System and method for health assessment of downhole tools

BACKGROUND OF THE INVENTION

[0001] Various tools are used in the exploration and production of hydrocarbons to measure properties of geologic formations during or after excavation of a borehole. The properties are measured by means of formation evaluation tool (FE tools) and other suitable devices, typically integrated into a bottomhole assembly. Sensors are used in FE tools to monitor various downhole conditions and formation characteristics.

[0002] Environments in which FE-tools, drilling equipment and other drill string components operate, is very harsh and include conditions such as high downhole temperatures (eg. In excess of 200 ° C) and events with high shock and vibration. Rig Operators are also currently in the process of using the tools to obtain routing profiles that have previously been impossible, thereby increasing the strain on utilities. At the same time require high reliability customers to help them to avoid costly wellbore failure.

[0003] Until now, periodic maintenance has been the most widespread method for maintaining tool reliability. As time goes on, there has been a shift towards condition-based maintenance which currently employs construction regulations and coarse thresholds for nominal operation to assess individual tools nature. The present techniques are however inferior in that a large amount of telemetry data collected during operation, has not yet been effectively treated.

BRIEF DESCRIPTION OF THE INVENTION

[0004] A system for evaluating the condition of a mechanism comprising: at least one sensor associated with the mechanism for generating observation data; a memory in operative communication with the at least one sensor, wherein the bearing includes a database to store observation data generated by the sensor; and a processor in operative communication with the memory, for receiving the observation data, the processor comprising: a detector capable of receiving the observation data and which is able to identify the mechanism operates in a normal or degraded manner, where the degraded manner is an indication of a fault in the mechanism; diagnostic means responsive to said observation data for identifying a fault type from at least one symptom pattern; and the projection equipment in operative communication with the at least one sensor, detector and diagnostic equipment, where the projection equipment is able to calculate a remaining useful life (RUL, remaining Useful life) for the mechanism based on information from at least one of the sensor, detector and diagnostic equipment, where projection device includes a population projection equipment for calculating the RUL based on a duration of use of the mechanism, cause-projection equipment for calculating the RUL value based on causal data and power-projection equipment for calculating the RUL value based on the efficacy data generated because of the error.

[0005] A method for evaluating the condition of a mechanism, comprising: receiving observation data generated by at least one sensor associated with the mechanism; To identify the mechanism operating at normal or degraded manner, where the degraded way an induction of an error in the mechanism; and in response to an identification of the degraded way, to identify a kind of error from at least one symptom pattern, and calculating a remaining lifetime (RUL, Remaining Useful Life) for the mechanism based on a comparison of observational data and examples of degradation data in connection with the wrong type wherein calculating the RUL is based on: a duration of use of the mechanism, causal data and efficacy data generated from error.

[0006] A computer-program product stored on machine readable media for assessing the condition of a mechanism by performing maskinimplementerbare instructions. The instructions perform: receiving observation data generated by at least one sensor adjacent to the mechanism; To identify the mechanism operates in a normal or degraded manner; where a deteriorating way is an indication of a fault in the mechanism; and in response to an identification of a deteriorating way to identify a kind of error of at least one symptom pattern, and calculate a remaining useful life (RUL) of the mechanism based on a comparison of observation data with examples of deterioration data adjacent to the error type.

BRIEF DESCRIPTION OF DRAWINGS

[0007] The following description should not be construed as limiting in any way. Referring to the accompanying drawings in which like elements have been number alike, and wherein:

Fig. 1 outlines one embodiment of a well logging system;

FIG. 2 outlines an embodiment of a system for evaluating the condition of a downhole tool;

Fig. 3 is a block diagram of another embodiment of the system of FIG. 2;

FIG. 4 is a flow chart that gives an example of a method for training the model of the system of FIG. 3;

Fig. 5 is a block diagram of a part of the system of FIG. 2 for generating an estimation of observation;

FIG. 6 is a block diagram of a part of the system of FIG. 2 for generating an alarm as an indication of an error;

Fig. 7 is a block diagram of a part of the system of FIG. 2 for generating a symptom observation;

Fig. 8 is a block diagram of a part of the system of FIG. 2 for generating an error class estimate;

FIG. 9 is a block diagram of a part of the system of FIG. 2 for generating a deterioration path and an associated lifetime;

FIG. 10 is a block diagram of a part of the system of FIG. 2 for generating an estimate of a residual life of the downhole tool;

FIG. 11 illustrates examples of deterioration courts;

FIG. 12 illustrates an observed deterioration path and examples of degradation paths in FIG. 11;

FIG. 13 is a flowchart to give an example of a method for classifying a deterioration in path and estimating the remaining life in connection with deterioration path; and

Fig. 14 outlines an alternative embodiment of a system for evaluating the condition of a downhole tool.

DETAILED DESCRIPTION OF THE INVENTION

[0008] There is provided a system and method for assessing the condition of a downhole tool or other mechanism. The method is a data-driven solution for estimating the state of the borehole tool units. The method includes analyzing data obtained from a formation evaluation tool (FE-tool) or another downhole device to determine: 1) whether there is a fault in the device or not, 2) if there is an error, error type, and 3) a residual life (RUL) of the tool. In a utførelsesfomn, the method includes comparing the collected telemetry data and associated statistics on data-driven models that have been trained for: 1) to differentiate between the nominal and degraded operation of error detection, 2) to differentiate between arrays of possible error classes of diagnosis, and 3) differentiate between odd and even deterioration courts for forecasting purposes (ie estimation of the remaining useful life).

[0009] A detailed description of one or more embodiments of the generated system and the obtained method, the here presented as examples and not limitative, with reference to the figures. Detailed description of the system and method are provided in the following publications: 1) Dustin R. Garvey and J. Wesley Hines, "Data Based Fault Detection, Diagnosis, and Prognosis of Oil Drill Steering Systems", The University of Tennessee, Knoxville, Department of Nuclear Engineering, and 2) J. Wesley Hines and Dustin R. Garvey, "Monitoring, Diagnostics, and prognostic for Drilling Operations: Final Report," August 2007, Nuclear Engineering Department; University of Tennessee, both hereby entirely incorporated by reference.

[0010] Referring to FIG. 1, an exemplary embodiment of a well logging system 10 includes a drillstring 11 that is shown disposed in a borehole 12 that penetrates at least one earth formation 14 to take measurements of properties of the formation 14 and / or borehole 12. The drilling fluid , or drilling mud 16 may be pumped through the borehole 12. As described herein, "formation" of the various features and materials as may be encountered in a subterranean environment. It is therefore to be considered that although the term "formation" generally refers to geologic formations of interest, that the term "formations," as used herein, in some cases includes all geological points or volumes of interest (such as an exploration area). In addition, it is noted that "drill string" as used herein, refers to any structure is suitable for lowering a tool through a borehole or connecting a drill to the surface and is not limited to the structures and configurations described herein.

[0011] In one embodiment, a downhole device (BHA) 18 arranged in well logging system 10 at or near the lower part of the drill string 11. The BHA 18 includes a plurality of formation evaluation (FE) downhole tool 20 for measuring one or more physical quantities in or around a borehole as a function of depth and / or time. Taking these measurements are referred to as "logging", and a record of such measurements are referred to as a "sign". Many types of measurements are taken to obtain information about the geological formations. Some examples of measurements include gamma ray logs, nuclear magnetic resonance logs, neutron log r, resistivity and sonic or acoustic logs.

[0012] Examples of login processes that can be performed using the system 10, includes processes for measurement-while-drilling (MWD) and logging-while-drilling (LWD) for measurements of properties of the formations and / or borehole are taken downhole during or after drilling. The data obtained during these processes, can be transferred to the surface and can also be stored in the downhole tool for later retrieval. Other examples include logging measurements for drilling, wireline logging and lowering shot logging.

[0013] The downhole tool 20 includes, in one embodiment one or more sensors or receivers 22 to measure various properties of the formation 14 as the tool 20 is lowered down the borehole 12. Such sensors 22 include, for example. nuclear magnetic resonance sensors (sensors NMR), resistivity, porøsitetssensorer, gamma ray sensors, seismic receivers and others.

[0014] Each of the sensors 22 may be a single sensor or multiple sensors located at a single position. In one embodiment, one or more of the sensors many sensors located near each other and assigned to a particular position on the drill string. In other embodiments, further comprises each sensor 22 further components such as watches, storage processors, etc.

[0015] In one embodiment, the tool 20 equipped with the transmitting equipment to eventually communicate with a processor 24 on the surface. Such transmission equipment may have any desired shape, and different transmission media and methods may be used. Exemplary compounds include wired, fiber optic, wireless connections or mud pulse telemetry.

[0016] In an embodiment, which includes processing unit 24 on the surface and / or the tool 20 components necessary to provide for storage and / or processing of data collected from the tool 20. Exemplary components include, without limitation, at least one processor , storage, cache, input devices, dispensers and the like. Processing unit 24 on the surface is preferably arranged to control the tool 20.

[0017] In one embodiment, includes tool 20 also a downhole clock 26 or other timing means for indicating a time at which each measurement was taken by the sensor 20. The sensor 20 and the downhole clock 26 may be included in a common housing 28. On the present description housing 28 can represent any structure used to support at least one sensor downhole clock 26 and other components.

[0018] Referring to FIG. 2, there is shown a system 30 for evaluating the condition of the downhole tool 20 or other device used in conjunction with the BHA 18 and / or drill string 11. The system can be incorporated into a computer or other processing unit that is able to receive data from the tool. Treatment unit may be incorporated in the tool 20 or be included as part of the treatment unit 24 on the surface.

[0019] In one embodiment, the system includes a computer 30 connected to tool 31, 20. Exemplary components include, without limitation, at least one processor, a memory, a cache memory, input devices, the output devices and the like. As these components are known to those skilled in the art, these are not described in detail here. The computer 31 may be arranged in at least one of the processing unit 24 on the surface and the tool 20.

[0020] Some of the descriptions contained herein, is reduced to an algorithm stored on machine-readable media. The algorithm is implemented using computer 31 and provide operators with desired outputs.

[0021] The tool 20 generates measured data which is stored in a memory associated with the tool and / or processing unit at the surface. The computer 31 receives data from the tool 20 and / or treating the surface unit for evaluating the condition of the tool 20. Although the computer 31 here is described as separate from the tool 20 and the processing unit 24 on the surface, the computer 31 may be a component of either the tool 20 or processing unit 24 on the surface, and consequently either the tool 20 or the processing unit 24 on the surface serve as a means for evaluating the condition of the tool.

[0022] Referring to FIG. 3, the system 30 includes memory 32 in which one or more databases 34, 36 and 38 are stored. The system 30 also includes a processor 40 which includes one or more analytical units includes empirical models 42, 44, 46 and 48. The models described here, the data-driven models, i.e. the data describing the input and utmatingskarakteristikkene define the model.

[0023] The data used by the system 30, is an abundance of data describing various aspects of how individual tools in a float function, is used and in some cases fails. In one embodiment, the data associated with a selected tool 20, categorized into three main types. Data types include storage dump data 34, operation data 36, ​​and maintenance data 38.

[0024] Bearing Dump Data 34 is a collection and / or display of the contents of an internal storage associated tool 20. Bearing Dump data 34 includes, for example. Sensor readings related to the sensed physical size and / or around the borehole, such as temperature, pressure and vibration. Operating data 36 includes measurements relating to the operation of the tool, such as electric power and motor or bit rotation. Maintenance data 38 includes data extracted from the tool after an error is observed.

[0025] The predictor 42 and the detector 44 is used to determine whether the tool 20 operates on a either nominal (ie normal) or degraded or not. Predictor 42 produces estimates of measured observations and generates estimate residues based on comparison with observation examples, and detector 44 evaluates whether the tool operates in a degraded way based on estimates remains. The diagnostic unit 46 is used to identify the type or class of any detected flaws from symptom patterns generated from observations. Symptom patterns include, but are not limited to, the predictor estimates residues, alarm patterns and signals that can be used to quantify the ambient or operating stresses. Forecasting unit 48 is used to derive the remaining useful life (RUL) to the tool 20 from observations of its deterioration pattern or history.

[0026] In one embodiment, the system is a non-parametric fuzzy interference system (NFI). The NFI is a fuzzy interference system (FIS) whose function centers and parameters are observations of examples of feeds and outputs.

[0027] In one embodiment, prior to use of system 30 for evaluation of tool state, the models 42, 44, 46, 48 trained based on error-free data to be able to detect the error, diagnose errors and determine the remaining lifetime. This training is in one embodiment performed through a training procedure 50.

[0028] FIG. 4 illustrates a method, i.e. a training procedure 50, to train the models in the system 10. The process 50 includes one or more stages 51, 52, 53 and 54.1 one embodiment, the method 50 embodiment of all of the steps 51, 52 , 53 and 54 in the described order. Some steps may be omitted, however, steps may be added or the order of the steps can be changed.

[0029] In the first step 51 is the predictor 42 trained by building up a legal base in the internal memory to the predictor 42. The predictor legal base is built up by selecting a number of examples of sightings referred to as "Sample Obs. # 1 - # NP" in FIG. 3, from signals collected from a clean tool operation. These signals are in one embodiment obtained from the storage dump data 34. The term "signal" or "observation" as used herein, refers to the measurement, observation or maintenance data received for the tool 20. Each signal consists of one embodiment of one or more data points over a selected time interval.

[0030] In one embodiment, each signal is processed using processes which include statistical analysis, data fitting, and data modeling to provide an observation curve. Examples of statistical analysis comprises calculating a sum, average, a variance, a standard deviation, t-distribution, a confidence interval, and others. Examples of data fitting include various regression methods, such as linear regression, least squares, segmented regression, hierarchal linear modeling, and others.

[0031] In the second step 52, the detector 44 trained by calculating a residual for each observation by calculating an error between the measured values ​​of the observation and predicted values. Each residue is fed to a statistical routine to construct a number of distribution functions for each residue such as probability distribution functions (PDF), which is representative of the nominal system operation. These examples of nominal distributions are represented as "nominal displaced. #P" In FIG. 3, where "P" refers to the number of residual signals.

[0032] In the third step 53, the results of prediktor- and detector workout combined with a selected signal, operation and maintenance data for creating diagnostic device scissors base that will be used to identify symptom patterns to the wrong classes.

[0033] In this step, data such as the residues, extracted from one or more of the databases 32, 34, 36 to provide symptom-patterns associated with a known fault type, that error class. These symptom patterns are then consolidated and included as examples in the diagnosis device 46. At this point, the diagnosis unit 46 effectively learned the relationship between the estimated and remains known error classes.

[0034] The fourth step 54, the analysis of results from the preceding stage, is combined with additional signaling, operation and maintenance data to generate forecasting unit's legal base that maps deterioration courts, such as absorbed vibration, the tool life. Deterioration Paths using data points from the predictor 42, the detector 44 and the diagnostic unit 46, such as observation data and alarm data over a time interval that includes time that the tool 20 failed. Further information from the storage dump data 34 may also be combined so that further signals or composite signals (eg. Running total over a threshold) to produce degradation pathways. Any suitable regression functions or data fitting techniques can be applied to the data received from the tool 20 to generate deterioration pitch.

[0035] Figures 5-10 illustrate methods for evaluating the condition of a downhole tool or other component in a formation evaluation / examination system, such as a tool used in drilling for performing a downhole measurement. Methods include various steps described here. The procedures can be performed continuously or intermittently as desired. The methods are described herein in connection with the downhole tool 20 although the processes may be performed in conjunction with any number and design of sensors and tools, as well as any means for lowering the tool and / or drilling a borehole. The methods can be performed using one or more processors or other devices that are able to receive and process measurement data such as computer 01/31 one embodiment, the method includes the execution of all the steps in the described sequence. Some steps may be omitted, however, steps may be added or the order of the steps can be changed.

[0036] Referring to FIG. 5, where the first step, the tool data dump 34 or other data collected from the equipment or another component of the well logging system 10, is collected from the tool internal stock to extract useful information. From these data, a plurality ask observations 58 (Obs. # 1 - #NQ), ie, measured observations, introduced in predictor 42.

[0037] In one embodiment, includes asking observations 58 any type of data on measured characteristics of the formation and / or borehole, as well as data relating to the operation of the tool. In one example, the data comprises pressure, electrical current, the engine RPM, the drill rotational speed, vibration and temperature measurements.

[0038] The predictor 42 calculates the estimated observations 60 ("Estimated Obs. # 1- # NQ), by determining which of the predictor observation examples which are most similar to each sampling observation 60.

[0039] In one embodiment, the predictor 42 NFI one predictor. This embodiment of the predictor 42 is a non-parametric, auto-associative model that performs signal correction by correlations inherent in the signals. This embodiment reduces the effects of noise or utstyrsanomaliteter and generates signal patterns similar to those of normal operating conditions. In another embodiment, the predictor 42 one autoassosiativ kjerneregresjons predictor (AAKRA predictor).

[0040] Because the predictor 42 has been previously trained on exclusive "good" data, ie data generated during the known nominal operation, teacher predictor 42 effectively the correlations that are present below nominal, flawless tools operation. When these correlations as amended, which often is the case when an error is present, the predictor 42 is still able to estimate what the signal values ​​should be if it had not been a change in the correlation. The system 30 thus provides a dynamic reference point that can be compared with measured observations so that as soon as a change in signalkorrelasjonene, there will be a corresponding divergence of estimates of observations. When an error is present in the well logging system 10, the estimates will generally be far from their observed values ​​for the affected signals.

[0041] In one embodiment utilizes predictor 42 different regression methods, including non-parametric regression as kjerneregresjon, to generate an observation estimate 60 that match a query observation 58. Kjerneregresjon (KR) comprises estimating value by calculating a weighted average of hisoriske observation examples. The herein described methods are not limited to any particular statistical analysis as any process, such as curve fitting, can be used.

[0042] For a number of observation examples, the example. KR-estimation performed by calculating a distance "d" to a query observation, i.e. a feed "x", from each of the observation examples "X" input of the distances in a core function that converts the distances to the weights, i.e. similarities, and estimation of output by calculating a weighted average of a utmatingseksempel.

[0043] The distance can be calculated using any known technique. An example of a distance is a Euclid distance represented by the following equation: d (Xi, x) = Xi - x, where "i" represents a number of feeds. Another example of distance is the adaptive Euclidean distance where distance calculation are excluded for the measured observations that lie outside the reach of the maximum and minimum innmatingseksemplene.

[0044] To transform the distance d to a weight or a like, is in one embodiment a core function "Kh (d)" used. An example of such a core function is part of the Gaussian kernel, which are represented by the following equation:

1 -dY

Kh (d) = Ti = e / 2h \ V 2xch2 where "h" refers to the core bandwidth and are used to regulate the effective distances that are considered equal. Other examples of core functions include the inverse spacing exponential, absolute exponential, uniform weighted, triangular, quartic and trikubiske churned.

[0045] In one embodiment, the calculated similarities of the interrogation input x combined with each of the sample values ​​X, to generate estimates of the output, that is, the estimated observations 60. This is in KR carried out for example by calculating a weighted average of the output examples by using the similarities between asking observation and innmatingseksemplene as weighted parameters, as shown in the following equation: ΣΙ (χ, -χ) ν;] YW - ^ -.

Σκ (χ, -χ) ii where "n" is the number of observation examples in KR model, "X," and Ύ "is input and output for it. Observation example, x is a query-input, K (X, - x ) is a core function and y (x) is an estimate of y, if x is provided.

[0046] In one embodiment, varying the number and types of feeds and outputs analyzed using various KR architectures. The variables and feeds as described herein, in an embodiment represented by vectors when more inputs are used. A inferensial KR model uses multiple feeds to derive an output, one heteroassosiativ KR model uses several inputs to predict multiple outputs, and one autoassosiativ KR model (ÅKRE model) uses feeds to predict the "correct" values feeds, where "correct" refers to the relationships and behaviors that are a content observation examples.

[0047] Referring to FIG. 6, where the estimated observations 60 in the second stage is used to determine if an error has occurred. A number of residues 62 corresponding to the number "Nq" of observations 58 is calculated by subtracting each observed estimate 60 from a corresponding ask observation 58. The resulting restobserva-tions 62 each have a value representing a change in the correlation of the error-free observation.

[0048] Each residue observation 62 is then passed to the detector 44 which employs a statistical test to determine if the appropriate sequence of residual observations 62 more likely generated by a nominal mode (measurement that there is no fault) or degraded manner (meaning that there is an error). In one embodiment, residue 62 observations is evaluated using a cumulative sum (CUSUM) or a sequential probability ratio test (SPRT) detection statistics to determine whether the tool operates in a normal or impaired manner.

[0049] In one embodiment, the threshold values ​​to determine whether the tool 20 operates in a degraded manner determined. In an example, the nominal manner defined during the training, and a plurality of ways are degraded number with respect to the nominal manner. Each degraded manner corresponding to a selected threshold. Mean oppforskjøvne and middle nedforskjøvne degraded models are eg. defined by shifting the nominal distribution of the respectively higher and lower mean value. A series of tests is then performed to indicate the distribution sequence is most likely generated by.

[0050] In one embodiment, a sequential analysis such as a sequential probability ratio test (SPRT) performed to determine whether residual observation 62 is a result of the nominal operating mode or degraded mode of operation. SPRT is used to determine if a sensor is more likely in a nominal mode "Ho" or in a degraded mode, "Hi". SPRT comprises calculating a likelihood ratio, "L_N", shown in the following equation:

<img img-format="tif" img-content="drawing" file="NO20101425AD00141.tif" id="idf0001" />

where {xn) r a sequence of "n" successive observations of x. Probability ratio is then compared with a lower (A) and an upper (B) limit, such as those defined by a false alarm probability (a) and a lost alarm probability (P) shown in the following equations:

<img img-format="tif" img-content="drawing" file="NO20101425AD00142.tif" id="idf0002" />

[0051] If the odds ratio is less than A, the residual observation 62 determined to belong to the system's normal operating mode the Main If the odds ratio is greater than B, the residual observation 62 determined to belong to the system degraded mode of operation Hi and an error is detected.

[0052] If no test result indicates that the residues likely not have been generated from the nominal operating mode, generates the detector 44 an alarm 64 which indicates that a fault in the tool 20 may have occurred. Such alarms 64 is referred to as "Alarm Obs. # 1- # Nq", and can be any number of alarms 64 between zero and NQ.

[0053] If the output from the detector 44 indicates that the tool 20 operates normally (ie no error or irregularity has occurred), then no maintenance or management action taken, and the system 60 examines the next observation. However, if the detector 44 indicates that the tool 20 works in a degraded way, the prediction and detection results forwarded to the diagnostic unit 46 which assigns produced symptom patterns 66 (ie prediksjonsrester, signals, alarms, etc.) To known fault conditions to determine the nature of the error.

[0054] Referring to FIG. 7, wherein the symptom patterns 66 in the third step is produced by the processor 40 which encapsulates a sufficient amount of information to differentiate between the identified errors. Symptom patterns 66 are referred to as "Symptom Obs. # 1-NQS" in FIG. 7 in which "NQS" is a number less than or equal Nq. Symptom patterns 66 are calculated by combining the data from the predictor 42 and the detector 44, comprising one or more of the interrogation observations 58, estimate observations 60, restobserva-tions 62 and alarms 64 for each signal. In one embodiment, the additional information from the storage dump data 34 as further signals or a synthesis of additional signals, and / or signals that can be used to quantify environmental or operating stresses, combined with the data from the predictor 42 and the detector 44 to provide symptom observations 66.

[0055] In one embodiment, the residual observations 62, optionally in combination with alarm 64 is provided as the symptom patterns 66. Examples of symptom patterns 66 includes measured hydraulic unit signal values ​​alone and with associated residues, fringe signals (i.e., a rate at which a staying rotates its shaft) with corresponding estimates residues, and vibration signals corresponding estimate residues.

[0056] Referring to FIG. 8, where the observations in the fourth step, associated alarms and residues, is included in the diagnostic unit 46.1 one embodiment, the diagnostic unit 46 includes a NFI-diagnostic unit. In another embodiment, only data relating to observations that generates an alarm 64, introduced into the diagnostic unit 66.

[0057] In one embodiment, the symptom observations 66 introduced into the diagnostic unit 46 which derives the class or type of error for each symptom observation 66. Class of class (i.e., class "A" - "Z") is carried out by compare symptom observations 66 with examples of symptom patterns previously generated by the diagnostic unit 46, and then combining the results of this comparison to each example of the symptom pattern to generate an estimate 68 of the class. In one embodiment, each symptom observation 66 compared with symptom patterns, and is assigned to a class that is associated with the symptom pattern as it most closely resembles. This class estimate 68, referred to as "class-estimate Obs. # 1- # NQS "in FIG. 8, is produced for each observation 58 exhibiting a fault. In one embodiment, the frequency of classes (eg. Class A, Class B, etc..) In the estimated observations 60 determined to provide a definitive diagnosis of the tool 20 and / or its components.

[0058] Errors may occur for any of various reasons, and corresponding defect classes are designated. Examples of error classes include "mud invasion" (Ml) for drilling fluid 16 flowing into a tool 20 and cause errors, "pressure transducer offset" (PTO), wherein the sensor offset (negative and positive) causes problems in controlling the system 10 to ultimately resulting in system failure, and "pump start" (PS), wherein a pump fails after the drill is started.

[0059] In one embodiment, the "nearest neighbor" classification of (NN) classification of) used to determine the class of one symptom observation 66 falls into, involving allocation of a non-classified sample point where the classification of the nearest of a set with previously classified points are located. An example of nearest neighbor classification k-nearest neighbor (KNN). KNN refers to the klassifikasjonsanordningen which examines the number "k" nearest neighbors to a query pattern, and NN refers to the classification device that examines the nearest neighbor (i.e., k = 1). NN classification comprises calculating a distance between a query pattern and each example of the symptom pattern and associate ask pattern to a class that is associated with the example of symptom pattern which has the smallest distance.

[0060] KNN classification involves calculating distances for each example of symptom pattern, sort of distances and extraction of starting classes for the k. Minimum distances. The number of cases of each class represented by the k. Minimum distances are counted, and the class to ask pattern is designated as the class with the largest representation in the k nearest neighbors.

[0061] An example of nearest neighbor classification are described here. In this example, a number "n" with examples of symptom patterns collected for "p" in nnmatinger (i.e., variables) that are examples of a number of "nc" classes. "C" refers also to the in. The class and "n" denotes the number of samples for a class. Using these definitions, the sum of the number of examples for each class equals the number of samples of symptom patterns.

[0062] In this example, the training mind feeds (i.e. examples of symptom patterns) designated by X, and the outputs (i.e. classes) is denoted by Y. "memory matrix" or vectors are generated for input and output as follows:

<img img-format="tif" img-content="drawing" file="NO20101425AD00171.tif" id="idf0003" />

[0063] Classification of a query for observation p inputs, denoted by x is performed. Ask observation x is represented by the following equation:

<img img-format="tif" img-content="drawing" file="NO20101425AD00172.tif" id="idf0004" />

[0064] The distance, such as euklidavstanden, can be used to determine how close observation ask each of innmatingseksemplene. In mimetic form, the distance of the query to it. As given by:

<img img-format="tif" img-content="drawing" file="NO20101425AD00173.tif" id="idf0005" />

[0065] Distance calculation is repeated for the n samples, where the result is a vector of n distances:

<img img-format="tif" img-content="drawing" file="NO20101425AD00181.tif" id="idf0006" />

[0066] To classify x nearest naboklassifikasjonsenhet is output or the class classification example corresponding to the minimum distance.

[0067] The different types of methods of classification used herein are intended only as examples. Any number or any type of technique can be used for comparison of data patterns from a sensor or multiple sensors with known data patterns of misclassification.

[0068] Referring to FIG. 9, where a deterioration in path 70 and a corresponding life 72 in the last step is calculated for each signal. Deterioration paths 70 are referred to as "deterioration path # 1 to # Nqd" and lifetimes 72 is referred to as "lifetime # 1- # Nqd" where Nqd is the number of degradation paths 70. From these data, the remaining useful life of the tool was calculated. Deterioration path 70 is generated by combining data from the predictor 42, the detector 44 and the diagnostic device 46, comprising one or more of signal observations 58, the signal estimates 60, estimate residues 62, alarms 64, symptom observations 66 and class estimates 68. Further information from the storage dump data 34 may also be combined, such as additional signals or composite signals (eg. running total over a threshold), to produce deterioration paths. Any suitable data fitting or regression technique may be applied to the data acquired from the tool, to generate degradation pitch. Many types of statistical analysis is used to calculate the deterioration path, such as polynomial regression, effektregresjon, etc. For simple data relationships, and application of the fuzzy interference system, neural networks, and so on for complex relationships.

[0069] Deterioration path 70 can be generated from any desired measurement data. Examples of such data used for degradation paths, comprising: crack length in the drilling string, the measured pressure, electric power, motor and / or bit rotation and temperature over a selected time period.

[0070] lifetimes 72, corresponding to each deterioration path 70 is generated. In one embodiment, a threshold value is determined for deterioration path 70, to indicate a failure. This threshold may be based on extrapolation of data from the existing deterioration path 70, or based on pre-existing examples of degradation pathways in conjunction with known failure times.

[0071] Referring to FIG. 10, where deterioration paths 70 and lifetimes 72 is introduced into the projection unit 48, which uses this information to generate estimates of remaining useful life (RUL) 74 according to each court. RUL-value for each path may be referred to as "RUL estimate # 1- # NQD". In one embodiment, the projection device 48 includes a NFIs-forecasting unit. Deterioration paths 70 are compared with Examples of deterioration paths, and the results of the comparison with samples of lifetimes are compared to generate an estimate 74 of the tool 20 and / or component RUL values. In one embodiment, a classification and estimasjonsmodell (PACE model) using an associated PACE algorithm used to generate RUL estimate 74.

[0072] PACE algorithm is used in situations where i) each deterioration path 70 includes a discrete error threshold that accurately predicts when a device will fail, and ii) degradation pathways 70 shows no clear error threshold. In one embodiment, for example, to deterioration paths 70 which exhibits well-established thresholds (eg. Seeded crack growth and controlled test environments, such as constant load or uniform cycling), the data are formatted so that the moment when the deterioration path 70 crosses the fault threshold, interpreted as an error event.

[0073] In other embodiments, a defined discrete error threshold is not always available. In some such embodiments, especially in many real applications where failure modes are not always well understood or may be too complex to be quantified by using a single threshold, erfeilgrensen gray for the best.

[0074] PACE algorithm involves two general observations: 1) to classify a current deterioration path 70 as belonging to one or more previously collected examples of degradation paths, and 2) to use the resulting membership to estimate RUL value.

[0075] Referring to FIG. 11, in which examples of deterioration signals 76 are shown, represented as "Yi (t)", and their associated time to failure (TTF,). In this example, it can be seen that there is a clear threshold for degradation path 70.1 one embodiment, such signals 76 Generalized by adopting an arbitrary function 78, referred to as "fj (t, Øi)," the data by regression, machine learning, or second adaptation techniques.

[0076] In one embodiment, the two information elements extracted from degradation pathways, especially TTF values ​​and "shape" of deterioration described using functional approximations fj (t, .PHI.I). These information items are used to construct an example of TTF values ​​and functional advances, as follows:

<img img-format="tif" img-content="drawing" file="NO20101425AD00201.tif" id="idf0007" />

where TTFi and fi (t, Øi) is TTF and the functional approximation of it. example of deterioration signal path, Øi the parameters for it. functional approximation of it. example of deterioration signal path, and Θ is all the parameters for any functional approximation.

[0077] In one embodiment, the deterioration path computed using a general path model (GPM General Path Model). GPM involves parameterization of a device's deterioration signal to calculate depreciation path and determine TTF value. In one embodiment, the TTF-value is described as a probability of failure depending on time. TTF-value can be determined by any selected probability for error or failure.

[0078] In one embodiment, the generic PDF values ​​appropriate for a deterioration signal to measure deterioration pitch and TTS-value. If, for example. N devices are tested and Nr is the total number of devices that have failed up to the present time T, then the proportion of devices which have failed interpreted as the probability of failure for all times less than or equal to the current time. More particularly, the cumulative probability of failure at time T, denoted by P (T &lt;t), the relationship between the actual number Failed devices (NT) and the total number of devices (N), as shown in the following equation:

<img img-format="tif" img-content="drawing" file="NO20101425AD00211.tif" id="idf0008" />

[0079] If a generic sannsynlighetsdensitetsfunksjon (PDF) is adapted observed error data, then the above equation is written in terms of a PDF, referred to as "f (t)" and its corresponding continuous distribution function (CDF, ​​continuous distribution function), referred to as "F (t)":

<img img-format="tif" img-content="drawing" file="NO20101425AD00212.tif" id="idf0009" />

[0080] The above equation can also be used to define the probability that an error or failure has not occurred for all times less than the current time t, referred to as the reliability function "R (t)":

<img img-format="tif" img-content="drawing" file="NO20101425AD00213.tif" id="idf0010" />

[0081] In one embodiment, additional reliability measures calculated using the TTF distribution data and reliability features for predicting and report the failure, namely the average time error (MTTF) and 100 p. Percentage reliability function. MTTF characterizes the expected error time of a test device drawn from a population. The following equation can be used to calculate MTTF for a continuous TTF distribution:

<img img-format="tif" img-content="drawing" file="NO20101425AD00214.tif" id="idf0011" />

and can be further defined in terms of the reliability function:

<img img-format="tif" img-content="drawing" file="NO20101425AD00215.tif" id="idf0012" />

[0082] In one embodiment, can as an alternative to MTTF, 100 p. Percentile reliability function is used to determine the time (tp) in which a specified percentage of the devices has failed. In mimetic shape, the time at which 100 p.% Of the devices failed, simply expressed as the time the reliability function has a value of p:

<img img-format="tif" img-content="drawing" file="NO20101425AD00221.tif" id="idf0013" />

where p has a value between zero and one.

[0083] Referring to FIG. 12, the RUL is calculated for an observed deterioration path 70. Deterioration path 70 has a value "y (t *)" degradation path 70 at a time T ". To estimate the RUL of the device via PACE model, the algorithm is represented in FIG. 13, was used.

[0084] Referring to FIG. 13, where an example of a method 80 for estimating the RUL, in one embodiment, a plurality of steps 81-83.

[0085] In the first step 81, the expected deterioration signal values ​​according to the examples of deterioration paths 76 estimated by evaluating the regression at t *. The current time t * is used to estimate the expected values ​​for deterioration path 70 according to Example paths 76.1 one embodiment, is the expected values ​​of deterioration path 70 according to court Examples 76 approximation functions 78 evaluated at time t *, as shown in the following equation:

<img img-format="tif" img-content="drawing" file="NO20101425AD00222.tif" id="idf0014" />

[0086] The values ​​of the above function evaluations can be interpreted as examples of deterioration path 70 at time t *. In this regard, the above vector is rewritten as follows:

<img img-format="tif" img-content="drawing" file="NO20101425AD00231.tif" id="idf0015" />

[0087] In step 82, becomes the expected RUL values ​​calculated by subtracting the current time t * of the observed TTF values ​​such as paths 76. This is shown, for example. in the following equation:

<img img-format="tif" img-content="drawing" file="NO20101425AD00232.tif" id="idf0016" />

[0088] In step 83, the observed deterioration path 70 at time t *, y (t *), classified based on a comparison to the expected deterioration signal values ​​Y (t *). Deterioration path 70 are classified as belonging to the class associated with the instance path 76 which is the closest in value. In one embodiment, the signal value y (t *) is compared with the expected deterioration signal values ​​Y (t *) using one of a plurality of classification algorithms to provide a vector with memberships Hy [y (t *)]. In this embodiment, memberships values ​​equal to zero or one, and fjyi [y (t *)] denotes membership to y (t *) to it. Instance court, as shown in the following equation:

<img img-format="tif" img-content="drawing" file="NO20101425AD00233.tif" id="idf0017" />

[0089] The vector of the memberships of the signal value y (t *) of the examples of deterioration paths 76 are combined with the vector of expected RUL values ​​to estimate RUL value for the individual device.

[0090] In one embodiment, the estimate of the RUL of a device generated using one or more of many types of projection devices, comprising a population prediction unit to estimate RUL value from the population-based error statistics, and individual projection devices including a causal-prediction unit estimating RUL value by monitoring the causes of component failure / failure (eg. by examining the voltage signals such as vibration, temperature, etc.), and a power prediction unit to estimate RUL value by examining the effects of component failure / failure each device by examining the output of a monitoring system. In one embodiment, multiple power units projection provided to estimate RUL value for each fault class.

[0091] In one example, utilizes causal-projection device data absorbed vibrational energy to estimate RUL value by examining the cause of failure. In another example, calculate the output projection device a cumulative sum of the alarms 64 which is used to estimate RUL value by examining the effect of the beginning of the failure.

[0092] In one example, the population projection device continuously used to estimate RUL value by calculating the expected RUL value given the current time as the device size has been used below. Stressorsignal data (eg. Vibration, temperature, etc.) Are additionally used as inputs to causal-projection units for each of the identified effects, which estimate RUL value by examining the amount of absorbed mechanical stress of the device. Relevant data signal is likewise also extracted from the collected device data and used as inputs to a monitoring system determines whether the device currently operating at a nominal or impaired manner. If the monitoring system indicates that the device operates in a degraded fashion, then the original signals and monitoring system outputs used as inputs to a diagnostic system that thereafter selects an appropriate effect projection unit based on the observed patterns. If, for example. diagnostic unit 46 classifies the current operation of the device representative of the in. wrong class, it will, in. power-projection device may be used to estimate RUL value.

[0093] Referring to FIG. 14 wherein an alternative exemplary system 80 includes a device database 82, a monitor 84, a diagnostic system 86, one population projection unit 88, one Ml-causal prediction unit 90, one PTO cause projection unit 92, one Ml effect prediction unit 94 and the PTO power prediction unit 96. The monitor 84 includes, for example. predictor 42 and detector 44. The diagnostic system 86 includes, for example. diagnostic unit 46.

[0094] Population prog rose unit 88 receives operating data and generates RUL value from these. Ml- and PTO causal forecasting units 90, 92 receive time data and kausaldata, such as vibration data, and predicts RUL value for the absolute vibration energy. Ml- and PTO power projection units 94, 96 receive data generated by the diagnostic system 86 and calculates RUL value from this. In one embodiment, the ML and PTO power projection units 94, 96 trained to estimate RUL value of mud invasion (Ml) and trykktransduserforskyvnings (TTO) errors. In one embodiment calculates ML and PTO power projection units 94, 96 RUL value of the cumulative sum of false alarms 64.

[0095] Although cause and effect projection units using ML and PTO error classes while generating the RUL-value, the system 80 is not limited to specific error classes. Although cause and effect prediction devices are likewise described in this embodiment as NFIs-projection units, may projection devices utilizing any suitable algorithm.

[0096] In one embodiment, to develop population prediction unit 88, data is collected from a number of devices that are exposed to normal operating conditions or accelerated life testing to extract information about the time error (TTF) for each device. The cumulative TTF distribution is then calculated. The first step in the development of population prediction unit 88 is to adapt a probability time density (PDF) for TTF-data, such as the cumulative TTF distribution. In one embodiment, to fit the data, a cumulative distribution function (CDF) associated PDF estimation value, and the resulting estimates are used to estimate the parameters of a general distribution. More PDF values ​​can be adapted to the data via, for example. least squares method, to determine the best model for the wrong times.

[0097] Other functions can be generated by population projection device 88. Population projection device 88 may for example. using accelerated life testing or proportional risk assessment for defining failure rate as a function of time. In one embodiment, the population risk model also take into account different mechanical stress variable in addition to time variable.

[0098] In one embodiment, an individual based projection unit used to determine RUL value. Examples of individually based projection units include cause and effect-projection units 88, 90, 92, 94 and 96. The individually-based forecasting devices uses some examples GPM and produce RUL or reliability estimates. In embodiments that use GPM, the device's degradation-treated as an initiation of a progression to a fault threshold. Examples of algorithms that use GPM include categorical data analysis, life consumption rings modeling and proportional hazard modeling, each producing either reliability estimates or RUL. Another example of an algorithm that uses GPM comprise different extrapolation methods used to produce RUL value. An example of an algorithm that does not use GPM, a neural network algorithm that is used to produce RUL value.

[00100] In one embodiment uses the individually based prognosealgoritmene following method. First, examples of degradation paths characterized by determination of the "shape" of the web and the critical error threshold. The term "form" refers to the parameter values ​​of the deterioration signal and the shape of a physical model for different aspects of a device such as deterioration, parameters and shape of the regression equation pitch. In this embodiment must examples of degradation paths are not generated by providing examples, but may be a product of the physical models of deterioration mechanism. Error threshold can be set manually if it is known or can be derived from orbit examples.

[00101] The results of the court-parameterization and the threshold is then used to construct an individual forecasting model. Unite test devices for estimating the reliability (that is estimating a probability of failure) or an RUL-value at a certain time t, is eventually the current progression of the test track presented as an input to the projection algorithm, which produces an estimate of the device's reliability or RUL value.

[00102] Various algorithms or models may be used to parameterize the examples and the measured signals deteriorate (eg. Environmental or driftspåkjennings signals) to generate degradation pathways, and to estimate RUL value. Examples of such algorithms are described here.

[00103] Categorical data analysis algorithms (CDA algorithms) uses logistic regression to assign observed deterioration parameters to one or two conditions, such as "no fault" (0) and "wrong" (1). CDA using logistic regression to establishing a relationship between a set of inputs (continuous or categorical) for categorical outputs.

[00104] In this method, the probability of error for an observation of deterioration signals estimated using a logistic regression model trained on historical data deterioration. For each degradation signal there is an associated critical threshold, and an error is considered to have occurred when any of deterioration signals crosses its associated threshold. This method provides a reliability estimate, but does not generate RUL value. In an embodiment, different time series analyzes, such as auto-regressive moving average (ARMA) or curve fitting used to extrapolate the deterioration signal to a svikttid where reliability is zero or where the extrapolated line crosses the threshold and thus estimates the RUL value.

[00105] By proportional hazard modeling (PH modeling, proportionel hazard modeling) depends on the error rate or hazard function of the current time as well as rows of variable stress factors that describe environmental and operational stresses to which a device is exposed. Another example for the estimation of RUL is life-consuming modeling (LCM, life consumption modeling). At LCM begins a new component's life in perfect condition / reliability. As the device is used and / or exposed to different operating conditions, the condition / reliability degraded with size is related to the damage that is absorbed by the device. An example of an LCM algorithm is accumulated damage modeling (ADM, Accumulated Damage modeling) using foul classes stresses to estimate the increment at which the component's condition has deteriorated after each use. Another similar solution is the cumulative wear model (CV model, the Cumulative wear model) that estimates the direct reliability of a device for incremental reduction of its reliability as it is used.

[00106] Extrapolation methods generally involves extrapolation of the state of the apparatus using a priori knowledge and observations of historical device operation. In general, extrapolation is carried out by either: 1) to predict future device stresses and then applying strain-states on a model of the device-degradation to estimate RUL value, or 2) use trend techniques to extrapolate path forringelses- or reliability signal to an error threshold.

[00107] Various prior knowledge can be used to estimate the future environmental and operational stresses. This knowledge may be in the form of several stress-functions (ie stressors), each over a specific time interval. A deterministic sequence, for example. used if future stress levels and exposure times is known, by iteratively feeding the predetermined stress levels and exposure times to a model for device degradation to estimate the future state of the device.

[00108] In population-based probabilistic sequence methods are historical data collected from a population of similar devices, used to estimate the probabilities of occurrence of specific stress levels and exposure times. In individually based probalistiske sequencing methods are historical data collected from the individual device used to estimate the probabilities. To estimate the distribution of RUL values ​​for a device given its current state, the simulations, such as Monte Carlo simulations covering the stress levels and exposure times are sampled according to the estimated probabilities. Finally, the RUL value for the individual device estimated by taking the expected value of the resulting PDF RUL values.

[00109] Other examples of forecasting algorithms including fuzzy-prediction algorithm such as fuzzy inferenssystemer (FIS) and adaptive neural fuzzy inferenssystemer (Anfisa). Different regression and neural networks and other analytical techniques can be used to estimate RUL value.

[00110] The systems and methods described herein, provides various advantages over prior art techniques. The systems and methods described herein, is simpler and less cumbersome than prior art techniques generally employing detailed physical models or expert systems troublesome. In contrast to methods that cause the data structure using physical models or detailed expert systems, derives the systems and methods described herein, the structure of data by allowing samples to define analysis components correctly.

[00111] Since the systems and methods described herein, as well as use data-driven techniques (ie data defining the model), the resulting systems simple to automate and are flexible enough to be adapted for changing deployment requirements.

[00112] As support for the teachings set forth herein, various analyzes and / or analytical components used, including digital and / or analog systems. The system may have components such as a processor, storage media, internal storage, input, output, communication links (hard wired, wireless, pulsed mud, optical or other), interface, software, signal processors (digital or analog) and other such components (such as resistors , capacitors, inductors and others) to provide for operation and analyzes of the devices and methods described herein at one of several ways well known in the art. It is intended that these teachings may be, but need not be, implemented in conjunction with a set of computer executable instructions stored on a computer readable medium, comprising internal memories (ROM, RAM), optical (CD-ROM) or magnetic (discs, hard -disker) or of any other type which, when executed, gets a computer to implement the method of the present invention. These instructions may provide for equipment operation, data collection and analysis and other functions deemed relevant for a system designer, owner, user or other similar staff, in addition to the functions described in this description.

[00113] Various other components may further be included and called upon for providing aspects of the specified invention. For example, a prøvetakningsledning, a sample-ekstraheringpumpe, a piston, a power supply (eg. At least one of a generator, a remote-supply and a battery), vacuum supply, a freezing unit (i.e., cooling device) or supply, a heating component , a driver (such as a translational force, a progress force or a rotational force), magnet, electromagnet, sensor, electrode, transmitter, receiver, transceiver / receiver, controller, optical unit, electrical unit or electromechanical unit may include, for support of the various aspects discussed herein or as a support for other functions beyond this disclosure.

[00114] One skilled in the art will recognize that the various components or technologies may provide these necessary or beneficial functionality or features. These functions and features as may be required for support of the appended claims and variations thereof, are accordingly considered inherently included as a part of the here specified learn and as part of the disclosed invention.

[00115] Although the invention has been described with reference to embodiments, the skilled artisan will appreciate that various changes can be made and equivalents may be substituted for elements without departing from the scope of the invention. Many modifications will also be found by those skilled in the art to adapt special instruments, situations or materials to the teachings of the invention without departing from the main frame for this. It is therefore intended not to be limited to the particular embodiment disclosed as the best way that can be envisaged for carrying out the invention, but that the invention will include all embodiments falling within the scope of the appended claims.