component 1 runs when N _Monitor (t) = 1. ..., Algorithm (x, N) represents the algorithm that component x runs when N _Monitor (t) = N (N _Monitor (t) is an integer greater than or equal to 0, because/> The three indicators are all integers greater than or equal to 0, which have been clearly defined in the previous indicator definitions); PHM developers set the data sampling rate and the time interval between two consecutive runs of the PHM algorithm for each component under each determined N _Monitor (t); sampling rate (1, 0) represents the data sampling rate set by component 1 when N _Monitor (t) = 0, monitoring interval (1, 0) represents the time interval between two consecutive runs of the algorithm by component 1 when N _Monitor (t) = 0, sampling rate (1, 1) represents the data sampling rate set by component 1 when N _Monitor (t) = 1, monitoring interval (1, 1) represents the time interval between two consecutive runs of the algorithm by component 1 when N _Monitor (t) = 1, ..., sampling rate (x, N) represents the data sampling rate set by component x when N _Monitor (t) = N, monitoring interval (x, N) represents the time interval between two consecutive runs of the algorithm by component x when N _Monitor (t) = N; thus, Table 4 can be obtained;

Table 4 PHM algorithm operation mode

STEPS. Determine the algorithm performance

According to the sampling rate and monitoring interval in Table 4, PHM engineers need to test the performance of the PHM algorithm under different computing resources when developing the PHM algorithm. The test content is: for algorithm α, when it runs on Class, computing power is/> When the computing resources are sufficient, the computing delay of a single execution can be determined./> and power consumption/> Assume that for algorithm α: Algorithm α in/> The minimum computing power requirement for this resource type is 3. The minimum computing power requirement for class resources is 2. The minimum computing power requirement for this resource type is 1; The computational delay of a single execution under the computing resource of class and computing power 3 is/> Power consumption is/> exist The computational delay of a single execution under the computing resource of class and computing power 2 is/> Power consumption is/> exist The computing delay of a single execution under the computing resource of class and computing power 3 is/> Power consumption is/> exist The computational delay of a single execution under a computing resource of class 1 is/> Power consumption is/> exist The computational delay of a single execution under the computing resource class with a computing power of 2 is/> Power consumption is/> In/> The computational delay of a single execution under the computing resource of class and computing power 3 is/> Power consumption is/> The specified algorithm cannot run on computing resources/> This resulted in Table 5.

Table 5 PHM algorithm α running performance

Similarly, based on the performance tests conducted by engineers, each algorithm can obtain a table that characterizes the algorithm's operating performance;

STEP 6. Determine the PHM system operation requirements

When the airborne PHM system is running, it can determine all the algorithms that need to be run in the entire PHM system at the current moment and their sampling rates, running intervals, and algorithm running performance according to the monitoring requirements of each airborne component at the current moment, and assume that each PHM algorithm is only run on one computing resource; assume that there are l PHM algorithms that need to be run at time t, then these PHM algorithms that need to be run and their sampling rates, running intervals, and algorithm running performance constitute a set. The running requirements and running performance of the airborne PHM system at time t are determined, that is, the running requirements of the airborne PHM system and the space for task allocation;

Step 2: PHM calculation task allocation and optimization of spatial structure

Assume that at a certain moment, the airborne PHM system needs to run K algorithms simultaneously.

Based on the requirements of PHM system operation and the space of task allocation (the PHM algorithm to be run at time t and its sampling rate, operation interval and algorithm operation performance determined in step 1, STEP 6), optimize the PHM calculation task allocation within the task allocation space;

The delay time of PHM algorithm execution and the power consumption generated by calculation are important indicators for evaluating the performance of the entire PHM system operation platform. The purpose of PHM calculation task allocation optimization is to minimize the overall delay and power consumption of the calculation tasks running on all computing resources, thereby improving the real-time performance of system monitoring data processing with low power consumption.

The performance index function J(D) is constructed by taking the delay and power consumption of all algorithms executed in the airborne PHM system as evaluation indicators:

in,

is a K-dimensional column vector, indicating that there are K algorithms that need to be run, which are respectively assigned to n ₁ , n ₂ , …, n _K computing resources for execution;

and/> They represent the delay time and power consumption of the i-th algorithm executed in the currently allocated computing resources respectively;

k ₁ and k ₂ are weight factors, indicating the weights of algorithm execution delay and computing power consumption in performance indicators, satisfying k ₁ +k ₂ =1, k ₁ , k ₂ ≥0;

The constraints are: the algorithm needs to run on computing resources that can meet the minimum computing power level requirements;

On the basis of the above allocation optimization conditions, various optimization algorithms are used, such as particle swarm optimization (Garcíagonzalo E, Fernándezmartínez JL. A brief historical review of particle swarm optimization (PSO) [J]. Journal of Bioinformatics & Intelligent Control, 2012, 1 (1): 3-16.), grey wolf algorithm (MIRJALILI S, MIRJALILI S M, LEWIS A. Grey wolf optimization [J]. Advances in Engineering Software, 2014, 69 (7): 46-61.), to allocate each PHM algorithm to be run to a computing resource, so that the performance index function J (D) in the above allocation optimization conditions is minimized, that is, the overall performance of the PHM system algorithm operation is optimal (the construction of the performance index function refers to the performance index function in the optimal control theory).

Claims (1)

1. An airborne distributed PHM computational modeling method, characterized in that it specifically includes the following steps: Step 1: Modeling the airborne PHM operation platform STEP 1. Modeling of airborne distributed computing resources The airborne PHM system consists of many computing resources that can run the PHM algorithm. Assume that a certain airborne PHM system has a total of M computing resources that can be used to run the PHM algorithm; (1) Computing Resource Classification C _Type These computing resources are different embedded platforms with different processor architectures. Each computing resource is a homogeneous system or a heterogeneous system. The computing resources are classified according to the embedded platform and system structure used by the computing resources of the airborne PHM system. Assuming that there are n types of computing resources in the whole PHM system, a label is assigned to each type of resource. i = 1, 2, ... n, as shown in Table 1; Table 1 Classification of airborne PHM computing resources in, Refers to computing resources of a single ARM architecture, /> Refers to the computing resources of the dual DSP architecture, /> Refers to the computing resources of DSP+FPGA heterogeneous architecture,/> Refers to the computing resources of ARM+DSP+FPGA heterogeneous architecture; the current trend of embedded platform development is heterogeneous platform. New embedded heterogeneous system architectures often appear under different functional requirements. It is impossible to list all architecture types. The subscript rules here are sorted according to the types of single-core architecture, multi-core homogeneous architecture, simple heterogeneous architecture, and complex heterogeneous architecture. When a new embedded system architecture is applied to the airborne PHM system, the new subscript number can be added to the class code; (2) Computing Power Classification C _Ability On the basis of the classification of computing resources, each type of computing resource is graded according to its computing power, and the computing power level of the i-th computing resource is specified as i = 1, 2, ..., M; Assuming that the computing resource with the smallest throughput in the whole PHM system is the computing resource No. 1, then its computing power level is specified/> And stipulate/> is the unit computing power; assuming that the computing resource with the largest throughput in the whole PHM system is the kth computing resource, refer to the definition of unit computing power to define its computing power level/> Then C _Abilitymax is the maximum value of the computing power level of the computing resources in the whole aircraft PHM system; the computing resources are rounded down with reference to the unit computing power to obtain the computing power level corresponding to all PHM algorithm model running platforms of the whole aircraft PHM system/> i = 1, 2, ..., M; According to Table 1 and the computing power classification, each computing resource in the airborne PHM system can correspond to and i＝1,2,…,n,j＝1,2,…,M, thus we get Table 2; Table 2 Performance of airborne PHM computing resources STEP 2. PHM algorithm performance modeling For a certain airborne component, the operational requirements of the PHM algorithm need to be modeled; (1) Determine real-time performance: When developing PHM algorithms, PHM engineers divide PHM algorithms into online algorithms and offline algorithms based on the timeliness of the data processed by the algorithm and whether the algorithm needs to use historical data. The algorithm model that processes data with timeliness and does not need to use historical data is an online algorithm with real-time performance; the algorithm model that processes data that reflects the status parameters of the component over a period of time or needs to use historical data is an offline algorithm that does not have real-time performance. (2) Determine the computing resource requirements: For the algorithm α that needs to be run on the airborne PHM system, PHM engineers need to test the algorithm on various computing resources and computing resources of various computing power levels based on the algorithm complexity, logical structure, data length, and data cache requirements when developing the PHM algorithm. Based on the test results, determine the algorithm's performance on various resources. When running on a class resource, at least/> α represents different algorithms, i = 1, 2, ..., n, and computing resources of a certain computing power level can meet the requirements of algorithm operation. Therefore, an operation requirement can be obtained for each algorithm, which is shown in Table 3; Table 3 Algorithm α operation requirements STEP 3. Component monitoring requirements modeling Assumption: An aircraft can perform m types of flight missions. The aircraft's onboard PHM system monitors the condition of x onboard components in total. The onboard PHM system obtains its health status parameters to characterize the health status of the components. First, the parameters are constructed to characterize the monitoring requirements of the components. (1) Component Importance i = 1, 2, ..., x: The importance of components is divided according to the degree of harm caused by component failure to flight safety. The greater the harm caused by component failure to flight safety, the greater the importance of components. The bigger; (2) Task importance j = 1, 2, ..., m: Each flight mission of the aircraft has a number j, j = 1, 2, ..., m, where m is the type of flight mission that the aircraft can perform. The components are divided into task importance according to the harm of component failure to the completion of the flight mission. The greater the harm of component failure to the completion of a certain flight mission, the greater the The larger the size, the more likely it is that the same component will perform different flight missions. Probably the same; (3) Component health level N _Health (t): t represents N _Health (t) as a function of time. As time goes by, the performance of the component gradually degrades and N _Health (t) changes; The failure of airborne components is a gradual degradation process. In the early stage of component degradation, the degradation process is often stable. The components are divided into health levels based on the current assessment of the component health status level. The greater the possibility of component failure, the greater N _Health (t); (4) Monitoring requirement N _Monitor (t): The constructed monitoring requirement N _Monitor (t) represents the urgency of a component’s need for monitoring; STEP 4. Set the PHM algorithm operation mode For a certain component, engineers need to select several algorithms for it according to its degradation model during the development phase of its PHM algorithm, corresponding to different monitoring sampling rates and fault diagnosis and life prediction algorithms under different monitoring requirements; The settings of the algorithm, sampling rate, and monitoring interval follow the following rules: ⑦ When the monitoring demand is low, the selected algorithm has the characteristics of low power consumption and low computing power requirements. The algorithm can be an offline algorithm or an online algorithm. The sampling rate of the sensor is low and the monitoring interval is long. The goal of the algorithm is to monitor and discover the early fault characteristics of the components. ⑧ As the monitoring demand increases, the real-time performance and calculation accuracy of the selected algorithm are improved, with the online algorithm as the main one, the sampling rate of the sensor is increased, the monitoring interval is shortened or even continuous monitoring; the goal of the algorithm is to accurately identify the accelerated degradation process of the component after the early fault characteristics of the component appear; ⑨ When the monitoring demand continues to increase, an algorithm with high precision and real-time characteristics should be selected, the sensor sampling rate should be the highest, and the monitoring method should be real-time monitoring; the purpose of the algorithm is to accurately predict the remaining service life of the component during the accelerated degradation stage of the component, and to be able to immediately and accurately identify and locate the fault when the component fails; Among all x airborne components, the highest monitoring requirement is The PHM developer finds a PHM algorithm for each component under each determined N _Monitor (t). Algorithm (1, 0) represents the algorithm that component 1 runs when N _Monitor (t) = 0, and algorithm (1, 1) represents the algorithm that component 1 runs when N _Monitor (t) = 1. Similarly, algorithm (x, N) represents the algorithm that component x runs when N _Monitor (t) = N. The PHM developer sets the sampling rate of the PHM algorithm runtime data and the time interval between two consecutive runs of the algorithm for each component under each determined N _Monitor (t). Table 4 is obtained. Table 4 PHM algorithm operation mode STEP 5. Determine the algorithm performance According to the sampling rate and monitoring interval in Table 4, PHM engineers need to test the performance of the PHM algorithm under different computing resources when developing the PHM algorithm. The test content is: for algorithm α, when it runs on Class, computing power is/> When the computing resources are sufficient, the computing delay of a single execution can be determined./> and power consumption Table 5 is obtained; Table 5 PHM algorithm α running performance STEP 6. Determine the PHM system operation requirements When the airborne PHM system is running, all algorithms that need to be run in the entire PHM system at the current moment and their sampling rates, running intervals and algorithm running performance can be determined according to the monitoring requirements of each airborne component at the current moment, and it is assumed that each PHM algorithm is only run on one computing resource; assuming that there are a total of l PHM algorithms that need to be run at time t, then these PHM algorithms that need to be run and their sampling rates, running intervals and algorithm running performance constitute a set; the running requirements and running performance of the airborne PHM system at time t are determined, that is, the running requirements of the airborne PHM system and the space for task allocation; Step 2: PHM calculation task allocation and optimization of spatial structure Assume that at a certain moment, the airborne PHM system needs to run K algorithms simultaneously; Based on the requirements of PHM system operation and the space of task allocation, the PHM computing task allocation is optimized within the task allocation space; The performance index function J(D) is constructed by taking the delay and power consumption of all algorithms executed in the airborne PHM system as evaluation indicators: in, is a K-dimensional column vector, indicating that there are K algorithms that need to be run, which are respectively assigned to n ₁ , n ₂ , …, n _K computing resources for execution; and/> They represent the delay time and power consumption of the i-th algorithm executed in the currently allocated computing resources respectively; k ₁ and k ₂ are weight factors, indicating the weights of algorithm execution delay and computing power consumption in the performance index, satisfying k ₁ +k ₂ =1, k ₁ , k ₂ ≥0; The constraints are: the algorithm needs to run on computing resources that can meet the minimum computing power level requirements; On the basis of the above allocation optimization conditions, the optimization algorithm is used to allocate each PHM algorithm that needs to be run to a computing resource, so that the performance indicator function J(D) in the above allocation optimization conditions is minimized, that is, the overall performance of the PHM system algorithm operation is optimal.