CN116449135B - Method and system for determining health state of electromechanical system component and electronic equipment - Google Patents
Method and system for determining health state of electromechanical system component and electronic equipment Download PDFInfo
- Publication number
- CN116449135B CN116449135B CN202310421466.8A CN202310421466A CN116449135B CN 116449135 B CN116449135 B CN 116449135B CN 202310421466 A CN202310421466 A CN 202310421466A CN 116449135 B CN116449135 B CN 116449135B
- Authority
- CN
- China
- Prior art keywords
- component
- degradation
- health
- electromechanical
- association relation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000036541 health Effects 0.000 title claims abstract description 71
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000006731 degradation reaction Methods 0.000 claims abstract description 59
- 230000015556 catabolic process Effects 0.000 claims abstract description 57
- 230000004927 fusion Effects 0.000 claims abstract description 46
- 230000003862 health status Effects 0.000 claims abstract description 15
- 230000008569 process Effects 0.000 claims abstract description 14
- 238000012544 monitoring process Methods 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 58
- 230000009471 action Effects 0.000 claims description 24
- 238000005457 optimization Methods 0.000 claims description 24
- 230000002787 reinforcement Effects 0.000 claims description 16
- 230000003993 interaction Effects 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 7
- 230000007613 environmental effect Effects 0.000 claims description 7
- 230000005540 biological transmission Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 abstract description 10
- 238000004364 calculation method Methods 0.000 abstract description 4
- 230000000875 corresponding effect Effects 0.000 description 16
- 238000010586 diagram Methods 0.000 description 6
- 238000012549 training Methods 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000011005 laboratory method Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/003—Environmental or reliability tests
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/092—Reinforcement learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Abstract
The invention discloses a method and a system for determining the health state of an electromechanical system component and electronic equipment, and relates to the technical field of monitoring of the health state of the component. The invention provides a method for determining the health state of an electromechanical system component, which comprises the steps of firstly obtaining the observed quantity of an electromechanical system sensor; then, obtaining a one-dimensional health indication capable of representing the health state of the electromechanical system component through a built association relation model of the component degradation hidden variable and the observed quantity of the system sensor; finally, an accurate assessment of the health status of the electromechanical system component can be achieved based on the health indication. In addition, in the practical application process, the method can adapt to different electromechanical components of the electromechanical system, has the characteristics of low calculation complexity and flexible fusion mode, and can realize accurate component health state evaluation on the basis of the conventional system sensor layout.
Description
Technical Field
The present invention relates to the field of health status evaluation and management technologies of electromechanical systems, and in particular, to a method and a system for determining health status of an electromechanical system component, and an electronic device.
Background
Electromechanical system components are components of an electromechanical system, the components being coupled to one another to form a complex system of mechanical, electrical, hydraulic and control interactions, i.e., an electromechanical system. With the widespread use of electromechanical systems, it is common for failure in operation to lead to severe accidents. Failure of the electromechanical system is caused by failure of an electromechanical component, for example, damage of a blade can damage an air inlet flow field of an engine, so that surge is formed, and the engine is seriously stopped; the tension, compression and torsional deformation of the roller screw lead to the reduction of the transmission precision of the electromechanical servo system, and serious erroneous instructions. In summary, the health and reliability of the electromechanical components plays a vital role in the safe and reliable operation of the entire electromechanical system. Therefore, the health state of the electromechanical system component is accurately estimated, the safety state of the electromechanical system can be judged in advance, and the method has a key meaning for guaranteeing the service safety of the electromechanical system and supporting maintenance and guarantee services.
Although the electromechanical components are subjected to necessary reliability experiments in a laboratory before use, so that the reliability requirements of the electromechanical systems are met, the actual health states of the components still need to be monitored by corresponding sensors in the application process of the electromechanical systems. However, in the actual electromechanical system, only a limited number of sensors remain due to various reasons such as structural limitations, weight saving, and the like. Based on this, the health status of most electromechanical components cannot be estimated using laboratory methods, but can only be obtained indirectly through system-level observations monitored by system sensors. For example, in the electromechanical servo system, the health state of the bearing can be obtained through the vibration sensor in the experiment of the reliability of the component, but in the system, the actual health state of the component can only be indirectly estimated by measuring the current, voltage and other system-level observation signals of the subsystem where the bearing is positioned. Furthermore, unlike laboratory environments that can employ a sufficient amount of rich historical data, real systems have only a small amount of data about limited sensors, making state of health assessment of electromechanical system components based on system-level observed signals difficult to develop.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a method, a system and electronic equipment for determining the health state of an electromechanical system component.
In order to achieve the above object, the present invention provides the following solutions:
a method of determining a state of health of an electromechanical system component, comprising:
constructing an association relation model between the degradation hidden variable of the electromechanical component and the observed quantity of the electromechanical system sensor; the component degradation hidden variables comprise bearing wear of mechanical components, energy transmission efficiency of electronic components and other degradation physical quantities; the observed quantity of the system sensor comprises input and output voltage, current and conventional sensor output quantity.
Estimating parameters of the association relation model; the estimation method is based on a reinforcement learning method under the constraint of strong robustness.
Inputting the degradation characteristics of the electromechanical system component into the nonlinear fusion model to obtain a one-dimensional health indication;
a health status of an electromechanical system component is determined based on the one-dimensional health indication.
Optionally, constructing the association model between the electromechanical component degradation hidden variable and the electromechanical system sensor observables includes:
acquiring degradation characteristics which are characterized by observed quantity of a system sensor and are related to a component degradation hidden variable; the degradation feature includes: the voltage effective value ratio of the input and output system, the current peak value ratio of the input and output system, the power loss of the system, the output voltage difference in a preset time interval and the characteristic quantity constructed based on the sensor monitoring data;
constructing an association relation model with nonlinear fusion characteristics, inputting degradation characteristics of the component, and outputting one-dimensional health indications representing degradation hidden variables of the component; the nonlinear fusion characteristics comprise linear fusion based on fusion weights and nonlinear transformation of nonlinear function action;
determining an objective function, and abstracting the construction of the one-dimensional health indication into an optimization problem based on the degradation characteristic, the objective function and the association relation model structure;
solving the optimization problem based on a reinforcement learning method under strong robustness constraint to obtain a fusion weight and a nonlinear function which maximize the objective function value;
and completing the construction of the association relation model based on the fusion weight and the nonlinear function which make the objective function value maximum.
Optionally, solving the optimization problem based on reinforcement learning to obtain a fusion weight and a nonlinear function that maximize the objective function value, specifically including:
and transforming the optimization problem into an interaction process of the agent and the environment based on reinforcement learning under strong robustness constraint so as to obtain a fusion weight and a nonlinear function which maximize the objective function value.
Optionally, the interaction process of the agent with the environment includes:
the agent observes the environmental state at the current moment from the environment and determines a strategy based on the knowledge of the environment at the current moment; the environment state at the current moment is a fusion weight and a nonlinear function at the current moment;
selecting an action taken by the environmental state based on the policy, the action changing the environment to produce a new environmental state while deriving a prize value from the changed environment;
updating the strategy based on the reward value, selecting a new environment state to take action based on the updated strategy, and so on until the reward value reaches the maximum.
Optionally, the policy is updated based on the prize value using a SAC network architecture incorporating strong robustness constraints.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
according to the method for determining the health state of the electromechanical system component, after the association relation model between the degradation hidden variable of the electromechanical component and the observed quantity of the electromechanical system sensor is built, the observed quantity of the system sensor is input into the built association relation model, so that one-dimensional health indication capable of representing the health state of the electromechanical system component can be obtained, and finally, accurate evaluation of the health state of the electromechanical system component can be realized based on the one-dimensional health indication. In addition, in the practical application process, different degradation characteristics can be obtained according to different electromechanical system components, and the method has the characteristics of low calculation complexity and flexible fusion mode.
Corresponding to the method for determining the health state of the electromechanical system component provided by the invention, the invention also provides the following implementation structure:
an electromechanical system component health status determination system, comprising:
the component degradation characteristic acquisition module is used for acquiring degradation characteristics which are characterized by observed quantity of a system sensor and are related to component degradation hidden variables; the degradation feature includes: the voltage effective value ratio of the input and output system, the current peak value ratio of the input and output system, the power loss of the system, the output voltage difference in a preset time interval and the characteristic quantity constructed based on the sensor monitoring data;
the association relation model construction module is used for constructing an association relation model;
the health state determining module is used for inputting the degradation characteristics into the association relation model to obtain one-dimensional health indications so as to determine the health state of the electromechanical system component;
an electronic device, comprising:
a memory for storing a computer program;
and the processor is connected with the memory and is used for calling and executing the computer program so as to implement the method for determining the health state of the electromechanical system component.
Optionally, the memory is a computer readable storage medium.
The implementation structure provided by the invention has the same effect as that achieved by the method for determining the health state of the electromechanical system component provided by the invention, so that the description is omitted here.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for determining the health status of an electromechanical system component provided by the present invention;
FIG. 2 is a schematic diagram of an embodiment of a method for determining the health status of an electromechanical system component according to the present invention;
FIG. 3 is a schematic diagram of the reinforcement learning-based optimization problem solving method;
FIG. 4 is a schematic diagram of a policy network provided by the present invention;
FIG. 5 is a schematic diagram of an evaluation network in a SoftActor-Critic (SAC) method provided by the invention;
FIG. 6 is a state and state space diagram provided by the present invention;
FIG. 7 is a schematic diagram of the motion and motion space provided by the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention aims to provide a method and a system for determining the health state of an electromechanical system component and electronic equipment, which can accurately evaluate the health state of the electromechanical system component.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
As shown in fig. 1 and 2, the method for determining the health status of an electromechanical system component provided by the invention includes:
step 100: and obtaining the observed quantity of the electromechanical system sensor. The observed quantity of the system sensor obtained in the step is based on a conventional sensor of an electromechanical system, and no additional sensor is added for monitoring the health state of the component.
Step 101: and constructing an association relation model of the component degradation hidden variable and the observed quantity of the system sensor. The association model constructed in this step is not limited to the observation data and the degradation law of the health state. The constructed association relation model takes the observed quantity of the electromechanical system sensor as input and takes the one-dimensional health indication as output. The degradation rules of the component degradation hidden variables are described by one-dimensional health indications.
Degradation characteristics associated with component degradation hidden variables, which are characterized by observed quantity of a system sensor, are obtained. Taking current and voltage signals as examples, the degradation characteristic of the structure is as follows:wherein l 1 To output the effective value ratio of the voltage of the input component, l 2 To output the current peak ratio of the input component, l 3 For power loss through the component, l 4 For a certain time interval of output voltage difference l 5 ,l 6 ,...,l N Is a characteristic quantity constructed based on other relevant sensors. The normalization takes place taking into account the different magnitudes of the different degradation characteristics. Given a period of history j=1, 2..α, where α is the length of the history. Then the degradation characteristic can be normalized to { x } 1,j ,x 2,j ,...,x N,j }. Further, the standardized degradation features obtain one-dimensional health indications through an association relation model with nonlinear fusion characteristics.
Based on the above description, the nonlinear fusion characteristics include linear fusion based on fusion weights and nonlinear transformation of nonlinear function actions. The specific process flow is as follows:
let the fusion weight corresponding to the degradation feature be ω= [ ω ] 1 ,ω 2 ,...,ω k ,...,ω N ]Each fusion weight omega k ∈[0,1]. Linear fused one-dimensional variable o j Expressed as:
where N is the number of degradation features, ω k X is the kth fusion weight k,j And the k degradation characteristic normalized by the moment j.
Based on linear fusion, a nonlinear function f (·) e Γ= { sigmoid (), cos (), log (),.+ -) is introduced for mining the potential law of nonlinear degradation of the health state, and the mathematical description is as follows:
in the formula, h j And (3) as a one-dimensional health indication value corresponding to the moment j, Γ is a function set, sigmoid () is a sigmoid function, cos () is a cosine function, and log () is a logarithmic function. Wherein, the distribution of different fusion weights omega and the selection of a nonlinear function f (-) form different association relation models pi i I=1, 2,.. 1 ,π 2 ,…,π n }。
Further, the selection of the fusion mode is converted into an optimization problem with the aim of outputting one-dimensional health indication reflecting the change of the health state of the electromechanical system component
This optimization problem is to find a set of linear fusion weights and nonlinear functions so that the corresponding one-dimensional health indicators can maximize the objective function. To establish one-dimensional health indication and health careThe invention introduces evaluation index Mon, tre, rob related to degradation characteristics, as the health status of the electromechanical system component is reduced and degraded into positive correlation, for the objective function G in the optimization problem. Taking into account the natural advantages of the SNR structure, the invention constructs a matrix (evaluation matrix of health indicators) SNR of similar structure d To characterize the objective function G, namely:
G＝SNR d ＝u/v＝(Mon+Tre)/Rob
where u is the mean of the signal in the signal-to-noise ratio definition and v is the variance of the signal; the processing method of each evaluation index Mon, tre, rob is as follows:
A. monotonicity of
The electromechanical system component degradation process is unidirectional, and thus the change in the state of health of the electromechanical system component should also be unidirectional. The mathematical description of monotonicity Mon is as follows:
wherein delta () is a unit step function, mon is a mathematical sign of a defined monotonicity evaluation index, and alpha is the length of a given period of history time.
B. Robustness (robustness)
In practical applications, there are individual differences in individual component degradation processes, and there are disturbances such as noise. Thus, the fused one-dimensional health indicator should reduce the perturbation and uncertainty of the original features. The invention provides the variance of the residual error of the original signal and the smoothed trend as the description of the robustness, and the mathematical description of the robustness Rob is as follows:
wherein,for the residuals of one-dimensional health indications and their post-smoothing trends, rob describes health indications and smoothingVariance of fitting error of the post trend, exp () is an exponential function.
C. Trend of
The degree of degradation of the component is positively correlated with time, so a one-dimensional health indicator should have a time-series correlation. The mathematical description of Tre trend is as follows:
step 102: based on the reinforcement learning training association model, model parameters (corresponding to a set of optimal linear fusion weights and nonlinear functions) are obtained. Considering that the optimizing variables comprise continuous variables (linear fusion weights) and discrete variables (nonlinear functions), a conventional solver cannot be adopted for solving. And the genetic algorithm has high time complexity and poor stability of the solving result. Therefore, in order to overcome the problems, the invention provides an optimization problem solving scheme based on reinforcement learning under strong robustness constraint. The scheme is insensitive to the number of sensors and training data, and has small calculation complexity. In addition, the method is not limited by the specific form of the objective function, and can meet the flexible fusion objective.
Specifically, based on reinforcement learning, the solution of the optimization problem transitions to agent interactions with the environment. An agent is the subject of interaction with the surrounding environment. The specific interaction process is as follows: the agent observes from the environment the state at time t (i.e., the environmental state) s t e.OMEGA.OMEGA.is a defined state space. Based on the current knowledge, a policy is defined as pi (a|s), and an action a to be taken in the current state is selected t E, Λ is the action space. Action a t Further changing the environment, generating a new state s t+1 And return a prize r t . Further, the agent awards r t The policy pi (a|s) is updated to select actions that can get more rewards.
As shown in fig. 3, the proxy finds the fusion weights ω and the nonlinear function f (·) that maximize the objective function G in T steps. In this process, action a t In particular to the t step for fusing weight omega and nonlinear functionThe adjustment amplitude of the number f (·) and the state s t+1 As passing action a t The adjusted fusion weight omega and the nonlinear function f (. Rewards r t Is positively correlated with the objective function G, i.e. based on the state s t+1 The higher the obtained health indication is, the corresponding reward r is obtained t The larger. After T step, the agent completes one-time optimization solution to be state s T+1 . The above procedure is a complete task Y, with the policy pi (a|s) updated after each task Y. The above task is repeated until the optimal fusion weight omega and nonlinear function f (·) are found (corresponding to) I.e. get the maximum rewards +.>The specific algorithm flow for solving the optimization problem based on reinforcement learning under strong robustness constraint is shown in the following table 1.
TABLE 1 Process Table for reinforcement learning based optimization problem solving
In the table 1, the contents of the components,the prize values are summarized for successive iterative processes.
In the above process, the policy update adopts a SAC (Soft activator-Critic) network structure based on strong robustness constraint, and the composition includes: policy network pi shown in FIG. 3 θ (a|s); v-value network V shown in FIG. 4 φ (s) target V value network V χ (s), and two Q value networksWherein (1)>And the network parameters are respectively corresponding to the network.
In the SAC algorithm, the optimization objective of the policy network is defined as follows:
wherein, psi is the accumulation buffer area, to correspond to Q value network->Is a function of the normalization of (a).
Based on the above description, the present invention proposes an improved strategy network optimization objective for adapting to the requirements of the optimization problem on solving robustness. Since the solution result corresponds to state s t To ensure stability of the solution, for similar state set groups s, s near There should be similar actions. Thus, the optimization objective of the policy network improves as:
wherein, beta is a weight adjustment factor for balancing optimization objectives.
Having generally described the algorithm flow, the design of agents and environments is specifically described.
The agent includes a state space Ω, an action space Λ, a reward r t Policy pi (a|s).
A. State space Ω
The state space Ω includes all candidatesA selected optimizing variable (ω U Γ) for representing the state s at each observation time t t ＝(s 1 ,s 2 ,...,s i ,...,s N-1 ,s N ) I.e. the fusion weight ω and the nonlinear function f (·) of the current choice. Specifically, ω is shown in FIG. 6 1 ,...,ω i ,...,ω N-1 Represents the fusion weight of N-1 features to be fused, and f (·) represents the nonlinear function selected at time t. In a complete task ψ, state s t (t=1, 2,) T-1) as a temporally selected fusion weight ω and nonlinear function f (·). State s T+1 Is a solution to the optimization problem after a complete task.
B. Motion space lambda
The action space Λ defines the action a taken at each observation time t t ＝(a 1 ,a 2 ,...,a i ,...,a N-1 ,a N ) Is not limited in terms of the range of (a). Specifically, Λ= { Λ 1 ，Λ 2 ，...，Λ i ，...，Λ N-1 ，Λ N }，a i ∈Λ i . Each action component a i Corresponding to adjusting the corresponding state component s in fig. 6 i . As shown in fig. 7, (a) 1 ,a 2 ,...,a i ,...,a N-1 ) In successive intervals [0,1 ]]Sampling, and realizing continuous adjustment of fusion weights. a, a N A nonlinear function f (·) is selected in the discrete space Γ. Wherein, in figure 6,is a real space of 1×n dimensions.
C. Policy pi (a|s)
Policy pi (a|s) represents the probability distribution of the selection action in the current state. Given state s t And action a t Selected from the probability distribution of pi (a|s) according to a random strategy. In the present invention, pi (a|s) is represented by pi using actor network as shown in FIG. 4 θ (a|s) represents. For the continuous motion component (a 1 ,a 2 ,...,a i ,...,a N-1 ) The actor network outputs the mean and variance of the corresponding probability distribution through the middle layer. For discrete action component a N The actor network outputs a discrete probability distribution through a softmax activation function.
D. Rewards r t
Rewards r t A strategy for encouraging agent learning to achieve the maximum objective function value G. In the course of agent learning strategies, it is desirable that pre-training agents be more prone to explore different strategies, so strategies that bring about small rewards promotion are not favored. Since objective function values are difficult to boost in the later stages of training, small rewards should be encouraged. The invention introduces an exponential function exp (), and realizes the balance of the two targets. The mathematical description of the reward function is as follows:
r temp,t ＝exp(SNR d )-bias,t＝1,...,T。
where bias is a constant bias value used to balance the prize value size. r is (r) temp,t To be a bonus function value.
Further, the invention rewards r corresponding to different observation moments t t Distinction is made. t=1..a.state s corresponding to T-1 t For driving reinforcement learning only, while the state s through the T step T The corresponding fusion weights and nonlinear functions are solutions to the optimization problem. Therefore, the reward value r is prevented from going through the T steps by distinguishing the reward value T Is influenced by r of intermediate process t And (3) counteracting. The final bonus function is described as follows:
in the present invention, the environment satisfies the markov process, the observed quantity coincides with the state quantity, and no unobservable portion exists. Specifically, given a current state s t Action a t The next time the state is the determined state, the mathematical description is as follows:
step 103: and inputting the degradation characteristics of the electromechanical system component into the association relation model to obtain one-dimensional health indication, so as to determine the health state of the electromechanical system component.
Based on the description, the method for determining the health state of the electromechanical system component provided by the invention realizes the health state evaluation of the key electromechanical component by mining and fusing the system observation data monitored by the electromechanical system sensor. The method starts with system observation data related to the components and utilizes the interactive learning advantage of reinforcement learning to construct an association relation model with nonlinear fusion characteristics. The method is not limited to the fusion mode of the observation data, and can adapt to any fusion mode. In addition, the sensor is insensitive to the number of sensors and training data, and the calculation complexity is low. The method can realize accurate component health state evaluation on the basis of the conventional system sensor layout, and has strong guiding significance for actual electromechanical system management.
Further, corresponding to the method for determining the health status of the electromechanical system component provided by the invention, the invention also provides the following implementation structure:
an electromechanical system component health status determination system, comprising:
the component degradation characteristic acquisition module is used for acquiring degradation characteristics which are characterized by observed quantity of a system sensor and are related to component degradation hidden variables; the degradation feature includes: the voltage effective value ratio of the input and output system, the current peak value ratio of the input and output system, the power loss of the system, the output voltage difference in a preset time interval and the characteristic quantity constructed based on the sensor monitoring data;
the association relation model construction module is used for constructing an association relation model;
and the health state determining module is used for inputting the degradation characteristic into the association relation model to obtain a one-dimensional health indication so as to determine the health state of the electromechanical system component.
An electronic device, comprising:
and a memory for storing a computer program.
And the processor is connected with the memory and is used for retrieving and executing the computer program to implement the method for determining the health state of the electromechanical system component.
Furthermore, the computer program in the above-described memory may be stored in a computer-readable storage medium when it is implemented in the form of a software functional unit and sold or used as a separate product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method of the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a U disk, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (5)
1. A method of determining a state of health of an electromechanical system component, comprising:
constructing an association relation model between the degradation hidden variable of the electromechanical component and the observed quantity of the electromechanical system sensor; the component degradation hidden variables comprise bearing wear of mechanical components, energy transmission efficiency of electronic components and other degradation physical quantities; the observed quantity of the system sensor comprises input and output voltage, current and conventional sensor output quantity;
estimating parameters of the association relation model; the estimation method is a reinforcement learning method based on strong robustness constraint;
inputting the observed quantity of the system sensor into the association relation to obtain a one-dimensional health indication for describing the degradation hidden variable of the component;
determining a health status of an electromechanical system component based on the one-dimensional health indication;
constructing an association relation model between a component degradation hidden variable and a system sensor observed quantity, which comprises the following steps:
acquiring degradation characteristics which are characterized by observed quantity of a system sensor and are related to a component degradation hidden variable; the degradation feature includes: the voltage effective value ratio of the input and output system, the current peak value ratio of the input and output system, the power loss of the system, the output voltage difference in a preset time interval and the characteristic quantity constructed based on the sensor monitoring data;
constructing an association relation model with nonlinear fusion characteristics, inputting degradation characteristics, and outputting one-dimensional health indications representing degradation hidden variables of the components; the nonlinear fusion characteristics comprise linear fusion based on fusion weights and nonlinear transformation of nonlinear function action;
determining an objective function, and abstracting the construction of the one-dimensional health indication into an optimization problem based on the degradation characteristic, the objective function and the association relation model structure;
solving the optimization problem based on a reinforcement learning method under strong robustness constraint to obtain a fusion weight and a nonlinear function which maximize the objective function value;
completing construction of the association relation model based on the fusion weight and the nonlinear function which make the objective function value maximum;
solving the optimization problem based on reinforcement learning under strong robustness constraint to obtain a fusion weight and a nonlinear function which maximize the objective function value, wherein the method specifically comprises the following steps:
converting the optimization problem into an interaction process of the agent and the environment based on reinforcement learning so as to obtain a fusion weight and a nonlinear function which maximize the objective function value;
the interaction process of the agent and the environment comprises the following steps:
the agent observes the environmental state at the current moment from the environment and determines a strategy based on the knowledge of the environment at the current moment; the environment state at the current moment is a fusion weight and a nonlinear function at the current moment;
selecting an action taken by the environmental state based on the policy, the action changing the environment to produce a new environmental state while deriving a prize value from the changed environment;
updating the strategy based on the reward value, selecting a new environment state to take action based on the updated strategy, and so on until the reward value reaches the maximum.
2. The method of claim 1, wherein the policy is updated based on the prize value using a SAC network architecture that incorporates strong robustness constraints.
3. An electromechanical systems component health status determination system adapted for use in an electromechanical systems component health status determination method as claimed in claim 1, comprising:
the component degradation characteristic acquisition module is used for acquiring degradation characteristics which are characterized by observed quantity of a system sensor and are related to component degradation hidden variables; the degradation feature includes: the voltage effective value ratio of the input and output system, the current peak value ratio of the input and output system, the power loss of the system, the output voltage difference in a preset time interval and the characteristic quantity constructed based on the sensor monitoring data;
the association relation model construction module is used for constructing an association relation model;
and the health state determining module is used for inputting the degradation characteristic into the association relation model to obtain a one-dimensional health indication so as to determine the health state of the electromechanical system component.
4. An electronic device, comprising:
a memory for storing a computer program;
a processor, coupled to the memory, for retrieving and executing the computer program to implement the method of determining the health of an electromechanical system component according to any of the claims 1-2.
5. The electronic device of claim 4, wherein the memory is a computer-readable storage medium.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310421466.8A CN116449135B (en) | 2023-04-19 | 2023-04-19 | Method and system for determining health state of electromechanical system component and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310421466.8A CN116449135B (en) | 2023-04-19 | 2023-04-19 | Method and system for determining health state of electromechanical system component and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116449135A CN116449135A (en) | 2023-07-18 |
| CN116449135B true CN116449135B (en) | 2024-01-30 |
Family
ID=87123371
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310421466.8A Active CN116449135B (en) | 2023-04-19 | 2023-04-19 | Method and system for determining health state of electromechanical system component and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116449135B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018075617A1 (en) * | 2016-10-19 | 2018-04-26 | SmartKable, LLC | Method and apparatus for predicting life cycle of a splice |
| CN109141881A (en) * | 2018-07-06 | 2019-01-04 | 东南大学 | A kind of rotating machinery health evaluating method of depth autoencoder network |
| CN112034789A (en) * | 2020-08-25 | 2020-12-04 | 国家机床质量监督检验中心 | Health assessment method, system and assessment terminal for key parts and complete machine of numerical control machine tool |
| CN112241595A (en) * | 2020-10-23 | 2021-01-19 | 西京学院 | Airborne power supply reliability increase performance degradation analysis method based on logistic regression |
| CN114239377A (en) * | 2021-11-10 | 2022-03-25 | 合肥赛为智能有限公司 | Method and system for evaluating health state of urban rail electromechanical equipment and storage medium |
| CN114819054A (en) * | 2022-03-11 | 2022-07-29 | 西北工业大学 | Power electronic system state monitoring method based on physical information neural network |
| CN114936504A (en) * | 2021-12-22 | 2022-08-23 | 山东浪潮工业互联网产业股份有限公司 | Equipment residual life prediction method and system based on Bayesian multi-source data fusion |
| CN115270591A (en) * | 2022-03-18 | 2022-11-01 | 北京航空航天大学 | Method and system for monitoring health state of electromechanical actuator based on composite space |
| CN115661246A (en) * | 2022-10-25 | 2023-01-31 | 中山大学 | Attitude estimation method based on self-supervision learning |
| CN115828611A (en) * | 2022-12-15 | 2023-03-21 | 北京控制工程研究所 | Method and device for evaluating health state of electromechanical component of spacecraft |
| CN115860106A (en) * | 2022-12-23 | 2023-03-28 | 四川物通科技有限公司 | Intelligent transformer substation capacitor fault early warning method based on deep Q learning |
| WO2023059811A1 (en) * | 2021-10-06 | 2023-04-13 | Google Llc | Constrained device placement using neural networks |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11783195B2 (en) * | 2019-03-27 | 2023-10-10 | Cognizant Technology Solutions U.S. Corporation | Process and system including an optimization engine with evolutionary surrogate-assisted prescriptions |
| CN111985155B (en) * | 2020-08-10 | 2023-08-04 | 武汉大学 | Circuit health state prediction method and system based on integrated deep neural network |
| US20220138613A1 (en) * | 2020-10-29 | 2022-05-05 | Qualcomm Incorporated | Reinforcement learning based scheme for tuning memory interfaces |
-
2023
- 2023-04-19 CN CN202310421466.8A patent/CN116449135B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018075617A1 (en) * | 2016-10-19 | 2018-04-26 | SmartKable, LLC | Method and apparatus for predicting life cycle of a splice |
| CN109141881A (en) * | 2018-07-06 | 2019-01-04 | 东南大学 | A kind of rotating machinery health evaluating method of depth autoencoder network |
| CN112034789A (en) * | 2020-08-25 | 2020-12-04 | 国家机床质量监督检验中心 | Health assessment method, system and assessment terminal for key parts and complete machine of numerical control machine tool |
| CN112241595A (en) * | 2020-10-23 | 2021-01-19 | 西京学院 | Airborne power supply reliability increase performance degradation analysis method based on logistic regression |
| WO2023059811A1 (en) * | 2021-10-06 | 2023-04-13 | Google Llc | Constrained device placement using neural networks |
| CN114239377A (en) * | 2021-11-10 | 2022-03-25 | 合肥赛为智能有限公司 | Method and system for evaluating health state of urban rail electromechanical equipment and storage medium |
| CN114936504A (en) * | 2021-12-22 | 2022-08-23 | 山东浪潮工业互联网产业股份有限公司 | Equipment residual life prediction method and system based on Bayesian multi-source data fusion |
| CN114819054A (en) * | 2022-03-11 | 2022-07-29 | 西北工业大学 | Power electronic system state monitoring method based on physical information neural network |
| CN115270591A (en) * | 2022-03-18 | 2022-11-01 | 北京航空航天大学 | Method and system for monitoring health state of electromechanical actuator based on composite space |
| CN115661246A (en) * | 2022-10-25 | 2023-01-31 | 中山大学 | Attitude estimation method based on self-supervision learning |
| CN115828611A (en) * | 2022-12-15 | 2023-03-21 | 北京控制工程研究所 | Method and device for evaluating health state of electromechanical component of spacecraft |
| CN115860106A (en) * | 2022-12-23 | 2023-03-28 | 四川物通科技有限公司 | Intelligent transformer substation capacitor fault early warning method based on deep Q learning |
Non-Patent Citations (2)
| Title |
|---|
| 一种新型深度自编码网络的滚动轴承健康评估方法;佘道明;贾民平;张菀;;东南大学学报(自然科学版)(05);全文 * |
| 面向大型载人航天器的系统健康状态评估方法研究;陈润锋;杨宏;;载人航天(05);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116449135A (en) | 2023-07-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109472110B (en) | Method for predicting residual service life of aeroengine based on LSTM network and ARIMA model | |
| CN111339712A (en) | Method for predicting residual life of proton exchange membrane fuel cell | |
| CN110889085A (en) | Intelligent wastewater monitoring method and system based on complex network multiple online regression | |
| Xu et al. | RUL prediction for rolling bearings based on Convolutional Autoencoder and status degradation model | |
| CN115018209B (en) | Long-term prediction method and equipment for operation error of digital electric energy metering system | |
| CN114397043B (en) | Multi-point temperature intelligent detection system | |
| Wang et al. | Evaluation and prediction method of rolling bearing performance degradation based on attention-LSTM | |
| Yu et al. | Pump remaining useful life prediction based on multi-source fusion and monotonicity-constrained particle filtering | |
| Zhang et al. | Health status assessment and prediction for pumped storage units using a novel health degradation index | |
| CN114266201B (en) | Self-attention elevator trapping prediction method based on deep learning | |
| CN117077327A (en) | Bearing life prediction method and system based on digital twin | |
| CN115630101A (en) | Hydrological parameter intelligent monitoring and water resource big data management system | |
| CN116449135B (en) | Method and system for determining health state of electromechanical system component and electronic equipment | |
| CN113487086A (en) | Method and device for predicting remaining service life of equipment, computer equipment and medium | |
| Cheng et al. | Interactive hybrid model for remaining useful life prediction with uncertainty quantification of bearing in nuclear circulating water pump | |
| CN115616163A (en) | Gas accurate preparation and concentration measurement system | |
| CN112257893A (en) | Complex electromechanical system health state prediction method considering monitoring error | |
| CN115952916A (en) | Artificial intelligence-based wind power prediction error correction method, device and equipment | |
| CN114358244B (en) | Big data intelligent detection system of pressure based on thing networking | |
| CN116304587A (en) | Rolling bearing degradation trend prediction method based on CAE and AGRU | |
| CN114722721A (en) | Power transformer health state evaluation method based on deep coding and decoding convolutional network | |
| CN114372618A (en) | Student score prediction method and system, computer equipment and storage medium | |
| CN113591391A (en) | Power load control device, control method, terminal, medium and application | |
| Lu et al. | Relation-aware attentive neural processes model for remaining useful life prediction | |
| CN118036435A (en) | Method, device and storage medium for predicting remaining usable life of compressor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |