Fleet management system

Abstract

The invention relates to a fleet management system. A method implemented by a computing device of monitoring a set of machines. The method includes receiving a plurality of candidate statistical distributions for a set of machines. Each of the plurality of candidate statistical distributions describes first operational data and second operational data characterizing one or more aspects of at least one machine in the set of machines. The method further includes updating and comparing the plurality of candidate statistical distributions based on a combination of the first operational data and the second operational data. The method further includes selecting a baseline statistical distribution from the plurality of candidate statistical distributions based on the comparison. Further, the method includes outputting a baseline statistical distribution, where the baseline statistical distribution is predicted to optimally describe both the first operational data and the second operational data, and sending an alert to indicate the found baseline statistical distribution.

Description

Technical Field

[0001] The present disclosure relates generally to evaluating machine operations, and more particularly to evaluating data from the operations of a collection of machines to better manage the collection. Background Art

[0002] Vehicle fleets are monitored for a wide range of statistics that are observed or calculated as key performance indicators (KPIs). KPIs are data metrics used to track various aspects of a fleet of equipment, such as, but not limited to, the safety and performance of the vehicles. Examples of KPIs include fleet vehicle utilization, vehicle life, vehicle reliability, maintenance completion time, repair effectiveness, customer satisfaction, and customer retention. A large amount of data is used to analyze the fleet and predict indicators. The data can be obtained in a variety of different ways, including but not limited to real-time monitoring of individual vehicles, historical data points from previously recorded flights, and simulated tests. Analysis includes identifying or expressing indicators in a function or statistical distribution as discussed herein. A statistical distribution is a mathematical expression that describes the probability that a random variable will assume a specific value or set of values.

[0003] Large amounts of data need to be analyzed to evaluate and predict a large number of indicators. This analysis should provide meaningful results without requiring high computing power.

[0004] Unless explicitly identified as such, no claim is made herein that it qualifies as prior art merely by virtue of its inclusion in the Technical Field and / or Background sections. Summary of the Invention

[0005] One aspect of the present disclosure relates to a method for monitoring a set of machines implemented by a computing device. The method includes receiving a plurality of candidate statistical distributions for the set of machines. Each of the plurality of candidate statistical distributions describes first operating data and second operating data that characterize one or more aspects of at least one machine in the set of machines. The method also includes updating the plurality of candidate statistical distributions based on a combination of the first operating data and the second operating data. The method also includes comparing the plurality of candidate statistical distributions based on a combination of the first operating data and the second operating data. In addition, the method includes selecting a baseline statistical distribution from the plurality of candidate statistical distributions based on the comparison. The method includes outputting a baseline statistical distribution, wherein the baseline statistical distribution is predicted to best describe both the first operating data and the second operating data. The method also includes sending an alert to indicate the discovered baseline statistical distribution.

[0006] In some aspects of the method for monitoring a collection of machines, triggering an alarm results in a change in machine maintenance. For example, in response to the alarm, one or more steps in the maintenance procedure are changed. Examples include, but are not limited to, changing the frequency of maintenance and replacing one or more components at a new predetermined time (e.g., six months of use, 100,000 miles on the vehicle).

[0007] The features, functions, and advantages that have been discussed can be achieved independently in various aspects or may be combined in yet other aspects further details of which can be seen with reference to the following description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 is a schematic diagram of a computing device that receives operational data from a fleet of vehicles and one or more remote nodes, according to some aspects of the present disclosure.

[0009] Figure 2 is a schematic diagram of a fleet management system according to some aspects of the present disclosure.

[0010] Figure 3 is a flow chart illustrating a method of monitoring operational data of a fleet of vehicles according to some aspects of the present disclosure.

[0011] Figure 4 is a flow chart illustrating a method of monitoring operational data using distribution hypothesis testing according to some aspects of the present disclosure.

[0012] Figure 5 is a flowchart illustrating a method of performing candidate distribution refinement according to some aspects of the present disclosure.

[0013] Figure 6 is a flowchart illustrating a method of determining a candidate distribution selection according to some aspects of the present disclosure.

[0014] Figure 7 is a schematic diagram of a computing device for executing the methods disclosed herein, according to some aspects of the present disclosure. DETAILED DESCRIPTION

[0015] This application relates to analyzing operational data from a collection of machines. A variety of different machines are applicable, including but not limited to various vehicles (e.g., such as aircraft, ships, trucks, and cars), manufacturing equipment, computing equipment, and office equipment. An example of a collection is a fleet of vehicles.

[0016] In some examples, processing is applied to a fleet management system to analyze operational data associated with vehicles in a fleet to identify trends and behaviors in the data. The fleet management system determines an updated static statistical baseline distribution that best describes the operational data. The fleet management system uses the updated baseline distribution to monitor and track a large amount of operational data. The updated baseline distribution improves monitoring and tracking of the operational data by providing an accurate baseline for comparing incoming operational data.

[0017] The flight management system performs a comparative analysis between the updated baseline distribution and the candidate distribution when executing or rerunning an operation. If the current baseline distribution fails to closely match the operational data, the updated baseline distribution replaces the current faulty baseline distribution. Alerts can be generated to highlight new data trends or behaviors detected in the operational data. In this way, new trends or behaviors in the operational data are detected when the current baseline distribution no longer represents the operational data with statistical certainty.

[0018] Fleet management systems need to be aware of unexpected changes in operational data. They can also provide context and improved representations. These representations take the form of predefined parameter distributions with optimized parameters. Iteratively improving the distributions allows the fleet management system to further adapt to evolving operations. The fleet management system can also notify recipients of alerts to rerun analyses based on the statistical representations.

[0019] In some examples, a fleet management system treats each indicator in the operational data as a random variable. A baseline representation / prediction of the indicator's behavior is established based on engineering analysis or historical precedent. The observed indicator is compared to the baseline representation. When the baseline representation is determined to be invalid based on the fleet data, an alert is generated. In some examples, the management system also provides the statistical distribution that best describes the fleet data. The new distribution of the data is used to represent and predict the indicator.

[0020] Figure 1 A fleet 110 of vehicles 101 is shown. In this example, the vehicles 101 are aircraft configured to transport passengers and / or cargo or perform missions. The number and type of vehicles 101 comprising the fleet 110 can vary. Operational data 115 indicating the operation of the fleet 110 is collected by a computing device 120. Examples of operational data 115 include, but are not limited to, the number of trips (e.g., flights), the distance traveled, the number of passengers, the amount of cargo, fuel usage, maintenance records, and weather conditions. Operational data 115 can be collected from a variety of different sources, including, but not limited to, sensors on the vehicles 101, crew input, and remote nodes 109, such as third-party data (e.g., the Federal Aviation Administration), airport authorities, airline personnel, and weather services (e.g., the National Weather Service).

[0021] The computing device 120 includes Figure 2 Fleet monitoring system 205 is shown. Fleet monitoring system 205 is a computer program having instructions for performing hypothesis testing, candidate distribution refinement, and candidate distribution selection by hypothesis testing unit 210, candidate distribution refinement unit 220, and candidate distribution selection unit 230, respectively.

[0022] Figure 2 Also shown is the relationship between the hypothesis testing unit 210, the candidate distribution refinement unit 220, and the candidate distribution selection unit 230 included in the fleet monitoring system 205. In some examples, each unit is interconnected for analyzing the operational data 115 based on Bayesian statistics. The hypothesis testing unit 210, the candidate distribution refinement unit 220, and the candidate distribution selection unit 230 can work together to determine whether a distribution from an array of competing evolving statistical distributions (i.e., a candidate distribution) can outperform a single fixed distribution (i.e., the baseline distribution M0). In some examples, the hypothesis testing unit 210, the candidate distribution refinement unit 220, and the candidate distribution selection unit 230 rely on the hypothesis that the best-characterized distribution has the highest likelihood of generating a given operational data set. In some examples, each unit is interconnected for analyzing the operational data 115 based on frequentist statistics.

[0023] Hypothesis testing is a process in which fleet monitoring system 205 determines whether a baseline distribution M0 or a candidate distribution has the highest likelihood of matching operational data 115. Baseline distribution M0 is set based on a believed behavior of the operation of one or more vehicles 101. In some examples, baseline distribution M0 is set by engineers who design the aircraft or based on historical precedent. Candidate distributions include an array of one or more distributions that differ from the current baseline distribution M0.

[0024] Hypothesis testing unit 210 tests the goodness of fit of a baseline distribution or other frequentist assumptions. In some cases, the test determines whether the baseline distribution reasonably fits operational data 115. In some examples, when the baseline distribution does not adequately characterize operational data 115, an alert 240 is generated by hypothesis testing unit 210.

[0025] Candidate distribution refinement is a process in which the candidate distribution refinement unit 220 optimizes possible distribution options to better represent the operational data 115. In some examples, candidate distribution options are pre-selected from a predetermined set of distributions and are subsequently refined.

[0026] The candidate distribution selection selects one of the candidate distributions as the new baseline distribution M0 because it better represents the operational data 115. In some examples, candidate distribution selection is initiated after the hypothesis testing unit 210 has determined that the baseline distribution is no longer a reasonable representation of the operational data 115. In other examples, candidate distribution selection can operate simultaneously with hypothesis testing. The candidate distribution selection unit 230 sends the new baseline distribution M0 245 to the hypothesis testing unit 210 for use in future hypothesis testing.

[0027] Figure 2 Further shown is the operational data 115 received by the hypothesis testing unit 210 and the candidate distribution refinement unit 220. The candidate distribution refinement unit 220 outputs the selection of candidate distributions 215 to the candidate distribution selection unit 230. Figure 2 In some of the examples shown, candidate distribution options are also output to a hypothesis testing unit for testing.

[0028] Figure 3 1 is a flow chart of a method 300 for monitoring a fleet 110 of vehicles 101. Operational data 115 is received from one or more of the vehicles 101 and / or remote nodes 109 (block 310). In some examples, the operational data 115 is historical data stored at a computing device or operational data captured and measured in real time from the computing device.

[0029] Examples of such operational data 115 may include, but are not limited to:

[0030] - Operational availability - measured as the fraction of the total time that an aircraft is not ready for operation based on schedule. It reflects the reliability and maintainability levels implemented in the design, the fidelity of the manufacturing process, maintenance policies, in-theater assets, order / shipping times, etc.

[0031] - Budget - cost of fuel, parts, personnel, etc.

[0032] - Vehicle Configuration - A measure of total missions / flights as a percentage of hours in each vehicle configuration.

[0033] - Network Connectivity - A measure of the percentage of time that the network is functioning normally versus the percentage of time that the network is "down" and not functioning.

[0034] - O&S Costs - Various cost metrics including cost per flight hour, annual aircraft cost, and annual fleet cost.

[0035] - Critical to non-critical maintenance - measured as the ratio of critical to non-critical maintenance actions.

[0036] - Equipment availability - used for capacity planning of equipment to include spares, parts and flight equipment.

[0037] - Flight time before flight abort - The flight time before flight abort is measured in hours.

[0038] In some examples, method 300 further includes performing hypothesis testing by comparing the baseline distribution M0 to the operational data 115 and further comparing the candidate distribution to the operational data (block 320). In some examples, the comparison is based on using a calculated Bayes factor and / or a difference in the Kullback-Liebler divergence of the baseline distribution relative to the true but unknown distribution, and the Kullback-Liebler divergence of the candidate distribution relative to the true but unknown distribution. The comparison that produces a better fit to represent the operational data 115 is preferred.

[0039] The method 300 also includes outputting an alert in response to determining that the current baseline distribution is not reasonably valid (block 330). In other words, the candidate distribution is a very good representation of the operational data compared to the baseline distribution, or the baseline distribution is an implausible representation of the data.

[0040] The method 300 also includes performing candidate distribution refinement on the received operational data (block 340). The candidate distribution refinement optimizes one or more candidate distributions that may better represent the operational data 115. The possible candidate distributions are then sent to the hypothesis testing unit 210 and the candidate distribution selection unit for hypothesis testing. In this way, the possible candidate distributions selected by the candidate distribution selection unit 230 and the hypothesis testing unit 210 for testing are continuously refined over time and reflect recent operations.

[0041] In some examples, candidate distribution refinement (block 340 ) is performed on the received operational data before hypothesis testing (block 320 ) is performed.

[0042] The method determines a new baseline distribution from one or more candidate distributions (block 350).Once the new baseline distribution is determined, further hypothesis testing is performed on the new baseline distribution.

[0043] Hypothesis Testing

[0044] The hypothesis testing unit 210 determines whether the original hypothesis (i.e., the baseline distribution M0) is no longer a reasonable representation of the operational data 115. In some examples, the baseline distribution is the null hypothesis, which is rejected using the nomenclature of a null hypothesis significance test (NHST). In other examples, the test is not constructed as an NHST, but is implemented as a Bayesian hypothesis test (BHT), in which the baseline distribution is tested against an array of candidate distributions M1, M2, ..., M n+1Each candidate distribution is tested against the baseline distribution to determine whether it has a higher likelihood of generating the operational data 115. In some examples, this is done by setting a static threshold on the ratio of these likelihoods, called a Bayes factor (see Equation 1). The Bayes factor is equivalent to the ratio of evidence for selecting one distribution relative to another using Bayesian logic. j and M0 are different models (i.e., distributions). As an example, M j is the candidate model and M0 is the baseline model. This equation compares the generative performance (called evidence) of the candidate model with respect to the operational data D with the generative performance of the baseline model with respect to the same operational data D. P is the likelihood function. The resulting ratio or Bayes factor is a value that will indicate which distribution is more likely to match the operational data 115. The larger the Bayes factor, the better the candidate distribution M. j The more likely it is, the better the model. The opposite is also true, where a lower Bayes Factor indicates that the model baseline distribution M0 is more likely to be a better representation. This occurs when the Bayes Factor is <1.

[0045] (Equation 1)

[0046]

[0047] In some examples, a decision strategy is set based on the difference in the sample Kullback-Liebler (KL) divergence relative to the true but known distribution of the baseline distribution and the candidate distribution. The KL divergence is a measure of the expected logarithmic difference in probability density / mass. Intuitively, the distribution with the smallest logarithmic deviation relative to the true but unknown distribution is optimal. Conversely, a strategy derived from the likelihood ratio has an equivalent decision strategy based on the cumulative logarithmic deviation provided by the dataset for each of the two distributions.

[0048] The KL divergence-based strategy has many positives, especially in experiments with high misleading evidence rates. Below is an example of the strategy with derivation and analysis.

[0049] Let x be a sample of the operational data 115 for indicator X, and ζj:=ln(P(X|M j ))-ln(P(X|M0)), where E(ζ j ):=μ, and E(ζ j 2 ):=σ 2 As an extension of Sanov's theorem, the KL-divergence of M0 and M j The difference in KL-divergence is equal to E(ζ j ). Therefore, if μ is determined to be only positive, the candidate distribution is significantly better than the baseline hypothesis. The experiment then ends.

[0050] Using the principle of maximum entropy, we can construct Bayesian inference, where, without any additional information, ζ j The variation of is described as a normal distribution N(μ,σ). Because the value of μ or σ is unknown, a non-informative prior is set, namely the reference prior. Below, Bayesian inference of (μ,σ) is established in Equation 2, where n is the number of samples of X.

[0051] (Equation 2)

[0052]

[0053] in

[0054]

[0055] If the posterior providing an inference of possible values that μ could take given the operational data 115 indicates that μ is likely to be negative or zero, with very low probability α, then the experiment is complete, and M j is undoubtedly a “better” distribution of operational data 115. Mathematically, this strategy is described by Equation 3, where F t is the cumulative distribution function (CDF) of the Student's t distribution with n-1 degrees of freedom (dof).

[0056] (Equation 3)

[0057]

[0058] This leads to a balance due to the properties of the Student's t distribution: the high-density region (HDR) becomes narrower with smaller sample standard deviations and larger sample sizes. This is achieved by using the Bayes factor required for comparison, which is equal to nζ j This is demonstrated by cumulative evidence of the equivalent reduction in

[0059] Figure 4 4 is a flow chart of a method 400 implemented by a computing device for monitoring a collection of machines. The method 400 includes receiving operational data observed from the collection of machines, wherein the operational data is expected to be represented by a baseline distribution characterizing one or more aspects of operation of at least one machine in the collection of machines (410).

[0060] The method also includes comparing the operational data to each of a plurality of distributions including a baseline distribution and a candidate distribution within a set of candidate distributions, wherein the candidate distribution is predicted to characterize one or more aspects of operation of at least one machine in the set of machines (420).

[0061] The method also includes discovering a new trend in the operational data if it is determined that one of the candidate distributions more accurately represents the operational data than the baseline distribution or that the baseline distribution is unlikely (430). In some examples, the computing device determines that the baseline distribution is an unlikely representation of the process regardless of the candidate distribution. This allows for the use of NHST with composite hypotheses.

[0062] In response to the determination, one candidate distribution is assigned as the baseline distribution (440).

[0063] The method also includes triggering an alert to indicate a new trend discovered in the operational data (450).

[0064] In some examples, candidate distribution refinement unit 220 may periodically update the plurality of candidate distributions based on operational data periodically received by candidate distribution refinement unit 220. In some examples, candidate distribution updating may be performed concurrently before or after hypothesis testing is performed.

[0065] In some examples, the comparison is based on a calculated Bayes factor, which represents a ratio indicating the probability of selecting one distribution relative to another using Bayesian logic. In some examples, the comparison is based on a calculated difference in the Kullback-Liebler divergence of samples from two different distributions, where the KL divergence represents a measure of the expected logarithmic deviation of a distribution when random samples are drawn from a true but unknown distribution.

[0066] Candidate distribution refinement

[0067] Candidate distribution refinement determines one or more candidate distributions that better represent the operational data 115. In some examples, this occurs after the hypothesis testing algorithm rejects the baseline distribution. In some examples, distribution refinement occurs before hypothesis testing to provide an up-to-date representation of the operation.

[0068] The candidate distribution provides a variety and a large amount of operational data 115. In some examples, the large amount of operational data 115 is small (e.g., a given cockpit warning light occurs once in tens of thousands of operating hours). Other data sets have much larger data sets (e.g., the number of flight hours for each classification for each flight record).

[0069] Statistical methods for parameter estimation using Bayesian methods may require less data than those based on maximum likelihood estimation (MLE), which is common in frequentist methods. Frequentist methods assume complete uncertainty at initialization because Bayesian methods are seeded with exogenous information (see Equation 4). Let θ be the unknown parameter we believe to be distributed according to a probability density function (PDF) P(θ), called the "prior." We can then use the prior, the distribution likelihood P(θ|D), and the marginal likelihood P(D) to construct the "posterior" distribution P(θ|D).

[0070] (Equation 4)

[0071]

[0072] In some examples, such as in aerospace, original design estimates provide context and value to set P(θ) for given operational data 115. For example, a given line replaceable unit (LRU) on an aircraft platform may be identical or highly similar in construction to an already installed component. Expectations are set for characteristics, such as flight hours between removals (FHBR) compared to similar components. In some examples, conservative estimates from engineering analysis and design can be used for new classes of components.

[0073] In some examples, the current implementation of distribution refinement is based on maximum a posteriori (MAP) estimation. In some examples, due to computational convenience, a specific class of P(θ) is currently used to form the estimate: the conjugate prior. The conjugate prior decomposes the evolution of the prior to the posterior into an algebraic relation. Thus, numerical methods that can be computationally heavy, such as those based on Monte Carlo sampling, are avoided. This is the main driver for the distribution options provided in Table 1. These candidate distributions represent a broad, predetermined set of continuous distributions where the domain is the positive real numbers with small exceptions.

[0074] Table 1: Candidate distribution

[0075]

[0076] Figure 5 A method 500 for performing candidate distribution refinement is shown. The method includes receiving operational data 115 (block 510). Distribution candidates describing the operational data 115 are then determined (block 520).

[0077] In some examples, distribution refinement is based on MAP estimates formed using conjugate priors. In some examples, distribution refinement is based on MLE estimates. In some examples, distribution refinement is based on full a posteriori estimates of the parameters.

[0078] Candidate distribution selection

[0079] Candidate Distribution Selection evaluates candidate distributions and selects them as the new baseline distribution. The analysis uses the posterior distribution P(θ|D) built using the entire operational data 115 (D) to calculate the likelihood P(D|M). To reduce the computational burden, the analysis uses a recursive approximation as shown in Equation 5, which is P(D|M j ,θ) is a mixture of the integrals over the prior and posterior distributions. This reduces the computational burden by requiring only incremental modifications of the total model evidence with the most recent sample evidence. The use of the posterior distribution of the parameters reflects the evolution of the parameters in continuous Bayesian inference. k is an integer equal to the number of times the candidate distribution selection unit 230 has been run. is the set of operational data 115 that has been determined after the last run of the candidate distribution selection unit 230. Bayes' theorem is then used to establish the recurrence relationship shown in Equation 5, where λ∈[0,1] is the geometric decay. The geometric decay allows for a soft range to be set on the operational data 115, which is helpful when the process is slowly changing, and therefore not stationary, and counteracts the long-term effects of poor prior specifications.

[0080] (Equation 5)

[0081]

[0082] in,

[0083]

[0084] Figure 6 An example of a method 600 implemented by a computing device monitoring a collection of machines is shown. The method 600 receives a plurality of candidate statistical distributions for the collection of machines, each of the plurality of candidate statistical distributions describing first operational data and second operational data characterizing one or more aspects of at least one machine in the collection of machines (block 610).

[0085] In some examples, the second operational data is used to set a prior distribution for each of a plurality of candidate statistical distributions.

[0086] In some examples, the second operating data is received before receiving the first operating data.

[0087] In some examples, a candidate statistical distribution in the plurality of candidate statistical distributions is selected from a group of distributions consisting of: chi-square, chi-squared, Erlang, exponential, gamma, generalized gamma, half-normal, inverse gamma, inverse Gaussian, lognormal, Nakagami, normal, Rayleigh, and reciprocal inverse Gaussian.

[0088] In some examples, the refined distribution candidates are determined based on a MAP estimate formed using a conjugate prior. In some examples, the refined distribution candidates are determined based on an MLE estimate. In some examples, the distribution candidates are determined based on a posterior distribution of the parameter.

[0089] The method 600 includes updating a plurality of candidate statistical distributions based on a combination of the first operational data and the second operational data (block 620 ).

[0090] In some examples, the updating and comparing are performed recursively at a kth time, where k is an integer equal to the number of times the new first operation data has been received.

[0091] The method also includes comparing a plurality of candidate statistical distributions based on a combination of the first operational data and the second operational data (block 630).

[0092] In some examples, the comparison is performed in real time as the data is received.

[0093] In some examples, as shown in Equation 5, the selection from the plurality of candidate distributions is based on the selection with the maximum P(D k+1 |M j ) is recursively executed based on the candidate j associated with it, where P(.|M j ) is the model M j Evidence of D k is the second operation data, and is the first operation data, θ k and θ k+1 It is model M j is a parameter of , and λ is a constant geometric forgetting factor.

[0094] The method 600 further includes selecting a baseline statistical distribution from the plurality of candidate statistical distributions based on the comparison (640). The method further includes outputting the baseline statistical distribution, wherein the baseline statistical distribution is predicted to best describe both the first operational data and the second operational data (650).

[0095] Upon selecting and outputting the discovered baseline statistical distribution, method 600 further includes sending an alert to indicate that the discovered baseline statistical distribution has been selected (660). In some examples, the alert may be sent to an internal monitoring system for monitoring trends in operational data 115. In some examples, the alert may be sent to an external monitoring system for monitoring trends in operational data 115. In some examples, the alert may be sent wirelessly to a remote node via a wireless communication network.

[0096] According to some aspects, methods 300, 400, 500, and 600 are performed by a computing device for monitoring a fleet of vehicles. The computing device includes processing circuitry and memory containing instructions executable by the processing circuitry.

[0097] According to some aspects, a non-transitory computer-readable medium stores a computer program product for controlling a computing device, the computer program product comprising software instructions that, when executed on the computing device, cause the computing device to perform the steps of methods 300 , 400 , 500 , and 600 .

[0098] Various processes receive operational data about a collection of machines. The operational data is used to verify a baseline distribution and candidate distributions. When a baseline distribution is determined to be invalid, a new baseline distribution is required to define the operational data. The new baseline distribution provides information on future actions for managing and operating the collection of machines. Multiple different distributions are analyzed to determine which best fits the operational data. The selected distribution is then assigned as the baseline distribution. An alert indicating the discovered baseline statistical distribution may be sent to an internal or external monitoring system. In some examples, the alert may be sent wirelessly to a remote node via a wireless communication network.

[0099] In some examples, a triggered alarm results in a change to one or more operational aspects of a machine. The alarm indicates that a change is necessary because the alarm indicates a problem with one or more current procedures. In one specific example of a method monitoring a collection of machines, one or more changes are made to the way the machines are operated in response to the alarm. One specific example includes replacing parts in the machine more frequently.

[0100] In some examples, methods 300, 400, 500, 600 are used to monitor a fleet of vehicles, and specifically to schedule maintenance for the vehicles. An initial maintenance schedule is developed and used for the fleet. When new trends are identified using the methods, the maintenance schedule is changed to take into account the trends. The ability to modify the maintenance schedule based on the new trends provides better performance for individual vehicles and the fleet as a whole. In one example, the initial maintenance schedule includes performing an oil change every 6,000 miles. After maintaining the fleet for a period of time, a trend is discovered that engine wear is occurring at a higher rate than expected. In view of the new trend, the maintenance schedule is adjusted to change the oil in the vehicle every 5,000 miles. In a similar example, the maintenance schedule is adjusted to maintain oil changes every 6,000 miles, but the type of oil used in the vehicle is changed to take into account the increased engine wear. Identifying trends (i.e., increased engine wear) and adjusting the maintenance schedule based on the trends provides a longer life expectancy for the vehicles and the fleet as a whole.

[0101] For example, a process is used to determine the service of a group of machines. Operational data regarding the machines is expected to perform according to a baseline distribution. Expected maintenance and various other performance aspects of the machines are set based on the baseline distribution. Received operational data is compared to the baseline distribution and various other distributions. If another distribution is determined to better represent the operational data, that distribution is assigned as the new baseline. Performance aspects of the machines are set based on the new distribution. One advantage of this process is that a more accurate baseline is determined relatively early in the process. Instead of continuing to base decisions on the baseline distribution, a more accurate distribution is used that better represents the operational data and provides a more accurate prediction of future issues with the machines. When there is a change, an alert indicating the change can be sent.

[0102] For example, operational data may be safety or performance related. In addition to the examples outlined herein, operational data may be inefficient operation of a machine. Operational data may characterize the utilization of each machine, the reliability of each machine, the operating hours of each machine, the fuel or electricity usage of each machine, or the maintenance records of each machine. In some examples, operational data may indicate a degradation in the performance of a machine. Additionally or alternatively, in an example, operational data may indicate an increase in vibration sensed in the machine, or an increase in fuel or electricity usage. Operational data may optionally be provided by sensors of the machine. In some examples, the sensors are health and usage monitoring sensors, strain sensors, vibration sensors, shock sensors, wear sensors, or corrosion sensors.

[0103] In some examples, maintenance may include any of inspection of the machine, repair of the machine, and / or scrapping of the machine. In an example, the collection of machines may be a fleet of vehicles, and the fleet may include one or more fleet vehicles. In some examples provided herein, the operational data may represent the distance traveled by each vehicle and / or the time of operations completed by each vehicle. In some examples, the fleet of vehicles may be a fleet of aircraft or other ground-based vehicles (e.g., vehicles based on wheels or tracks).

[0104] In some of the examples described above, scheduling maintenance on one or more machines (eg, a vehicle or a fleet) and / or maintaining a set of machines is performed according to a schedule.

[0105] In some examples, operational data is safety- or performance-related. Monitoring this data can be used to maintain or operate the machine in a manner that ensures safe operation and / or safe usage. Performance-related aspects provide benefits including, but not limited to, more efficient use, longer lifespan, and appropriate maintenance scheduling. Examples include, but are not limited to, machine utilization, machine lifespan, machine reliability, maintenance completion time, repair effectiveness, customer satisfaction, and customer retention.

[0106] In some examples, operational data indicates anticipated failure of a machine component or inefficient operation of the machine. Monitoring these aspects allows for maintenance of the machine before operational or efficiency issues arise. Additionally or alternatively, the affected machine can be removed from service before failure occurs. This results in more efficient use of the vehicle and safer and more effective operation.

[0107] In some examples, the operational data may include information about the utilization of each machine, the reliability of each machine, the operating hours of each machine, the fuel or electricity usage of each machine, or the maintenance history of each machine. This operational data may be used to optimize the efficiency of vehicles and fleets and ensure proper operation of the vehicles. Using this information may allow for the removal of machines from service before they experience performance issues.

[0108] In some examples, operational data indicates performance degradation of a machine. Monitoring this operational data provides for more efficient use of the machine and the entire predetermined group of machines. Furthermore, this operational data can extend the life of the machine, as maintenance can be performed before the machine is damaged due to degradation. Additionally or alternatively, the affected machine can be removed from service before degradation exceeds a threshold. In some examples, this provides for more efficient use and safer operation of the machine.

[0109] In some examples, operational data indicates increased vibration or increased fuel or electricity usage sensed in the machine. This operational data may be an indicator of a future problem with the machine. Monitoring this data can provide the ability to remove the machine from service and provide maintenance before it reaches a point in use that could result in damage to the machine and / or inefficient use of the machine. Vibration and increased fuel or electricity usage have been identified as indicators of impending machine problems. In some examples, the machine is removed to provide a more comfortable ride for passengers.

[0110] In some examples, operational data is provided by sensors of the machine. The sensors are configured and positioned to provide the accurate data needed for accurate analysis. When the underlying data is not an accurate reflection of the actual aspects of the machine, the monitoring aspects are ineffective. In some examples, the sensors are health and usage monitoring sensors. These sensors indicate the usage of the vehicle to monitor operational aspects and ensure operation within expected ranges. In some examples, the sensors are strain sensors, vibration sensors, shock sensors, wear sensors, or corrosion sensors. These types of sensors provide accurate data on a range of aspects that can be monitored and analyzed. This type of data has been determined to be a valid indicator of future problems when the data exceeds certain thresholds.

[0111] In some examples, maintenance includes any of inspection, repair, or retirement of a machine. Maintenance aspects of a machine provide insightful data that provides accurate and useful output. Problems that have affected a machine provide valuable data that can determine future operational aspects of other machines.

[0112] In some examples, machines scheduled for maintenance are prioritized based on how far the indicator exceeds a threshold. In some examples, the indicator has been found to provide an accurate estimate of when maintenance is due. The further beyond the indicator the machine continues to operate, the more likely the machine is to have an operational problem.

[0113] In some examples, the collection of machines is a fleet of vehicles. The vehicles can be monitored for various operational data indicative of future operational aspects of the vehicles.

[0114] In some examples, the operational data represents the distance traveled by each vehicle and / or the number of hours of operation completed by each vehicle. These aspects provide a good indication of the expected future operation of the vehicle. For example, the more miles and / or flights an aircraft has traveled, the more likely it is that maintenance will be required to ensure the aircraft continues to operate efficiently. In some examples, thresholds are set for the distance and / or hours of operation, and once these thresholds are reached, the vehicle is removed from service for maintenance.

[0115] In some examples, for purposes of illustration, the vehicle fleet is a fleet of aircraft, but may also include the various vehicles described below. Operational data for the aircraft may be obtained, which has been found to be effective in predicting how the aircraft and fleet of aircraft will operate in the future. In some examples, aircraft maintenance may be scheduled and / or performed based on the monitored operational data.

[0116] Figure 77 is a schematic block diagram illustrating a computing device 120 according to one or more aspects of the present application. The example computing device includes processing circuitry 710, memory circuitry 730, and interface circuitry 735. The processing circuitry 710 is communicatively coupled to the memory circuitry 730 and the interface circuitry 735, for example, via one or more buses 720. The processing circuitry 710 may include one or more microprocessors, microcontrollers, hardware circuits, discrete logic circuits, hardware registers, digital signal processors (DSPs), field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), or combinations thereof. For example, the processing circuitry 710 may be programmable hardware capable of executing a software program 740, for example, stored as a machine-readable computer program in the memory circuitry 730. The memory circuitry 730 of various aspects may include any non-transitory machine-readable medium known in the art or that may be developed, whether volatile or non-volatile, including but not limited to solid-state media (e.g., SRAM, DRAM, DDRAM, ROM, PROM, EPROM, flash memory, solid-state drives, etc.), removable storage devices (e.g., Secure Digital (SD) cards, miniSD cards, microSD cards, memory sticks, thumb drives, USB flash drives, ROM cartridges, universal media disks), fixed drives (e.g., magnetic hard drives), etc., all or in any combination.

[0117] The interface circuit 735 may be a controller hub configured to control input and output (I / O) data paths of the computing device 120. Such I / O data paths may include data paths for exchanging signals over a communication network and data paths for exchanging signals with electronic devices, nodes, or users. For example, the interface circuit 735 may include output circuitry (e.g., a transmitter circuit 745 configured to send communication signals over a communication network) and input circuitry (e.g., a receiver circuit 755 configured to receive communication signals over a communication network).

[0118] Interface circuitry 735 may be implemented as a single physical component or as multiple physical components arranged serially or separately, any of which may be communicatively coupled to any other component or may communicate with any other component via processing circuitry 710 .

[0119] Other aspects include a non-transitory computer-readable medium (e.g., memory circuit 730) storing a computer program (e.g., software instructions 740) comprising software instructions that, when executed on processing circuit 710 of computing device 120, cause computing device 120 to perform any of the methods disclosed herein.

[0120] In some examples, aspects disclosed herein are applicable to a management system 100 that monitors a fleet 110. Fleet 110 can include various vehicles 101, including, but not limited to, manned aerial vehicles, unmanned aerial vehicles, manned space vehicles, unmanned space vehicles, manned rotary-wing aircraft, unmanned rotary-wing aircraft, satellites, rockets, missiles, manned land vehicles, unmanned land vehicles, manned water vehicles, unmanned water vehicles, manned underwater vehicles, unmanned underwater vehicles, and combinations thereof.

[0121] Additionally, this disclosure includes examples according to the following clauses:

[0122] Item 1. A method implemented by a computing device for monitoring a collection of machines, the method comprising: receiving a plurality of candidate statistical distributions for the collection of machines, each of the plurality of candidate statistical distributions describing first operating data and second operating data characterizing one or more aspects of at least one machine in the collection of machines; updating the plurality of candidate statistical distributions based on a combination of the first operating data and the second operating data; comparing the plurality of candidate statistical distributions based on a combination of the first operating data and the second operating data; selecting a baseline statistical distribution from the plurality of candidate statistical distributions based on the comparison; outputting the baseline statistical distribution, wherein the baseline statistical distribution is predicted to best describe both the first operating data and the second operating data; and sending an alert to indicate the discovered baseline statistical distribution.

[0123] Clause 2. The method of clause 1, wherein the updating and comparing are recursively performed a k-th time, where k is an integer equal to the number of times new first operation data has been received.

[0124] Clause 3. The method of clause 2, wherein the second operational data is used to set a prior distribution for each candidate statistical distribution in the plurality of candidate statistical distributions.

[0125] Clause 4. The method of clause 1, wherein the second operational data is received before receiving the first operational data.

[0126] Clause 5. The method of clause 1, wherein one of the plurality of candidate statistical distributions is selected from a group consisting of: chi-square, chi-squared, Erlang, exponential, gamma, generalized gamma, half-normal, inverse gamma, inverse Gaussian, lognormal, Nakagami, normal, Rayleigh, and reciprocal inverse Gaussian.

[0127] Clause 6. The method of clause 1, wherein the comparing is performed in real time as the data is received.

[0128] Clause 7. The method of clause 1, wherein the selection from the plurality of candidate distributions is based on selecting the distribution with the largest P(D k+1 |M j ) is recursively executed based on the candidate j associated with it, where P(.|M j ) is the model M j Evidence of D k is the second operation data, and is the first operation data, θ k and θ k+1 It is model M j The parameter is , and λ is a constant geometric forgetting factor:

[0129]

[0130] in,

[0131]

[0132] Clause 8. The method of Clause 1, further comprising updating the plurality of candidate statistical distributions based on estimates formed using a conjugate prior.

[0133] Clause 9. The method of clause 1, wherein the first operational data and the second operational data are security or performance related.

[0134] Clause 10. The method of clause 1, wherein the first operating data and the second operating data represent utilization of the machine, reliability of the machine, operating time of the machine, fuel or electricity usage of the machine, or maintenance records of the machine.

[0135] Clause 11. The method of clause 1, wherein the first operational data and the second operational data indicate a degradation in performance of the machine.

[0136] Clause 12. The method of any of clauses 1 to 11, wherein the operational data is collected in real time during commercial aircraft flight operations, military aircraft maintenance procedures, military aircraft flight operations, or during commercial aircraft maintenance procedures.

[0137] Clause 13. The method of any one of clauses 1 to 12, wherein the operational data indicates increased vibration or increased fuel or electricity usage sensed in the machine.

[0138] Clause 14. The method of any one of clauses 1 to 13, wherein the operational data is provided by sensors of the machine.

[0139] Clause 15. The method of clause 14, wherein the sensor is a health and usage monitoring sensor.

[0140] Clause 16. The method of clause 14 or 15, wherein the sensor is a strain sensor, a vibration sensor, a shock sensor, a wear sensor, or a corrosion sensor.

[0141] Clause 17. The method of any one of clauses 1 to 16, wherein the operational data includes maintenance, the maintenance including any of an inspection of the machine, a repair of the machine, or a scrapping of the machine.

[0142] Clause 18. The method of any one of clauses 1 to 17, wherein the collection of machines is a fleet of vehicles.

[0143] Clause 19. The method of clause 18, wherein the operational data represents a distance traveled by each of the vehicles and / or an operation time completed by each of the vehicles.

[0144] Clause 20. The method of clause 18 or 19, wherein the fleet of vehicles is a fleet of aircraft.

[0145] Clause 21. A method of maintaining a collection of machines, the method comprising: scheduling the method of maintenance according to any one of clauses 1 to 20; and maintaining the collection of machines according to the schedule.

[0146] Clause 22. A computing device comprising a hardware processor and a memory containing computer program instructions that, when executed by the hardware processor, cause the computing device to perform the method of any one of clauses 1 to 21.

[0147] Clause 23. A computer program comprising computer program instructions that, when executed by one or more hardware processors of a computing device, cause the computing device to perform the method of any one of clauses 1 to 22.

[0148] Clause 24. A non-transitory computer-readable medium having stored thereon the computer program of Clause 23.

[0149] Item 25. A computing device for monitoring a collection of machines, the computing device comprising: a processing circuit and a memory, the memory containing instructions executable by the processing circuit, whereby the computing device is configured to: receive a plurality of candidate statistical distributions for the collection of machines, each of the plurality of candidate statistical distributions describing first operating data and second operating data characterizing one or more aspects of at least one machine in the collection of machines; update the plurality of candidate statistical distributions based on a combination of the first operating data and the second operating data; compare the plurality of candidate statistical distributions based on a combination of the first operating data and the second operating data; select a baseline statistical distribution from the plurality of candidate statistical distributions based on the comparison, wherein the baseline statistical distribution is predicted to best describe both the first operating data and the second operating data; and output an alert indicating the baseline statistical distribution.

[0150] Clause 26. The computing device of Clause 25, wherein updating and comparing are recursively performed a k-th time, where k is an integer equal to the number of times new first operation data has been received.

[0151] Clause 27. The computing device of Clause 26, wherein the second operational data is used to set a prior distribution for each candidate statistical distribution in the plurality of candidate statistical distributions.

[0152] Clause 28. The computing device of Clause 25, wherein the second operating data is received before receiving the first operating data.

[0153] Clause 29. A computing device according to clause 25, wherein one of the plurality of candidate statistical distributions is selected from a group of distributions consisting of: chi-square, chi-squared, Erlang, exponential, gamma, generalized gamma, half-normal, inverse gamma, inverse Gaussian, lognormal, Nakagami, normal, Rayleigh, and reciprocal inverse Gaussian.

[0154] Clause 30. The computing device of Clause 25, wherein the comparison is performed in real time as the data is received.

[0155] Clause 31. The computing device of clause 25, wherein the computing device is further configured to: select from the plurality of candidate distributions based on the selection of the maximum P(D k+1 |M j ) is recursively executed based on the candidate j associated with it, where P(.|M j ) is the model M j Evidence of D k is the second operation data, and is the first operation data, θ k and θk+1 It is model M j The parameter is , and λ is a constant geometric forgetting factor:

[0156]

[0157] in,

[0158]

[0159] Clause 32. The computing device of Clause 25, further configured to update the plurality of candidate statistical distributions based on the estimates formed using the conjugate prior.

[0160] Item 33. A non-transitory computer-readable medium storing a computer program product for controlling a computing device, the computer program product comprising software instructions that, when executed on the computing device, cause the computing device to: receive a plurality of candidate statistical distributions for a set of machines, each of the plurality of candidate statistical distributions describing first operating data and second operating data characterizing one or more aspects of at least one machine in the set of machines; update the plurality of candidate statistical distributions based on a combination of the first operating data and the second operating data; compare the plurality of candidate statistical distributions based on a combination of the first operating data and the second operating data; select a baseline statistical distribution from the plurality of candidate statistical distributions based on the comparison; trigger an alarm to indicate the baseline statistical distribution; and wherein the baseline statistical distribution is predicted to best describe both the first operating data and the second operating data.

[0161] Clause 34. A method of scheduling maintenance for a collection of machines, the method comprising: the method of monitoring a collection of machines according to any one of clauses 1 to 20; and modifying a maintenance schedule for the collection of machines to account for new trends in operational data.

[0162] Clause 35. A method of maintaining a collection of machines, the method comprising: the method of scheduling maintenance according to Clause 34; and maintaining the collection of machines according to the schedule.

[0163] Item 36. A computer-implemented method for scheduling maintenance for a collection of machines, the method comprising: receiving a plurality of candidate statistical distributions for the collection of machines, each of the plurality of candidate statistical distributions describing first operating data and second operating data characterizing one or more aspects of at least one machine in the collection of machines; updating the plurality of candidate statistical distributions based on a combination of the first operating data and the second operating data; comparing the plurality of candidate statistical distributions based on a combination of the first operating data and the second operating data; selecting a baseline statistical distribution from the plurality of candidate statistical distributions based on the comparison; outputting the baseline statistical distribution, wherein the baseline statistical distribution is predicted to best describe both the first operating data and the second operating data; and modifying the maintenance schedule for the collection of machines to take into account new trends in the operating data.

[0164] Clause 37. The method of clause 36, wherein the first operational data and the second operational data are security or performance related.

[0165] Clause 38. The method of clause 37, wherein the first operating data and the second operating data represent utilization of the machine, reliability of the machine, operating time of the machine, fuel or electricity usage of the machine, or maintenance record of the machine.

[0166] Clause 39. The method of clause 37 or 38, wherein the first operating data and the second operating data indicate a degradation in performance of the machine.

[0167] Clause 40. The method of any one of clauses 37 to 39, wherein the operational data indicates increased vibrations sensed in the machine, or increased fuel or electricity usage.

[0168] Clause 41. The method of any one of clauses 37 to 40, wherein the operational data is provided by sensors of the machine.

[0169] Clause 42. The method of clause 41, wherein the sensor is a health and usage monitoring sensor.

[0170] Clause 43. The method of clause 41 or 42, wherein the sensor is a strain sensor, a vibration sensor, a shock sensor, a wear sensor, or a corrosion sensor.

[0171] Clause 44. The method of any one of clauses 36 to 43, wherein the operational data includes maintenance, the maintenance including any of inspection of the machine, repair of the machine, or scrapping of the machine.

[0172] Clause 45. The method of any one of clauses 36 to 44, wherein the collection of machines is a fleet of vehicles.

[0173] Clause 46. The method of clause 45, wherein the operational data represents a distance traveled by each of the vehicles and / or an operation time completed by each of the vehicles.

[0174] Clause 47. The method of clause 45 or 46, wherein the fleet of vehicles is a fleet of aircraft.

[0175] Clause 48. The method of any of clauses 36 to 47, wherein the operational data is collected in real time during commercial aircraft flight operations, military aircraft maintenance procedures, military aircraft flight operations, or during commercial aircraft maintenance procedures.

[0176] Clause 49. The method of any one of clauses 36 to 48, wherein the updating and comparing are performed recursively up to a k-th time, wherein k is an integer equal to the number of times new first operation data has been received.

[0177] Clause 50. The method of Clause 49, wherein the second operational data is used to set a prior distribution for each of the plurality of candidate statistical distributions.

[0178] Clause 51. The method of any one of clauses 36 to 50, wherein the second operational data is received before receiving the first operational data.

[0179] Clause 52. A method according to any one of clauses 36 to 51, wherein one of the plurality of candidate statistical distributions is selected from a group of distributions consisting of: chi-square, chi-squared, Erlang, exponential, gamma, generalized gamma, half-normal, inverse gamma, inverse Gaussian, lognormal, Nakagami, normal, Rayleigh, and reciprocal inverse Gaussian.

[0180] Clause 53. The method of any one of clauses 36 to 52, wherein the comparison is performed in real time as the data is received.

[0181] Clause 54. The method of any one of clauses 36 to 53, wherein the selection from the plurality of candidate distributions is based on selecting the distribution with the largest P(D k+1 |M j ) is recursively executed based on the candidate j associated with it, where P(.|M j ) is the model M j Evidence of D k is the second operation data, and is the first operation data, θ kand θ k+1 It is model M j The parameter is , and λ is a constant geometric forgetting factor:

[0182]

[0183] in,

[0184]

[0185] Clause 55. The method of any one of clauses 36 to 54, further comprising updating the plurality of candidate statistical distributions based on estimates formed using a conjugate prior.

[0186] Clause 56. A method of maintaining a collection of machines, the method comprising: the method of scheduling maintenance according to any one of clauses 36 to 55; and maintaining the collection of machines according to the schedule.

[0187] Clause 57. A computing device comprising a hardware processor and a memory containing computer program instructions that, when executed by the hardware processor, cause the computing device to perform the method of any one of clauses 36 to 56.

[0188] Clause 58. A computer program comprising computer program instructions that, when executed by one or more hardware processors of a computing device, cause the computing device to perform the method of any of clauses 36 to 56.

[0189] Clause 59. A non-transitory computer-readable medium having stored thereon the computer program of Clause 58.

[0190] This application also covers the following examples:

[0191] Example 1. A method implemented by a computing device (120) for monitoring a collection of machines, the method comprising:

[0192] receiving a plurality of candidate statistical distributions (215) for the set of machines, each of the plurality of candidate statistical distributions (215) describing first operational data and second operational data characterizing one or more aspects of at least one machine in the set of machines;

[0193] recursively updating the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data;

[0194] recursively comparing the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data;

[0195] selecting a baseline statistical distribution (245) from the plurality of candidate statistical distributions (215) based on the comparison;

[0196] outputting the baseline statistical distribution (245), wherein the baseline statistical distribution (245) is predicted to best describe both the first operational data and the second operational data; and

[0197] An alert is sent (240) to indicate the discovered baseline statistical distribution (245).

[0198] Example 2. The method of example 1, wherein k is an integer equal to the number of times the plurality of candidate statistical distributions (215) have been previously updated and evaluated.

[0199] Example 3. The method of example 2, wherein the second operational data is used to set a prior distribution for each of the plurality of candidate statistical distributions.

[0200] Example 4. The method of example 1, wherein the second operation data is received before receiving the first operation data.

[0201] Example 5. The method of example 1, wherein one of the plurality of candidate statistical distributions (215) is selected from a group of distributions consisting of:

[0202] Chi-square, chi-squared, Erlang, exponential, gamma, generalized gamma, half-normal, inverse gamma, inverse Gaussian, lognormal, Nakagami, normal, Rayleigh, and reciprocal inverse Gaussian.

[0203] Example 6. The method of example 1, wherein the comparing is performed in real time as the data is received.

[0204] Example 7. The method of Example 1, wherein selecting from the plurality of candidate distributions is based on selecting the distribution with the largest P(D k+1 |M j ) is recursively executed based on the candidate j associated with it, where P(.|M j ) is the model M j Evidence of D k is the second operation data, and is the first operation data, θ k and θ k+1 It is model M j The parameter is , and λ is a constant geometric forgetting factor:

[0205]

[0206] in,

[0207]

[0208] Example 8. The method of Example 1, further comprising:

[0209] Updating the plurality of candidate statistical distributions (215) is based on the estimates formed using the conjugate prior.

[0210] Example 9. The method of Example 1, wherein the first operational data and the second operational data are security or performance related.

[0211] Example 10. The method of Example 1, wherein the first operating data and the second operating data represent utilization of the machine, reliability of the machine, operating time of the machine, fuel or electricity usage of the machine, or maintenance records of the machine.

[0212] Example 11. The method of example 1, wherein the first operating data and the second operating data indicate a degradation in performance of the machine.

[0213] Example 12. A computing device (120) for monitoring a collection of machines, the computing device (120) comprising:

[0214] processing circuitry (710) and memory (730), the memory (730) containing instructions executable by the processing circuitry (710), whereby the computing device (120) is configured to:

[0215] receiving a plurality of candidate statistical distributions (215) for the set of machines, each of the plurality of candidate statistical distributions (215) describing first operational data and second operational data characterizing one or more aspects of at least one machine in the set of machines;

[0216] recursively updating the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data;

[0217] recursively comparing the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data;

[0218] selecting a baseline statistical distribution (245) from the plurality of candidate statistical distributions (215) based on the comparison, wherein the baseline statistical distribution (245) is predicted to best describe both the first operational data and the second operational data; and

[0219] An alarm (240) is outputted indicating the baseline statistical distribution (245).

[0220] Example 13. The computing device (120) of Example 12, wherein k is an integer equal to the number of times the candidate statistical distribution (215) has been previously updated and evaluated.

[0221] Example 14. The computing device (120) of Example 13, wherein the second operational data is used to set a prior distribution for each candidate statistical distribution in the plurality of candidate statistical distributions (215).

[0222] Example 15. The computing device (120) of Example 12, wherein the second operational data is received before receiving the first operational data.

[0223] Example 16. The computing device (120) of Example 12, wherein one of the plurality of candidate statistical distributions (215) is selected from a group of distributions consisting of:

[0224] Chi-square, chi-squared, Erlang, exponential, gamma, generalized gamma, half-normal, inverse gamma, inverse Gaussian, lognormal, Nakagami, normal, Rayleigh, and reciprocal inverse Gaussian.

[0225] Example 17. The computing device (120) of Example 12, wherein the comparison is performed in real time as the data is received.

[0226] Example 18. The computing device (120) of Example 12, further configured to:

[0227] The selection from the plurality of candidate distributions is based on the selection with the maximum P(D k+1 |M j ) is recursively executed based on the candidate j associated with it, where P(.|M j ) is the model M j Evidence of D k is the second operation data, and is the first operation data, θ k and θ k+1 It is model M j The parameter is , and λ is a constant geometric forgetting factor:

[0228]

[0229] in,

[0230]

[0231] Example 19. The computing device (120) of Example 12, further configured to:

[0232] The plurality of candidate statistical distributions are updated based on the estimates formed using the conjugate prior (215).

[0233] Example 20. A non-transitory computer-readable medium storing a computer program product for controlling a computing device (120), the computer program product comprising software instructions that, when executed on the computing device (120), cause the computing device (120) to:

[0234] receiving a plurality of candidate statistical distributions (215) for a collection of machines, each of the plurality of candidate statistical distributions (215) describing first operational data and second operational data characterizing one or more aspects of at least one machine in the collection of machines;

[0235] recursively updating the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data;

[0236] recursively comparing the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data;

[0237] selecting a baseline statistical distribution (245) from the plurality of candidate statistical distributions (215) based on the comparison; and

[0238] triggering an alarm (240) to indicate said baseline statistical distribution (245),

[0239] wherein the baseline statistical distribution (245) is predicted to best describe both the first operational data and the second operational data.

[0240] A variety of different machines are applicable, including but not limited to various vehicles (e.g., such as aircraft, ships, trucks, cars), manufacturing equipment, computing equipment, and office equipment. In some of the examples disclosed above, various aspects are disclosed in the context of monitoring a fleet of aircraft. This is an example, but the disclosure is not limited to this application. Examples of vehicles include vehicles that can be used in various environments (e.g., ground-based, water-based, air-based, and space-based). Vehicles include but are not limited to manned aircraft, unmanned aircraft, manned spacecraft, unmanned spacecraft, manned rotorcraft, unmanned rotorcraft, satellites, rockets, missiles, manned land vehicles, unmanned land vehicles, manned water vehicles, unmanned water vehicles, manned underwater vehicles, unmanned underwater vehicles, and combinations thereof.

[0241] These aspects can also be used in other contexts besides fleet management. For example, distribution can be used in wireless telecommunications, finance, healthcare, and a variety of other applications.

[0242] Of course, without departing from the essential characteristics of the present disclosure, the various aspects of the present disclosure may be implemented in other ways than those specifically set forth herein. The aspects of the present invention are to be considered in all respects as illustrative and not restrictive, and all changes within the meaning and equivalency range of the appended claims are intended to be encompassed therein.

Claims

1. A method for monitoring a collection of machines implemented by a computing device (120), the method comprising: receiving a plurality of candidate statistical distributions (215) for the set of machines, each of the plurality of candidate statistical distributions (215) describing first operational data and second operational data characterizing one or more aspects of at least one machine in the set of machines; recursively updating the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data; recursively comparing the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data; selecting a baseline statistical distribution (245) from the plurality of candidate statistical distributions (215) based on the comparison; outputting the baseline statistical distribution (245), wherein the baseline statistical distribution (245) is predicted to best describe both the first operational data and the second operational data; and An alert is sent (240) to indicate the discovered baseline statistical distribution (245).

2. The method according to claim 1, wherein k is an integer equal to the number of times the plurality of candidate statistical distributions (215) have been previously updated and evaluated.

3. The method according to claim 2, wherein: The second operational data is used to set a prior distribution for each of the plurality of candidate statistical distributions.

4. The method according to claim 1, wherein The second operation data is received before the first operation data is received.

5. The method according to claim 1, wherein One of the plurality of candidate statistical distributions (215) is selected from a group of distributions consisting of: Chi-square, chi-squared, Erlang, exponential, gamma, generalized gamma, half-normal, inverse gamma, inverse Gaussian, lognormal, Nakagami, normal, Rayleigh, and reciprocal inverse Gaussian.

6. The method according to claim 1, wherein The comparison is performed in real time as the data is received.

7. The method according to claim 1, wherein The selection from the plurality of candidate distributions is based on the selection with the maximum P(D k+1 |M j ) is recursively executed based on the candidate j associated with it, where P(.|M j ) is the model M j Evidence of D k is the second operation data, and is the first operation data, θ k and θ k+1 It is model M j The parameter is , and λ is a constant geometric forgetting factor: in, 8. The method according to claim 1, further comprising: Updating the plurality of candidate statistical distributions (215) is based on the estimates formed using the conjugate prior.

9. A computing device (120) for monitoring a collection of machines, the computing device (120) comprising: processing circuitry (710) and memory (730), the memory (730) containing instructions executable by the processing circuitry (710), whereby the computing device (120) is configured to: receiving a plurality of candidate statistical distributions (215) for the set of machines, each of the plurality of candidate statistical distributions (215) describing first operational data and second operational data characterizing one or more aspects of at least one machine in the set of machines; recursively updating the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data; recursively comparing the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data; selecting a baseline statistical distribution (245) from the plurality of candidate statistical distributions (215) based on the comparison, wherein the baseline statistical distribution (245) is predicted to best describe both the first operational data and the second operational data; and An alarm (240) is outputted indicating the baseline statistical distribution (245).

10. A non-transitory computer-readable medium storing a computer program product for controlling a computing device (120), the computer program product comprising software instructions that, when executed on the computing device (120), cause the computing device (120) to: receiving a plurality of candidate statistical distributions (215) for a collection of machines, each of the plurality of candidate statistical distributions (215) describing first operational data and second operational data characterizing one or more aspects of at least one machine in the collection of machines; recursively updating the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data; recursively comparing the plurality of candidate statistical distributions (215) k times based on a combination of the first operational data and the second operational data; selecting a baseline statistical distribution (245) from the plurality of candidate statistical distributions (215) based on the comparison; as well as triggering an alarm (240) to indicate said baseline statistical distribution (245), wherein the baseline statistical distribution (245) is predicted to best describe both the first operational data and the second operational data.

Images