Method and computing system for performing prognostic health analysis on assets

Abstract

To perform the prognostic health analysis (11-13) of the asset, a plurality of independent stochastic simulations are performed using transition probabilities of a discrete Markov chain model. A prognostic asset health state evolution over a time horizon is calculated from the plurality of independent stochastic simulations. An output is generated based on the calculated prognostic asset health state evolution.

Description

Technical Field

[0001] This invention relates to techniques for assessing asset health. More particularly, it relates to methods and apparatus for prognostic assessment of asset health. Background Technology

[0002] Power systems (e.g., generation, transmission, and / or distribution systems) and industrial systems both include assets. Transformers, electric generators, and distributed energy resources (DER) units are examples of such assets. Assets deteriorate during operation. For planning purposes, scheduling maintenance or replacement work, it is necessary to estimate the remaining useful life (RUL) of the asset.

[0003] RUL estimation can be based on sensor data from a group of assets of the same or similar type as the asset whose RUL is to be estimated. Sensor data can be tagged with fault characteristics to indicate whether the sensor data corresponds to a normal operating, deteriorated, or failed state of the asset. Combining different types of sensor data for RUL estimation can be challenging when, for example, manufacturer-provided or accessory sensors installed with the asset can be used for different assets within the group.

[0004] The paper "Remaining Useful Life Prediction Using Elliptical Basis Function Network and Markov Chain" by Y. Yu et al. (World Academy of Science, Engineering and Technology, 47, 2010) introduces a method for predicting remaining useful life using elliptical basis function (EBF) networks and Markov chains. The EBF structure is trained using a modified expectation-maximization (EM) algorithm to account for the set of missing covariates. Internal covariates do not require explicit extrapolation, and the Markov chain is constructed to represent the evolution of external covariates. Summary of the Invention

[0005] Improved technologies are needed to predict the temporal evolution of asset degradation. In particular, there is a need for a technology that allows for prognostic predictions of asset health degradation without requiring sensor data captured for the specific asset being predicted. Alternatively or additionally, there is a need for a technology that can make predictions using historical sensor data from the asset group, even when different types of sensor data are available for different assets within a group.

[0006] According to embodiments of the present invention, a method and computing system as described in the independent claim are provided. Dependent claims define preferred embodiments.

[0007] According to embodiments of the present invention, discrete Markov chain models can be used to predict asset performance degradation. A large number of simulations can be performed using the Markov Chain Monte Carlo (MCMC) method. The Markov chain model can have a set of discrete states corresponding to different asset health states.

[0008] The potentially large amount of historical sensor data captured can be compressed into a small number of transition probabilities. For illustration, a Markov chain model can be constructed such that only two or three transition probabilities control the transitions between states, which can indicate a healthy asset state, a deteriorated asset state where the asset is still operational, and a failed asset state where the asset has failed. This invention can also be applied to situations where little or no historical data is available. The transition probabilities of the Markov chain model can then be set by human experts.

[0009] The transition probabilities between states in a Markov chain model can be determined based on sensor data labeled with fault characteristics and captured on an asset group.

[0010] Different transition probabilities can be used in simulations running in parallel. Different transition probabilities can be associated with different operating conditions and / or environmental conditions.

[0011] Information about variance, confidence intervals, or other reliability metrics can be determined based on simulations.

[0012] The MCMC technique, utilizing a small state space and a limited number of transition probabilities, allows for the performance of prognostic asset health analyses across a wide range of timeframes, including those that can last for years or longer. Various other simulation methods, designed to model asset behavior in greater detail, can be applied to assess asset degradation on shorter timescales but may experience instabilities associated with asset health analyses of power system assets or industrial assets over longer periods.

[0013] A method for performing prognostic health analysis on an asset, particularly for determining the remaining useful life (RUL) or probability of failure (PoF), according to an embodiment, includes performing multiple independent stochastic simulations using transition probabilities from a discrete Markov chain model. The discrete Markov chain model has a state space containing a set of asset health states. Each of the multiple independent stochastic simulations represents the future evolution of the state space of the discrete Markov chain model within the prognostic range. The method may further include calculating the prognostic asset health state evolution within the prognostic range based on the multiple independent stochastic simulations. The method may also include generating an output based on the calculated prognostic asset health state evolution.

[0014] The assets can be power system assets.

[0015] The assets can be industrial assets.

[0016] Calculating the evolution of prognostic asset health status can include calculating RUL.

[0017] Calculating the evolution of prognostic asset health status can include calculating the probability of failure of an asset within the prognostic range as a function of time.

[0018] The method may also include calculating confidence information on the evolution of prognostic asset health status as a function of time within the prognostic range based on multiple independent random simulations.

[0019] Output can also be generated based on confidence information.

[0020] Confidence information can include the time evolution of the confidence interval within the prognostic range.

[0021] The method may also include calculating a function of time within the prognostic range based on variance information used for the evolution of prognostic asset health status from multiple independent random simulations.

[0022] Output can also be generated based on variance information.

[0023] Variance information can include the temporal evolution of the variance interval within the prognostic range.

[0024] Confidence or variance information may include the time evolution of the lower bound and the time evolution of the upper bound.

[0025] The lower bound can be associated with the first set of transition probabilities, and the upper bound can be associated with a second set of transition probabilities that are different from the first set of transition probabilities.

[0026] The output may include a representation of the evolution of asset health status, alarms or warnings generated based on the evolution of asset health status, and / or control signals for controlling the operation of the asset based on the evolution of asset health status.

[0027] The method may include deriving a first set of transition probabilities based on first sensor data of an asset operating under first conditions (e.g., first operating and / or environmental conditions), and deriving a second set of transition probabilities based on second sensor data of an asset operating under second conditions different from the first operating conditions (e.g., second operating and / or environmental conditions).

[0028] The state space can consist of three, four, or more states.

[0029] The state space may include at least one state in which the operation of an asset is not adversely affected by a fault.

[0030] The state space can include at least one state in which the operation of an asset is adversely affected by a failure, but the asset continues to operate.

[0031] The state space can include the state in which an asset is inoperable due to a failure.

[0032] Calculating the evolution of prognostic asset health status may include calculating the probability distribution in the state space over multiple time periods within the prognostic range and mapping that probability distribution to a scalar.

[0033] A scalar can be the probability that an asset will be in a state where it is rendered inoperable due to a failure, as determined by multiple independent simulations.

[0034] The evolution of prognostic asset health status can be obtained as a scalar time evolution.

[0035] The method may also include determining the transfer probability based on historical data, including sensor data for multiple assets.

[0036] Sensor data can be labeled with fault features that indicate at what time the corresponding asset is in which state in the state space of the Markov chain model.

[0037] Fault features can be automatically calculated, or they can be assigned based on input received from human experts, or a combination of the two methods can be used to assign fault features.

[0038] Determining the transition probability may include calculating a time-dependent scalar function based on sensor data used for multiple assets.

[0039] A heuristic approach that uses sensor measurements as input and outputs a scalar function can be used to compute the scalar function.

[0040] Determining transition probabilities can include using time-dependent scalar functions to identify transitions within the state space of a Markov chain model.

[0041] Determining transition probabilities can include calculating transition probabilities based on transitions within the state space of a Markov chain model.

[0042] The determination process may include comparing a scalar function with one or more thresholds.

[0043] Scalar functions can represent health indices, thus indicating the severity of deterioration.

[0044] The transition probability can be determined independently for different groups of sensor data, thereby indicating different operating conditions and / or environmental conditions of assets in the group.

[0045] The method may include determining multiple sets of transition probabilities, each set of transition probabilities being associated with different operating conditions and / or environmental conditions.

[0046] Although multiple sets of transition probabilities can be determined for different operating conditions and / or environmental conditions, the state space of a Markov chain model can remain the same.

[0047] The method may include selecting one set of multiple transition probabilities based on the operational and / or environmental conditions that the asset is to undergo, for use in performing a simulation.

[0048] The method may include performing multiple independent stochastic simulations such that one or more first simulations are performed using a first set of transition probabilities associated with a first operating condition and / or a first environmental condition, and one or more second simulations are performed using a second set of transition probabilities associated with a second operating condition and / or a second environmental condition, wherein the first transition probabilities are different from the second transition probabilities.

[0049] Different operating conditions can indicate different loads, different voltages, different currents, different operating points, and different insulating fluids, but are not limited to these.

[0050] Different environmental conditions can indicate different temperatures and relative humidity, but are not limited to these.

[0051] Multiple independent stochastic simulations can be Markov chain Monte Carlo (MCMC) simulations.

[0052] Markov chains can be homogeneous. Transition probabilities can be time-independent.

[0053] A Markov chain can have order 1, meaning that transitions can be based on the current state of the Markov chain model, regardless of previous transitions to that state.

[0054] Markov chain models allow for a monotonically ordered qualitative interpretation of states, meaning that two states can always be compared in terms of the severity of degradation.

[0055] Each state in the state space can have a non-zero transition probability to at most one other state in the state space (which describes a more severe degradation) and a non-zero transition probability to itself.

[0056] Markov chain models allow states in the state space that do not correspond to asset failures to have non-zero transition probabilities to only one other state in the state space.

[0057] Markov chain models ensure that the state space corresponding to the failure of an asset does not have any non-zero transition probability to any state other than the state itself.

[0058] Markov chain models can be finite Markov chain models.

[0059] The state space can consist of three, four, or more states.

[0060] The state space can consist of n states, where n is equal to three, four or greater than four. For the n-1 states of the state space corresponding to an operable asset, the transition probability to only one other state in the state space is non-zero, and for the states corresponding to a faulty asset, there is no non-zero transition probability to any other state in the state space.

[0061] The method may also include receiving sensor measurement data captured during the operation of the asset.

[0062] The method may also include updating the evolution of prognostic asset health status based on received sensor measurement data.

[0063] Multiple simulations can include simulations for different environments and / or operational scenarios.

[0064] Multiple simulations can be executed in parallel.

[0065] Multiple simulations can be executed simultaneously.

[0066] This method can be implemented by a computer.

[0067] This method can be executed by at least one integrated circuit.

[0068] This method can be executed by at least one integrated circuit of the central controller of a distributed control system.

[0069] This method can be executed by at least one integrated circuit of the local controller of a distributed control system.

[0070] The method may include receiving information about the transition probability on a communication network by at least one integrated circuit.

[0071] Assets can be power transformers, distributed energy DER units, or power generators.

[0072] The prognosis range can be 1 year or longer, 2 years or longer, 3 years or longer, 4 years or longer, 5 years or longer, 10 years or longer, 15 years or longer, or 20 years or longer.

[0073] The prognosis range may be 1 week or longer, 1 month or longer, etc.

[0074] The prognostic range can be measured over multiple periods and can include multiple periods, such as a certain number of flight periods, ship route periods, train route periods, etc.

[0075] A method for operating and / or maintaining assets includes performing a prognostic asset health analysis on the assets using a method according to an embodiment, and automatically taking control or output actions based on the prognostic asset health analysis.

[0076] Control or output actions may include performing at least one of the following: generating an alarm or warning based on the calculated prognostic asset health status evolution; generating a control signal for controlling the operation of the asset based on the calculated prognostic asset health status evolution; scheduling asset downtime based on the calculated prognostic asset health status evolution; scheduling maintenance work based on the calculated prognostic asset health status evolution; scheduling replacement work based on the calculated prognostic asset health status evolution; or changing the maintenance interval based on the calculated prognostic asset health status evolution.

[0077] Control or output actions may include outputting information via an interface about the probability of failure as a function of operating time, information about scheduled or rescheduled maintenance intervals, or information about scheduled replacement intervals.

[0078] The computational system operable according to the invention for performing prognostic health analysis on an asset is configured to perform multiple independent stochastic simulations using transition probabilities of a discrete Markov chain model, wherein the discrete Markov chain model has a state space comprising a set of asset health states, and wherein each of the multiple independent stochastic simulations represents the future evolution of the state space of the discrete Markov chain model within a prognostic range. The computational system is configured to compute the prognostic asset health state evolution within the prognostic range based on the multiple independent stochastic simulations. The computational system is configured to control the generation of outputs based on the computed prognostic asset health state evolution.

[0079] The computing system may include a local controller having one or more integrated circuits, which is operable to perform independent stochastic simulations and compute the evolution of prognostic asset health status. The one or more integrated circuits (ICs) are operable to perform the operations described in detail herein.

[0080] The computing system may include a central controller having one or more integrated circuits, which is operable to perform independent stochastic simulations and compute the evolution of prognostic asset health status. The one or more ICs are operable to perform the operations described in detail herein.

[0081] The assets can be power system assets.

[0082] The assets can be industrial assets.

[0083] The computational system can be operable to enable the computation of the evolution of the prognostic asset health status, which may include the computation of RUL.

[0084] The computational system can be operable such that calculating the evolution of the prognostic asset health status can include calculating the probability of failure of the asset within the prognostic range as a function of time.

[0085] The computational system is operable to calculate confidence information on the evolution of prognostic asset health status as a function of time within the prognostic range, based on multiple independent stochastic simulations.

[0086] The computing system is operable to the point that it can also generate output based on confidence information.

[0087] The computing system can be operable such that confidence information can include the temporal evolution of the confidence interval within the prognosis range.

[0088] The computational system is operable to calculate variance information of the evolution of prognostic asset health status as a function of time within the prognostic range based on multiple independent stochastic simulations.

[0089] The computing system is operable to the point that it can also generate output based on variance information.

[0090] The computing system may be operable such that the output may include a representation of the evolution of asset health status, alarms or warnings generated based on the evolution of asset health status, and / or control signals for controlling the operation of assets based on the evolution of asset health status.

[0091] The computational system can be operable such that variance information can include the temporal evolution of variance intervals within the prognostic range.

[0092] The computational system can be operable such that the confidence information or variance information can include the time evolution of the lower boundary and the time evolution of the upper boundary.

[0093] The computational system can be operable such that the lower boundary can be associated with a first set of transition probabilities and the upper boundary can be associated with a second set of transition probabilities that are different from the first set of transition probabilities.

[0094] The computing system may be operable to derive a first set of transition probabilities based on first sensor data of an asset operating under first conditions (e.g., first operating and / or environmental conditions), and to derive a second set of transition probabilities based on second sensor data of an asset operating under second conditions different from the first operating conditions (e.g., second operating and / or environmental conditions).

[0095] The computing system can be operable such that the state space can consist of three, four, or more states.

[0096] The computing system can be operable such that the state space can include at least one state in which the operation of the asset is not adversely affected by the failure.

[0097] The computing system can be operable such that the state space can include at least one state in which the operation of an asset is adversely affected by a fault, but the asset continues to operate.

[0098] The computing system can be operable such that the state space can include the state in which an asset is inoperable due to a failure.

[0099] The computational system can be operable such that computing the evolution of the health status of prognostic assets can include computing a probability distribution in the state space over multiple time periods within the prognostic range and mapping that probability distribution to a scalar.

[0100] The computational system can be operable such that a scalar can be the probability, as determined by multiple independent simulations, of an asset being in a state in which the asset is rendered inoperable due to a fault.

[0101] The computational system is operable to enable the evolution of prognostic asset health status as a scalar temporal evolution.

[0102] The computing system is operable to determine the probability of transition based on historical data, including sensor data for multiple assets.

[0103] The computing system can be operable to enable sensor data to be labeled with fault features that indicate when the corresponding asset is in which state in the state space of the Markov chain model.

[0104] The computing system can be operated to automatically calculate fault characteristics or receive input from human experts to determine fault characteristics.

[0105] The computing system can be operable such that determining the transition probability may include calculating a time-dependent scalar function based on sensor data for multiple assets.

[0106] The computing system is operable to enable the computation of scalar functions using heuristics that take sensor measurements as inputs and output scalar functions.

[0107] The computational system can be operable such that determining the transition probabilities can include identifying transitions within the state space of a Markov chain model using a time-dependent scalar function.

[0108] The computational system can be operable such that determining the transition probabilities can include calculating the transition probabilities by including transitions within the state space based on a Markov chain model.

[0109] The computational system may be operable such that determining the transition probability may include comparing a scalar function with one or more thresholds.

[0110] Scalar functions can represent health indices, thus indicating the severity of deterioration.

[0111] The computing system is operable to enable the determination of transition probabilities independently for different sets of sensor data, thereby indicating different operating conditions and / or environmental conditions of assets in the group.

[0112] The computational system can be operated to determine multiple sets of transition probabilities, each set of transition probabilities being associated with different operating conditions and / or environmental conditions.

[0113] Although multiple sets of transition probabilities can be determined for different operating conditions and / or environmental conditions, the state space of a Markov chain model can remain the same.

[0114] The computing system can be operated to select one of multiple sets of transition probabilities based on the operational and / or environmental conditions that the asset will undergo for use in performing simulations.

[0115] The computing system is operable to perform multiple independent stochastic simulations such that one or more first simulations are performed using a first set of transition probabilities associated with a first operating condition and / or a first environmental condition, and one or more second simulations are performed using a second set of transition probabilities associated with a second operating condition and / or a second environmental condition, wherein the first transition probabilities are different from the second transition probabilities.

[0116] The computing system may be operable to indicate different loads, voltages, currents, operating points, and insulating fluids under different operating conditions, but is not limited to these.

[0117] Different environmental conditions can indicate different temperatures and relative humidity, but are not limited to these.

[0118] The computational system is operable such that multiple independent stochastic simulations can be Markov chain Monte Carlo (MCMC) simulations.

[0119] The computational system can be operable such that the Markov chain can be homogeneous. The transition probabilities can be time-independent.

[0120] The computational system can be operable such that the Markov chain can have order 1, that is, the transition can be based on the current state of the Markov chain model, regardless of the previous transitions to that state.

[0121] The computational system can be operable such that each state in the state space can have a non-zero transition probability to at most one other state in the state space.

[0122] The computational system can be operable such that the Markov chain model allows states in the state space that do not correspond to asset failures to have non-zero transition probabilities to only one other state in the state space.

[0123] The computational system can be operable such that the Markov chain model allows the state in the state space corresponding to an asset failure to have no non-zero transition probabilities to states other than the state itself.

[0124] The computational system can be operable such that the Markov chain model is a finite Markov chain model.

[0125] The computational system can be operable such that the state space can consist of four or three states.

[0126] The computing system is operable such that the state space can consist of n states, where n is equal to three, four or greater than four, wherein for n-1 states in the state space corresponding to operable assets, the transition probability to only one other state in the state space is nonzero, and for states corresponding to faulty assets, there is no nonzero transition probability to any other state in the state space.

[0127] The computing system is operable to receive sensor measurement data captured during the operation of the asset.

[0128] The computing system is operable to update the evolution of prognostic asset health status based on received sensor measurement data.

[0129] The computing system can be operable such that multiple simulations can include simulations for different environments and / or operational scenarios.

[0130] The computing system can be operated to enable the parallel execution of multiple simulations.

[0131] The computing system can be operated to execute multiple simulations simultaneously.

[0132] The computing system has at least one integrated circuit to perform the operation. The operation can be performed in a distributed computing system, such as in a distributed control system, and / or using a cloud-based computing system.

[0133] The computing system may have an interface for receiving information about transition probabilities via a communication network.

[0134] Assets can be power transformers, distributed energy DER units, or power generators.

[0135] The prognosis range can be 1 year or longer, 2 years or longer, 3 years or longer, 4 years or longer, 5 years or longer, 10 years or longer, 15 years or longer, or 20 years or longer.

[0136] The prognosis range can be 1 week or longer, 1 month or longer, etc.

[0137] The prognostic range can be measured over multiple periods and can include multiple periods, such as a certain number of flight periods, ship route periods, train route periods, etc.

[0138] A control system for an asset includes: a computing system for performing a prognostic asset health analysis on the asset using a method according to an embodiment; and an output interface for automatically triggering or implementing control or output actions based on the prognostic asset health analysis.

[0139] Control or output actions may include performing at least one of the following: generating an alarm or warning based on the calculated evolution of the prognostic asset health status; generating a control signal for controlling the operation of the asset based on the calculated evolution of the prognostic asset health status; scheduling downtime for the asset based on the calculated evolution of the prognostic asset health status; scheduling maintenance work based on the calculated evolution of the prognostic asset health status; scheduling replacement work based on the calculated evolution of the prognostic asset health status; and changing the maintenance interval based on the calculated evolution of the prognostic asset health status.

[0140] Control or output actions may include outputting information via an interface about the probability of failure as a function of operating time, information about scheduled or rescheduled maintenance intervals, or information about scheduled replacement intervals.

[0141] Industrial or power systems include assets and computing systems used to perform prognostic asset health analyses on those assets.

[0142] The computing system can be a distributed controller for controlling assets in industrial or power systems.

[0143] Various effects and advantages are associated with this invention. Only a small number of parameters are required when using a Markov chain model with a discrete state space and non-zero transition probabilities to at most one other state corresponding to the next severity of degradation. This is particularly true when the Markov chain model has a state space that can consist of a fairly small number of states (e.g., three or four states, but optionally more), only a small number of transition probabilities between states are needed to perform the method. These transition probabilities can be set based on user input or automatically determined based on sensor data with fault characteristics, even if different types of sensor data are available for different assets in the swarm.

[0144] Randomized simulation techniques based on Markov chain models can be used to determine RUL curves or other prognostic asset health information starting from sensor data of any dimension.

[0145] The distribution of RUL curves or other prognostic asset health information can be determined based on stochastic simulations for different environmental and / or operating conditions. This allows for quantitative information not only on the expected RUL curve but also on its confidence interval. The precision and / or variance of the RUL distribution can be quantified and output.

[0146] High-speed computation can be achieved by integrating multiple stochastic simulations into a simulation step that allows the stochastic simulations to be executed in parallel.

[0147] Other information can be generated, such as quantitative information about the variance or confidence interval of the RUL curve, which can be used in descriptive tools or applications. Attached Figure Description

[0148] The subject matter of the invention will be explained in more detail with reference to the preferred exemplary embodiments shown in the accompanying drawings, wherein:

[0149] Figure 1 This is a schematic diagram of a power system having a computing system according to an embodiment.

[0150] Figure 2 This is a schematic diagram of a power system having a computing system according to an embodiment.

[0151] Figure 3 This is a diagram illustrating the Markov chain model used in the embodiments.

[0152] Figure 4 This is a flowchart of a method according to an embodiment.

[0153] Figure 5 This is a graph illustrating exemplary output generated by the method and computing system according to the embodiments.

[0154] Figure 6 It is a bar graph showing the time evolution of the state occupancy probability.

[0155] Figure 7 This is a graph illustrating exemplary output generated by the method and computing system according to the embodiments.

[0156] Figure 8A and Figure 8B This is a graph showing the results of a prognostic asset health analysis combined with the observed asset health status.

[0157] Figure 9 This is a flowchart of a method according to an embodiment.

[0158] Figure 10 This is a block diagram of a computing system according to an embodiment. Detailed Implementation

[0159] Exemplary embodiments of the invention will be described with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements. Although some embodiments will be described in the context of power system assets (e.g., distributed energy resources (DER) units or transformers), the embodiments are not limited thereto. Unless otherwise specifically indicated, features of the embodiments may be combined with each other.

[0160] Figure 1 and Figure 2 This is a schematic diagram of power systems 10 and 15. Power systems 10 and 15 include multiple assets. Assets may include generators (such as distributed energy resources (DER) units 11-13, 16-18), transformers, or other power system assets.

[0161] Power systems 10 and 15 include a control system comprising local controllers 21-23, each associated with an asset. The control system may include a central system 20. The central system 20 may communicatively connect to the local controllers. The central system 20 may also communicate with a remote (e.g., cloud-based) server system 24.

[0162] As will be described in more detail below, local controllers 21-23, central system 20, and / or remote server system 24 may be operable to perform prognostic asset health analysis using a Markov chain model. The Markov chain model may have specific configurations, as described below. Local controllers 21-23, central system 20, and / or remote server system 24 may be operable to perform multiple independent stochastic simulations, particularly Markov chain Monte Carlo (MCMC) simulations, to perform prognostic asset health analysis.

[0163] Local controllers 21-23, central system 20, and / or remote server system 24 can use the results of prognostic asset health analysis to schedule downtime, maintenance work, replacement work, or automate control operations. Local controllers 21-23, central system 20, and / or remote server system 24 can be operated to generate and output control or output data. Output can be provided via a human-machine interface (HMI) 26. The HMI can be connected to local controllers 21-23, central system 20, and / or remote server system 24 via the Internet or another wide area network (WAN).

[0164] For reference Figures 3 to 10 To explain in more detail, prognostic asset health analysis can involve simulating the temporal evolution of assets. MCMC simulation and further processing of MCMC results can be performed at local controllers 21-23. This facilitates the incorporation of local sensor measurements to update the prognostic asset health analysis.

[0165] The techniques described herein can also be used when no sensor data is available for the specific asset targeted for performing a prognostic asset health analysis. For illustration, even when no (not yet) sensor data is available, a historical data repository (including sensor data and fault information captured for similar assets, preferably for assets of the same type as the asset targeted for the prognostic asset health analysis) can still be used to calculate the remaining useful life (RUL) curve or other prognostic asset health predictions for the asset.

[0166] Figure 3 This is a graph of a Markov chain model that can be used in the methods and computational systems according to the embodiments. The state space of the Markov chain model consists of a set of n states S1, ..., Sn. In the present case, n = 4. However, it is possible to use a state space with a different number of states (e.g., n = 3, n = 5, or n > 5).

[0167] The states in the state space can be ordered as follows: the severity of asset health degradation increases from S1 to S2, from S2 to S3, and so on. That is, all states except the final state of the Markov chain model can be followed by another state representing a more severe degradation. The final state of the Markov chain model can represent the most severe degradation.

[0168] A Markov chain model can be constructed as follows: first state, second state, ..., (n-1)th state (41-43) with a non-zero transition probability p to only one other state in the state space. 12 p 23 p 34 The nth state 44 does not have a non-zero transition probability to a state other than itself.

[0169] To illustrate, the Markov chain model can have a finite transition probability p from the first state 41 to the second state 42. 12 The transition from state 42 back to state 41 has a zero probability. Maintaining state 41 in the iterations of the stochastic simulation has a probability of 1-p. 12 .

[0170] The Markov chain model allows for a finite transition probability p from the second state 42 to the third state 43. 23 The transition from state 43 back to state 42 has a zero probability. Maintaining state 42 in the iterations of the stochastic simulation has a probability of 1-p. 23 .

[0171] The Markov chain model allows for a finite transition probability p from state 43 to state 44. 34The transition from state 44 back to state 43 has a zero probability. Maintaining state 43 in the iterations of the stochastic simulation has a probability of 1-p. 34 .

[0172] The final state 44 of the Markov chain model can correspond to the state in which the asset has failed to the point of being no longer operational.

[0173] The other states 41-43 in the state space can correspond to different degrees of degradation.

[0174] For illustration, the first state 41 (which may also be referred to as the "unknown" state S1) can correspond to a state in which there is no known deterioration of the asset that would affect the operation of the asset. For illustration, the first state 41 can correspond to a state in which no fault is recorded or there is no fault available for recording.

[0175] The second state 42 (which may also be referred to as the "initial" state S2) corresponds to detectable faults that are of low severity so that they do not immediately affect the performance of the asset. Such initial faults are typically characterized by a short mean time to repair (MTTR), low repair costs, and a minor impact on overall performance. However, if not properly maintained, initial faults can escalate into more serious deterioration faults.

[0176] State 43 (also known as "deterioration" state S3) corresponds to a failure pattern that significantly reduces system performance but does not cause immediate asset downtime. Typically, this failure is caused by component damage. If left untreated, degradation will eventually lead to a critical failure.

[0177] State 44 (which can also be called “critical” state S4) corresponds to the most severe failure mode, which results in an immediate and complete downtime of the asset. It is typically characterized by a long and costly MTTR (due to total loss of production).

[0178] The transition probabilities of a Markov chain model can be received from a human expert via a user interface or determined using historical data, as described below. Different sets of transition probabilities p can be used. 12 p 23 p 34 To illustrate, the transition probabilities p of N>1, and especially N>2, can be considered. 12 p 23 p 34 Used to simulate the evolution of asset health status under different environmental and / or operating conditions.

[0179] Figure 4This is a flowchart of method 50 according to an embodiment. Method 50 can be automatically executed by one or more ICs from local controllers 21-23, central system 20, and / or remote server system 24.

[0180] In step 51, perform MCMC simulations or other stochastic simulations of the Markov chain model. Step 51 may include executing more than 100, more than 1000, more than 2000, more than 5000, more than 10000, more than 50000, more than 100000, more than 500000, or one million or more simulations. There is no upper limit to the number of simulations as computing power increases. These simulations can be executed in parallel.

[0181] Although any set of transition probabilities p in the Markov chain model can be used... 12 p 23 p 34 Perform a large number of simulations (e.g., more than 100 or more than 1000), but the transition probabilities do not need to be the same for all simulations. The transition probabilities p for different groups... 12 p 23 p 34 It can be used to quantitatively assess the impact of different operating conditions and / or environmental conditions.

[0182] Simulations can be performed within a specific timeframe. This timeframe can vary depending on the specific asset. For power system assets such as transformers, typical useful lives exceed 10 years, 20 years, or even longer. Therefore, stochastic simulations can be performed over timeframes exceeding 10 years, 20 years, or even longer. The timeframe can also be shorter depending on the asset. For illustration, the prognostic timeframe can be one week or longer, one month or longer, etc. The prognostic timeframe can be measured across multiple periods and can include multiple periods, such as a certain number of flight periods, ship route periods, train route periods, etc.

[0183] The initial state of the simulation can be selected based on available information about the asset. If no information about the asset is available, the simulation can begin entirely from the first state 41, which does not contain information about detectable faults. If information about the asset is available (e.g., sensor data collected after installation), that sensor data can be used to initialize the simulation. The initial state distribution for various MCMCs or other stochastic simulations can be selected based on whether the collected sensor data indicates the absence of identifiable faults affecting asset performance or the presence of detectable problems affecting asset performance.

[0184] Furthermore, initialization can be probabilistic. For example, if the available information is uncertain whether an asset is in state 41 or state 42 (where the chance of being in either state is equal), the system can be initialized using a Bayesian prior distribution such that the probability of the asset being in state 41 is 50% and the probability of the asset being in state 42 is 50%. Other probabilistic initializations with more states and different probabilities are also possible.

[0185] In step 52, the results of the stochastic simulation are processed. This may include calculating, as a function of time within a time frame, the probability that the Markov chain model has evolved to a critical state corresponding to an inoperable asset. The processing at step 53 may include calculating the time evolution of a health index, which, for any time within the prognostic time frame, represents the probability of the Markov chain model being in a first, second, ... nth state 41-44 of the Markov chain model. This processing may include calculating the RUL curve, or another output indicating the probability of asset failure, as a function of operating time.

[0186] In step 53, output may be generated. The output may include information about the asset's remaining useful life as a function of time. The output may include information about the asset's failure probability as a function of time. The output may include control and / or output data obtained through further processing of the simulation results, such as arrangements for asset inspection, maintenance, or replacement work.

[0187] Figure 5 This is a schematic diagram of output 60, which can be automatically generated and output. Output 60 indicates the probability that the Markov chain model has evolved into a critical state corresponding to an inoperable asset. For each of the multiple time points within the time range, output 60 can be determined by calculating the simulation score of the critical state S4 in the state space where the Markov chain model is located.

[0188] Additional or alternative outputs can be generated. For illustration, RUL curves or other information indicating asset deterioration can be processed to automatically schedule inspections, maintenance or replacement work, output scheduling information to operators, and / or automatically schedule downtime.

[0189] Alternatively or additionally, threshold comparisons or other triggers can be used to process RUL curves or other information indicating asset deterioration to determine whether and when to output alarms, warnings or other signals to operators.

[0190] Figure 6The diagram shows the overall random distributions 61-64 of the various states S1-S4 in the state space of the Markov chain model. Distribution 61 corresponds to the first time point where most of the Markov chain model simulations are still in state S1, which corresponds to assets without detectable degradation. Distributions 62 and 63 correspond to the subsequent second and third time points where states S2 and S3 have become more densely packed, corresponding to early or more severe degradation. Distribution 64 corresponds to the later fourth time point where the critical state S4 (which corresponds to asset shutdown) is most densely packed, reflecting that by then the asset is more likely to be in an inoperable state than an operable one.

[0191] While the prognostic prediction of asset health can be obtained based on the probability that the asset will be in the critical state S4 (which is the final state of the Markov chain model), the output processed from the results of stochastic simulation can be based on all probabilities p1, p2, ..., pn of the Markov chain model being in the corresponding first state, second state, ..., nth state, as determined by the stochastic simulation. n .

[0192] To illustrate, for any time j within the time range of performing the stochastic simulation, the scalar function d(j) = ∑ i＝1，...，n p i (j)×m i (1),

[0193] Where, p i (j) represents the probability that the Markov chain model at time j will be in state i, as determined by random simulation, and where m i Let m be the scalar value of a monotonic function, specifically a strictly monotonic function, of state label i. For illustration, all m... i We can choose from intervals such that m1≤m2≤…≤m n In particular, it makes m1 < m2 < ... < m n .

[0194] Even if the degradation that leads to a decrease in RUL has not yet caused the asset to reach the critical state S4, the degradation can be more fully reflected by the output function d(j) or by the information derived from the function d(j).

[0195] The function d(j) indicates deterioration, and when d(j) is restricted to values ​​between 0 and 1, the function d(j) can be correlated with the health index h(j) through h(j) = 1 - d(j).

[0196] Figure 7The output of curve 80 is shown, indicating asset degradation as a function of time, as determined by stochastic simulation. Curve 80 can be calculated based on a Markov chain model, as determined by stochastic simulation, to determine all probabilities p1, p2, ..., pn of the corresponding first, second, ..., nth states. n The evolution over time.

[0197] The various states of a Markov chain model can be associated with multiple intervals 71-74. For illustration, for a health index h within interval 71, it can be determined that the asset is in a state S1 where there is no known degradation. For a health index h within interval 72, it can be determined that the asset is in a state S2 where there is no initial degradation that does not affect performance. For a health index h within interval 73, it can be determined that the asset is in a state S3 where there is more severe degradation that affects performance but will not cause the asset to immediately shut down. For a health index h within interval 74, it can be determined that the asset is in a state S4 where the state is critical and will cause the asset to shut down immediately.

[0198] Thresholds TH1, ... TH n-1 The upper and lower boundaries of the interval 71-74 can be defined. When initializing a random simulation of assets, thresholds TH1, ... TH can be used. n-1 The comparison can be performed as follows. For illustration, the available sensor data of an asset can be processed into a scalar representing the asset's health index h or deterioration index d = 1 - h, and this scalar can then be compared with thresholds TH1, ... TH2. n-1 The comparison is used to determine how to initialize the simulation.

[0199] By performing stochastic simulations (such as MCMC), not only can the evolution of asset health status be automatically determined and output, but the reliability associated with the determined evolution can also be automatically determined and output.

[0200] Information regarding reliability can take various forms. For illustration, the evolution of the confidence interval around curves 60 and 80 can be determined as a function of time within the prognostic time range. For any time j within the prognostic time range, the temporal evolution of the confidence interval can indicate the lower and upper boundaries of the critical failure probability 60 or the health index h. The upper and lower boundaries can be determined such that at least a certain percentage (e.g., at least 70%, 80%, 90%, or 95%) of random simulations produce a critical failure probability 60 or the health index h within the range between the upper and lower boundaries. Exemplary upper boundaries 81 and lower boundaries 82 indicating the temporal evolution of the confidence interval are shown in... Figure 7 As shown in the image.

[0201] Alternatively or additionally, the upper boundary 81 and lower boundary 82 may reflect changes in operational and / or environmental conditions that the asset may experience. For illustration, curves 80, 81, and 82 may each be used as referenced. Figure 3 The explanation, however, is that Markov chain models with different group transition probabilities are obtained by performing multiple stochastic simulations.

[0202] Figure 8A and Figure 8B The temporal evolution of the health index 80, determined by random simulation, is shown, along with the upper and lower boundaries 81 and 82 indicating the temporal evolution of the confidence interval. Curves 80-82 have been obtained via MCMC using transition probabilities determined from historical sensor data with associated fault characteristics, thus allowing transitions between states S1-S4 to be identified in the historical sensor data.

[0203] Curves 84 and 86 represent observed actual degradation of assets of the same asset type, but are not included in the historical data used to determine the transition probabilities of the Markov chain model. Degradation is represented as the evolution of a continuous function, which is the degradation exponent. Different states of the Markov chain model can be associated with various ranges of the degradation exponent function.

[0204] The prognostic asset health analysis results 80-82 are randomized. Although the actual condition of assets 84 and 86 may evolve differently from what is predicted by curve 80 and / or confidence intervals 81 and 82, the prognostic asset health analysis results 80-82 reliably indicate the evolution of asset deterioration as determined on a large sample of random assets.

[0205] Figure 9 This is a flowchart of method 90 according to an embodiment. Method 90 can be executed automatically by local controllers 21-23, central system 20, and / or remote server system 24. In one implementation, steps 91-92 of the method can be executed by the central system or remote server system 24, while steps 93-94 can be executed by local controllers 21-23. In other variations, steps 91-94 can be distributed across multiple ICs of the processing system in other ways.

[0206] In step 91, sensor data is received. The sensor data may be historical sensor data of assets of the same asset type as one or more assets to which the prognostic asset health analysis is to be performed (e.g., photovoltaic panels with a specific rated power range, wind turbine generators with a specific rated power range, transformers with ratings in a specific range).

[0207] Sensor data can be correlated with time periods that last longer than the prognostic time range.

[0208] Sensor data can be labeled sensor data, which includes information about states, such as states S1-Sn. For illustration, for any set of sensor data, there can be information that associates the sensor data with one of the states S1, ..., Sn in the Markov chain model.

[0209] If the sensor data does not include fault characteristics, step 91 may include receiving information from a human expert via a user interface, who assigns the sensor data to the states S1, ..., Sn of the Markov chain model.

[0210] If the sensor data does not include fault characteristics, step 91 may include calculating a scalar function based on the sensor data (the scalar function representing a deterioration or health index of the corresponding asset in the group), and comparing the scalar function with one or more thresholds (e.g., Figure 7 The thresholds TH1-TH3 in the data are compared to determine the time when the transition between S1, ..., Sn has occurred based on sensor data.

[0211] Scalar functions can be calculated using heuristic methods based on sensor measurements.

[0212] Scalar functions can take sensor measurements captured at different times as input and can process these sensor measurements into scalar functions representing the observed evolution of asset health, such as those reflected by a health index h or a deterioration index d.

[0213] Various techniques can be used to calculate scalar functions for identifying transitions between discrete states. For example, sensor measurements can be compared to a range of operating values. Each sensor measurement that goes out of range can be penalized. Other processing methods can be used, such as weighted summation or combining the weighting factors used for the sensor measurements with products of values ​​based on the deviations of the sensor measurements from the normal operating range. The weighting factors are based on the respective sensors and indicate the importance of the measurements to asset health.

[0214] Tools that provide mappings of sensor measurements to continuous health or degradation functions for various assets are known, including but not limited to circuit breakers, batteries (e.g., lithium-ion batteries), or transformers. For illustration, tools such as Elliptic APM or RelCare process sensor measurements to provide functions with values ​​over a continuous range that indicate asset health. Normalization can be used to normalize the health or degradation function to a desired range (e.g., from 0 to 1).

[0215] The techniques disclosed in this paper allow any health or degradation function to be mapped to discrete states of a state model using optional normalization and threshold comparisons.

[0216] In step 92, the transition probabilities of the Markov chain model can be determined. These transition probabilities can be automatically determined based on sensor data and associated state labels.

[0217] In an exemplary implementation, the transition probability can be determined based on conditional probability. For illustration, the transition probability from state i to state i+1 (where 1 ≤ i ≤ n-1) at time j can be determined as:

[0218] p i→i+1 (j) = #(x) j+1 =S i+1 ∧x j =S i ) / #(x j =S i (2)

[0219] In equation (2), the numerator represents the number of assets that are in state i at time j and transition to state (i+1) at time j+1. The denominator represents the number of assets that are in state i at time j.

[0220] Transition probabilities can be set and / or adjusted based on user input. Averaging or other processing can be performed to obtain the probabilities of homogeneous Markov chain models.

[0221] When sensor data is available for different groups of assets with the same asset type (e.g., photovoltaic panels with a specific rated power range, wind turbine generators with a specific rated power range, transformers with ratings in a specific range) but subjected to different operating conditions and / or environmental conditions, the transfer probability can be determined independently for each group.

[0222] In step 93, the transition probability can be used to perform RUL calculations or otherwise determine the time evolution of degradation of one or more assets. This may involve MCMC simulations, and then processing the results of the MCMC simulations to determine the RUL curve or other metrics for time-related degradation of one or more assets.

[0223] Step 93 may include: receiving sensor data captured for the asset during asset operation; and adapting RUL calculations based on the sensor data when the sensor data becomes available. This may include updating the MCMC simulation based on the sensor data when the sensor data becomes available.

[0224] In step 94, control and / or output operations can be automatically performed based on the results of RUL calculation or other prognostic asset health analysis.

[0225] For illustration, the RUL curve can be output. Information on the time evolution of the confidence interval or variance can also be output simultaneously.

[0226] Alternatively or additionally, the operating point of an asset can be automatically adjusted by the local controllers 21-23 associated with the asset.

[0227] Alternatively or additionally, inspections, maintenance, and / or replacement work can be automatically scheduled.

[0228] Alternatively or additionally, downtime for inspection, maintenance, and / or replacement work can be automatically scheduled.

[0229] Alternatively or additionally, alerts, warnings, or other outputs may be generated via the HMI based on the evolution of the RUL curve or other prognostic asset health status.

[0230] Figure 10 This is a schematic diagram of computing system 100. Computing system 100 may include one or more ICs 103. The one or more ICs may include application-specific integrated circuits (ASICs), processors, controllers, field-programmable gate arrays (FGPAs), or combinations of multiple such integrated circuits.

[0231] One or more ICs 103 may reside in one of the central system 20, local controllers 21-23, or server system 24, or may be distributed across these entities.

[0232] One or more ICs 103 may be operable to execute a stochastic simulation engine 104 to simulate the future evolution of an asset using a Markov chain model. The stochastic simulation engine 104 may be adapted to perform MCMC simulations.

[0233] The transition probabilities for the Markov chain model used by the stochastic simulation engine 104 can be received via interface 101 (e.g., when one or more ICs 103 reside in one of the local controllers 21-23 and the central system 20 calculates the transition probabilities). The transition probabilities can be calculated by one or more ICs 103 based on historical sensor data of an asset group having the same asset type as the assets targeted by the prognostic asset health analysis to be performed. The historical sensor data can be received via interface 101 or can be stored locally in data storage device 102.

[0234] One or more ICs 103 may be operable to implement a prediction engine 105. The prediction engine 105 may calculate a RUL curve or other prognostic information associated with the evolution of asset health status based on simulation results performed by a stochastic simulation engine 104.

[0235] One or more ICs 103 may be operable to implement output engine 106. Output engine 106 may generate output data or output signals for controlling the HMI and / or implementing control operations on an asset or a system in which the asset is being used. For illustration, output engine 106 may be operable to generate data and output the data to the HMI, such that an output RUL curve is generated. Output engine 106 may be operable to generate data and output the data to the HMI, such that information about the time evolution of confidence intervals or variances can be output simultaneously.

[0236] Alternatively or additionally, the output engine 106 may be operable to automatically adjust the operation point of the asset in response to the output of the prediction engine 105.

[0237] Alternatively or additionally, the output engine 106 may be operable to automatically generate and output information about inspection, maintenance and / or replacement work.

[0238] Alternatively or additionally, the output engine 106 may be operable to automatically generate and output information about downtime for inspections, maintenance, and / or replacement work that can be automatically scheduled.

[0239] Alternatively or additionally, based on the evolution of the RUL curve or other prognostic asset health status, the output engine 106 may be operable to automatically generate alarms, warnings, or other outputs for use via HMI output.

[0240] Various effects and advantages are associated with this invention. By using a Markov chain model with a state space that can consist of a fairly small number of states (e.g., three or four states), this method requires only a small number of transition probabilities between states. RUL curves or other prognostic asset health information can be efficiently determined. Not only can quantitative information be provided about the expected RUL curve, but also about its confidence interval. Accuracy and / or variance information can be quantified and output.

[0241] The methods and systems according to the invention can be used in connection with, but are not limited to, power system assets (e.g., assets of power generation, distribution and / or transmission systems or assets of industrial systems).

[0242] Although the invention has been described in detail in the accompanying drawings and the foregoing description, such description should be considered illustrative or exemplary rather than restrictive. By studying the drawings, the disclosure, and the appended claims, those skilled in the art and those practicing the claimed invention can understand and implement variations of the disclosed embodiments. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite articles "a" or "an" do not exclude a plurality. The mere fact that certain elements or steps are recited in different claims does not imply that an advantage cannot be obtained by utilizing a combination of such elements or steps; in particular, any further meaningful combination of claims, in addition to actual claim dependencies, should be considered disclosed.

Claims

1. A method for performing prognostic health analysis on assets (11-13; 16-18), particularly for determining the remaining useful life (RUL) or probability of failure (PoF) of power system assets or industrial assets, the method comprising the steps of: Multiple independent stochastic simulations are performed using the transition probabilities of a discrete Markov chain model. The discrete Markov chain model has a state space that includes a set of asset health states (41-44). Each of the plurality of independent stochastic simulations simulates the future evolution of the state space of the discrete Markov chain model within the prognostic range. The different sets of transition probabilities between the asset health states (41-44) are used to simulate the multiple independent stochastic simulations running in parallel, and the different sets of transition probabilities are associated with different operating conditions and / or environmental conditions. The evolution of prognostic asset health status within the prognostic range was calculated based on the multiple independent stochastic simulations (60; 80). Outputs are generated based on the calculated evolution of prognostic asset health status; and Actions related to the assets (11-13; 16-18) are automatically executed based on the calculated evolution of the prognostic asset health status.

2. The method according to claim 1, wherein, Calculating the evolution of the prognostic asset health status includes calculating RUL(60; 80).

3. The method according to claim 1 or 2, further comprising: The confidence information or variance information (81, 82) of the evolution of the prognostic asset health status as a function of time within the prognostic range is calculated based on the plurality of independent random simulations, wherein the output is also generated based on the confidence information or the variance information (81, 82).

4. The method according to claim 3, wherein, The confidence information includes the future evolution of the confidence interval within the prognosis range, or the variance information includes the future evolution of the variance within the prognosis range.

5. The method according to claim 3, wherein, The confidence information or the variance information includes the time evolution of the lower boundary and the time evolution of the upper boundary, wherein the lower boundary is associated with a first set of transition probabilities and the upper boundary is associated with a second set of transition probabilities that is different from the first set of transition probabilities.

6. The method according to claim 1 or 2, wherein, The state space includes: The operation of the asset is not adversely affected by the failure in at least one state (41). The asset's operation is adversely affected by the failure, but the asset continues to operate in at least one state (42, 43). The state in which the asset is inoperable due to a malfunction (44).

7. The method according to claim 1 or 2, wherein, Calculating the evolution of the prognostic asset health status includes: calculating the probability distribution (61-64) in the state space for multiple time periods within the prognostic range and mapping the probability distribution to a scalar, wherein, optionally, the evolution of the prognostic asset health status is obtained as the temporal evolution of the scalar.

8. The method according to claim 1 or 2, further comprising: The transfer probability is determined based on historical data, including sensor data for multiple assets; Optionally, determining the transition probability includes: Calculate a time-dependent scalar function based on sensor data used for the multiple assets. Based on the time-dependent scalar function, the transitions within the state space of the Markov chain model are identified, and The transition probability is calculated based on the transitions within the state space of the Markov chain model. Alternatively, the determining step may include comparing the scalar function with one or more thresholds.

9. The method according to claim 1 or 2, wherein, The multiple independent stochastic simulations are Markov chain Monte Carlo (MCMC) simulations.

10. The method according to claim 1 or 2, wherein, The Markov chain is homogeneous, and each state (41-44) in the state space has a non-zero transition probability to at most one other state in the state space.

11. The method according to claim 1 or 2, further comprising: Receive sensor measurement data captured during the operation of the asset; as well as The evolution of the prognostic asset health status is updated based on the received sensor measurement data.

12. The method according to claim 1 or 2, wherein, The assets are power transformers, distributed energy DER units, or electric generators; and / or The prognosis range is 1 year or longer, 2 years or longer, 3 years or longer, 4 years or longer, 5 years or longer, 10 years or longer, 15 years or longer, or 20 years or longer.

13. An operation and / or maintenance asset (11-13; 16-18), particularly methods for power system assets or industrial assets, said methods including: Perform a prognostic asset health analysis on the asset using the method according to any one of the preceding claims; and Automatically perform at least one of the following: generate an alarm or warning based on the calculated prognostic asset health status evolution; generate control signals for controlling the operation of the asset based on the calculated prognostic asset health status evolution; schedule downtime for the asset based on the calculated prognostic asset health status evolution; schedule maintenance work based on the calculated prognostic asset health status evolution; schedule replacement work based on the calculated prognostic asset health status evolution; change the maintenance interval based on the calculated prognostic asset health status evolution.

14. A computing system (20-24; 100) operable to perform prognostic health analysis on assets (11-13; 16-18), particularly for determining the remaining useful life (RUL) or probability of failure (PoF) of power system assets or industrial assets, the computing system comprising at least one integrated circuit operable to: Multiple independent stochastic simulations are performed using the transition probabilities of a discrete Markov chain model. in, The discrete Markov chain model has a state space that includes a set of asset health states (41-44). Each of the plurality of independent stochastic simulations simulates the future evolution of the state space of the discrete Markov chain model within the prognostic range. The at least one integrated circuit is operable to perform multiple independent stochastic simulations using the transition probabilities of the discrete Markov chain model, wherein different sets of transition probabilities between the asset health states (41-44) are used to simulate the multiple independent stochastic simulations running in parallel, and the different sets of transition probabilities are associated with different operating conditions and / or environmental conditions. The evolution of prognostic asset health status within the prognostic range is calculated based on the multiple independent stochastic simulations; and The automatic control output is generated based on the calculated evolution of the prognostic asset health status.

15. An industrial or power system (10; 15) comprising: Assets (11-13; 16-18); as well as The computing system (20-24) according to claim 14; 100), the computing system is used to perform prognostic asset health analysis on the assets (11-13; 16-18), optionally wherein the computing system is a distributed controller of the industrial or power system for controlling the assets.

Images