Statistical method for analyzing the performance of oilfield equipment

[0001] This application is a continuation of U.S. application Ser. No. 11/461,911 filed Aug. 2, 2006, which is pending.

[0002] 1. Field of the Invention

[0003] The present invention generally relates to a statistical method which provides time-to-event estimates for oilfield equipment, and, more particularly a method which utilizes survival analysis techniques for analyzing time-to-event data.

[0004] 2. Description of the Prior Art

[0005] Currently oilfield tool (and equipment) performance prediction and analysis is conducted in an ad hoc fashion with varying degrees of sophistication and quality of interpretation. A major concern is that of bias being introduced into the analysis and hence into the results, either through the exclusion of data or by assumptions about the performance of equipment at the time of data sampling. At the time of an analysis, the analyst has a population of capital intensive oilfield equipment from which to draw data. Some of this equipment may have already failed at the time of the data was extracted, while other installed equipment is still fully operational and had not failed at the time the data was extracted. This latter subset of the population has been the subject of improper analysis in the past in two particulars: (i) complete exclusion of the data set; or (ii) the inaccurate assumption that, at the time the data was drawn, the equipment had failed.

[0006] Survival analysis is a statistical methodology and testing hypothesis of time-to-event data that has, for example, been applied in the medical field to analyze time-to-death of a patient after surgery, the cessation of smoking, the reoccurrence of disease. For most statistical applications, models for probability distributions are usually described in terms of:

• • [0007] Probability Density Function (pdf) f(t): a function whose integral over a given range is equal to the probability of taking a value in that range.
  • [0008] Distribution Function F(t) (cumulative density function): the probability of the event occurring by time t.
    For survival analysis, however, it is appropriate to work with different functions characterizing the distribution:
  • [0009] Survival Function S(t): the probability of surviving at least to time t [sometimes known as a reverse cumulative density function: 1−F(t)].
  • [0010] Hazard Function h(t): the potential of failure in the next instant given survival to time t.

[0011] An Explanatory Variable (EV) is a variable that may influence equipment behavior. In conventional product-limit analysis, the investigation of a single EV requires partitioning of data set into subsets for each level of the EV and analysis is then performed independently on each subset. This has the effect of thinning the data which may result in less reliable statistics. In an investigation of two or more EVs, this problem is compounded. For example, if it is desired to predicate how an Electrical Submersible Pump (ESP) System would behave in a deviated well (true/false) and an openhole well (true/false), four data subsets exist to examine independently. As the number of EVs in an analysis is increased, there will be some subsets that are sparsely populated.

[0012] A rigorous statistical methodology has been developed to provide time-to-event (e.g., failure) estimates for of oilfield equipment. This technique extracts unbiased information from equipment performance data and can consider parameter interactions without recourse to data thinning. The analysis explicitly accounts for items of equipment that are still operational at the time of the analysis, thus removing a significant source of bias in the results.

[0013] When the collection of the data on the equipment to be analyzed is made, certain items of the equipment have yet to fail. To ensure that the full population of equipment and tools is fully considered without any bias entering into the analysis, a method in accordance with the present invention comprises the step of assigning a censoring flag (e.g., 0=failure, 1=censored) to the equipment, e.g., an ESP. This step of assigning a censoring flag is a distinguishing feature of the present invention and permits the application of a tried and tested statistical survival analysis to the data. For oilfield equipment, survival analysis is effectively the only reasonable, bias-free and consistent approach to a performance analysis.

[0014] The first stage of a method according to the present invention comprises Exploratory Data Analysis (EDA). This first stage comprises the utilization of both Cox Proportional Hazard (CPH) and Kaplan-Meier (KM) modeling approaches to allow the user to become acquainted with the data and to recognize anomalies and outliers. The primary purpose of EDA is to obtain a reasonable initial model for the stepwise model selection procedure (Stage 2). This initial model is realized through the application of various statistical tests that identify significant Explanatory Variables (EVs) when considered singularly. The tests may, for example, be log-rank or Peto tests for KM and likelihood ratio test (LRT) for CPH. A process according to the present invention may further comprise the application of a CPH-related test to consider the assumption of proportional hazards used in the CPH model. EDA may also identify potential candidates for later transformation and grouping.

[0015] The second step in a method according to the present invention comprises Stepwise Model Selection. In one embodiment, a stepwise model selection is applied using the Akaike Information Criteria (AIC). All possible combinations of the EVs that were found to be significant in the EDA stage are the range of models examined in the stepwise search. The initial model is the most complex from this range, i.e., that which includes all of these EVs. A result of this second stage is a model that has identified the significant parameters in combination as opposed only identifying significant EVs individually.

[0016] The third step in a method according to the present invention is Factor Collapsing. In this step, the number of levels of a factor is economized, e.g. by using a backward elimination process using LRTs. The process iteratively considers the candidate models formed by all possible pair-wise joining of factor levels within the EVs in the current model. The least significant amalgamation is accepted for the current model of the next iteration. If all possible collapses have significant p-values, then the process is stopped.

[0017] The fourth step in a method in accordance with the present invention comprises the inclusion of interactions/other EVs. In this stage, consideration is given to model refinement by applying stepwise model selection using AIC from the current model. The range of models is bounded by the most complex that includes the current EVs, their pair-wise interactions and the excluded EVs, not just those identified in Stage 1. The inclusion of previously disqualified EVs allows the confirmation that they are indeed not necessary. The inclusion of pair-wise interaction parameters allows the capture of EV effects that are not behaving in an additive way. In other words, interactions allow an EV to have a different influence on survival over the different values provided by considering another EV.

[0018] The fifth step of a process according to the present invention is Model Checking where the proportional hazard assumption of the final model is tested. A global test is employed to see, if overall, the model violates this assumption. If it does, then proportional hazard assumption tests for individual EVs can suggest which of them violate the assumption. Any such variable needs to be declared as a strata variable in a stratified CPH model. Here separate baseline functions are fitted for the levels of each violating EV.

[0019] A method according to the present invention may also be utilized to apply survival analysis to any oilfield equipment components where time-to-event information has been recorded. The method of the present invention allows comparative reckoning between different components present in the system comprising several, or many, individual components, and allows analysis of these components either individually or simultaneously (in the presence of other components).

[0020] A method of performing a survival analysis on the components of an item of equipment comprises the step of representing the data on said components in a counting process formulation of a Cox Proportional hazards model. A method of the present invention next comprises the step of applying an extension of the Cox Proportional Hazard model to the data, and in one embodiment, the extension which is applied is as described in Lunn, M. and McNeil, D., “Applying Cox Regression to Competing Risks,” Biometrics (June 1995) [hereafter “Lunn extension”]. A method according to the present invention next comprises the step of performing individual survival analysis on each of the identified components by using the five-stage method described above. The final step of a method according to the present invention is to establish the time-to-event estimate for each of the recorded components, either taken individually or in the presence of the other components, such that an estimate of competing risks is obtained.

[0026] It will be appreciated that the present invention may take many forms and embodiments. In the following description, some embodiments of the invention are described and numerous details are set forth to provide an understanding of the present invention. Those skilled in the art will appreciate, however, that the present invention may be practiced without those details and that numerous variations and modifications from the described embodiments may be possible. The following description is thus intended to illustrate and not to limit the present invention.

[0027] A method in accordance with the present invention is described with respect to an Electrical Submersible Pump (ESP) system comprising 117 separate records for ESPs either currently or previously installed. Those skilled in the art will also appreciate that a method according to the present invention is not limited in its application to ESP systems, but rather may be applied to numerous types of oilfield equipment systems where any specific time-to-event understanding is desirable from both an operational assurance and financial standpoint. This is typically the case for equipment in place for extended periods of time, like ESPs, valves, permanent gauges, etc.

[0028] A method in accordance with the present invention utilizes a standard statistical software package. Several such packages are commercially available. In one embodiment, the R statistical software package is utilized in the method of the present invention.

[0029] As used in this specification and in the appended claims:

[0030] 1. The term “Explanatory Variable” (EV) means a variable that may influence equipment behavior. An EV may be a “factor” which is a categorized variable such as the type of equipment or a “covariate” which is a numerical variable.

[0031] 2. The term “parsimony” means “praiseworthy economy in use of means to an end; avoidance of excess.” The term “parsimonious” is used to describe a model which has parsimony.

[0032] A method according to the present invention may be applied to a set of data, e.g. EVs, that has been compiled or collected on the equipment to be analyzed. Since certain items of equipment had not yet failed at the time of data collection, a method according to the present invention assigns a censoring flag to each such item of equipment. For example, this censoring flag may equal “0” if the equipment has failed and “1” if the equipment is censored.

[0033] A basic data check is then undertaken to ensure compatibility in the analysis and to prevent system software and algorithmic failure and crash due, for example, to divisions by zero or an unrealizable numbers of permutations plus, spurious data, inconsistent data (i.e. characters where numerical values were expected, etc.). This basic data check involves checking for and discarding any EVs that are factors with low counts and the threshold delineating low counts can be set and changed by the user. In one embodiment, if a level has three or less members, then it is tagged accordingly. Such variables do not contain contrasting information that would allow for a variable effect to be estimated, and therefore, should be excluded from the analysis as separate levels; rather, they are gathered together into a new composite level terms “Other”.

[0034] The method of the present invention was applied to data collected from a system of ESPs from fields in Ecuador, South America operated by a single major operating company. This data was selected for demonstration due to its tractable size (small enough to enable testing and debugging) yet rich enough to demonstrate all the necessary concepts. Furthermore the data was “good” i.e. no data-holes or quality issues. Each ESP initially had 64 EVs associated with it. Table I below is a list of the final 57 EVs considered in the ESP analysis from the original 64 EVs associated with each ESP. The 57 were identified from a preliminary analysis of the data where perfectly correlated EV's were removed e.g. sysmanf and cabmanf were exactly the same, EV's with only 1 level were removed and also EVs with simply too many low-count (<=3) unique levels (i.e. well names) would result in just one level called “Other”. For the survival analysis that was applied to an ESP system, the lower part of Table I shows that eight EVs were removed from the main analysis for the reasons indicated in the preceding paragraph. Furthermore low-count data entries were grouped together into a composite class called “other” and are used in the analysis. These are distinct from zero-count entries, which are marked by “N/A”, and are not used.

[0035]

TABLE I
Index	Explanatory Variable	Type	Levels
1	Field ID	Factor	7
2	Well ID	Factor	12
3	Event ID	Covariate	N/A
4	Application	Factor	2
5	Deviated Well	Factor	2 (T/F)
6	Open Hole	Factor	2 (T/F)
7	Casing ID	Factor	3
8	Corrosion observed	Factor	2 (T/F)
9	Scale build-up in well	Factor	2 (T/F)
10	Abrasion observed	Factor	2 (T/F)
11	Application Engineer	Factor	6
12	Field Technician	Factor	8
13	Spooler Operator	Factor	6
14	Rig	Factor	9
15	Panel Type	Factor	8
16	Panel Manufacturer	Factor	2
17	Variable Speed Drive	Factor	3
18	Depth of ESP	Covariate	N/A
19	System Manufacturer	Factor	2
20	Pump Manufacturer	Factor	2
21	Pump Series	Factor	5
22	Pump Type	Factor	10
23	Pump Stages	Covariate	N/A
24	Number of Pumps	Covariate	N/A
25	Intake Manufacturer	Factor	3
26	Intake Serial	Factor	4
27	Intake Type	Factor	6
28	Protector Manufacturer	Factor	2
29	Protector Serial	Factor	4
30	Protector Type	Factor	11
31	Number of Protectors	Covariate	N/A
32	Motor Manufacturer	Factor	2
33	Motor Series	Factor	5
34	Number of Motors	Covariate	N/A
35	Motor Horsepower	Covariate	N/A
36	Motor Voltage	Covariate	N/A
37	Motor Amperage	Covariate	N/A
38	Pressure Instr. Manuf.	Factor	2
39	Cable Manufacturer	Factor	3
40	Cable Type	Factor	6
41	Actreas	Factor	7
42	Tubing ID	Factor	4
43	Tubing Type	Factor	5
44	Motor Controller Type	Factor	4
45	Wellhead ID	Factor	8
46	J-Box present	Factor	2 (T/F)
47	Meter Lead	Factor	5
48	Check Valve present	Factor	2 (T/F)
49	Bleeder Valve present	Factor	2 (T/F)
50	UMB present	Factor	2 (T/F)
51	Adaptor present	Factor	2 (T/F)
52	Y-Tool present	Factor	2 (T/F)
53	Centralizer present	Factor	2 (T/F)
54	Cable Protector present	Factor	2 (T/F)
55	Cable Band	Factor	3
56	Phasing	Factor	2 (T/F)
57	Grounded	Factor	2 (T/F)
		Removed due to
58	Packer present	Low count
59	Shroud Present	Low count
60	TVSS installed	Low count
61	Number of Intakes	Related to EV #34
62	Number of pump sections	Related to EV #24
63	Operator ID	Related to EV #11
64	Client ID	1 level only

[0036] The first stage of a method according to the present invention comprises Exploratory Data Analysis of each EV that was not excluded from the analysis as a result of the basic data check. In this first stage, both KM and CPH for all non-excluded EVs are generated and considered.

[0037] The resultant statistic of only one single EV (namely FieldID from the ESP study) is illustrated for demonstrative purposes. The EV FieldID had seven levels (i.e., there were seven separate producing fields). FIG. 1 illustrates the statistical curves for the EV FieldID for the test ESP System, while FIG. 2 illustrates the CPH curves for the same data as used in FIG. 1. Table II below presents the summary statistics of the KM analysis for this variable. Table III presents the same FieldID statistical data as Table II using the CPH model instead of KM.

[0038]

TABLE II
KM summary statistics for explanatory variable FieldID for seven
fields in the ESP System (117 records with 2 NAs).
	Kaplan-
	Meier	Mean	se	Median	0.95	0.95
FieldID	n	days	days	days	LCL	UCL
1	9	500	119.9	427	382	—
2	20	423	74.3	368	211	—
3	20	375	133.1	138	105	262
4	9	214	89.5	49	27	—
5	26	624	180.1	167	143	436
6	5	452	157.3	457	93	—
7	26	342	44.6	293	249	509
Statistical tests:
log-rank: p-value = 0.1600;
Peto: p-value = 0.432.
Key:
se: standard error of the mean,
LCL and UCL: lower- and upper-confidence limits of the median respectively (in days),
n: number of counts.
Note that “—” denotes that the upper bound is undefined for the data provided.

[0039]

TABLE III
CPH summary statistics for explanatory variable FieldID for
seven fields in the ESP System (117 records with two NAs).
CPH	Mean	se	Median	0.95	0.95
FieldID	n	days	days	days	LCL	UCL
1	115	783	133.6	498	242	—
2	115	569	83.5	319	214	1775
3	115	264	23.5	161	98	318
4	115	229	18.2	154	50	466
5	115	466	61.3	260	166	546
6	115	494	67.3	264	149	—
7	115	511	70.8	293	211	649
Statistical tests:
LRT: p-value = 0.1950;
CPH = 0.818.
Key: refer to Table B.1.

[0040] In accordance with the method of the present invention, stage 1 exploratory data analysis (such as described above for FieldID) is repeated for each EV in the data set. The statistical threshold (p-values=0.05) is used to demarcate the significance of each EV. Any parameter falling below the critical p-value threshold indicates that this parameter is most likely to have a significant impact on performance. The higher a parameter is above this threshold, the less likely it is to have an impact on performance.

[0041] FIG. 3 illustrates a plot of the p-value for each EV in the data set for the ESP system. Table IV below shows the seven primary EVs that were identified, namely those characterized by falling below the critical threshold p-value=0.05. An additional three EVs (with indexes 8, 52 and 53) with marginal statistical significance were also identified. While these do not show as much significance as those in the primary list, they may turn out to be significant when building models when involving more than one EV. Both the KM and CM approaches identified the same set of seven significant EVs.

[0042]

TABLE IV
List of seven significant explanatory variables for single parameter
analysis (no collapsing) for both KM and CPH models (found at the
end of Stage 1. These were identified by (mostly) falling below the
critical threshold of p-value = 0.05. CPHA = Cox's Proportional
Hazards assumption test.
	KM
	Log-	CPH
Index	Explanatory Variable	rank	LRT	CPHA
16	Panel Man	0.0150	0.0188	0.8900
17	Variable SD	0.0150	0.0255	0.9830
33	Motor Series	0.0000	0.0008	0.3960
39	Cable Man.	0.0002	0.0022	0.0472
40	Cable Type	0.0025	0.0188	0.0305
55	Cable Band	0.0016	0.0037	0.0471
57	Grounded	0.0107	0.0120	0.0397

[0043] In accordance with the present invention, a stepwise model selection is now performed, which is an iterative procedure involving adding or deleting an EV at each stage. The choice to add or delete is made by considering all of the AIC values of models formed by single variable addition or deletion from the current model. Any models that have a lower AIC value than the current one are deemed to be better. For each iteration, the best model (i.e. the one with the lowest AIC) is chosen to be the current model for the next iteration. If there are no models that have a lower AIC than the current model, the procedure terminates.

[0044] Table V below demonstrates the stepwise procedure from the initial model that includes all seven significant EVs identified in Stage 1. The full output of the AIC analysis (not shown) provides information at each iteration regarding the AIC of all models formed by the deletion of an EV in the current model and by the addition of an EV not in the current model. The subsequent iterations in Table V have smaller AIC scores (indicating better models). In the above example, the current model at each step is formed by successive deletion of an EV and at no subsequent steps are these EVs ever returned into the current model. As noted above some EVs identified as significant in Stage 1 will not be deemed significant when combined with other such EVs. A reasonable model of data for Stage 2, as summarized in Table V has now been identified.

[0045]

TABLE V
The stepwise modeling process using AIC. The initial model comprises
the seven significant EVs identified in Stage 1. AIC is used to identify
a parsimonious model that only includes three of these EVs.
Step	Model Variables (per Table 1)	AIC
Initial	16, 17, 33, 39, 40, 55, 57	685.98
1	16, 17, 33, 39, 55, 57 (removed #40)	678.09
2	16, 17, 33, 55, 57 (removed #17)	676.18
3	33, 39, 55, 57 (removed #16)	674.75
4	33, 39, 55 (removed #57)	673.44
Final	33 (=Motor Series) and	673.44
	39 (=Cable Manufacturer) and
	55 (=Cable Bands)

[0046] The stepwise process of Stage 2 suggests that only the three EVs identified in Table V are required in a CPH model to adequately represent the data.

[0047] A method according to the present invention further includes the stage of factor collapsing, which may, for example, be implemented using an iterative process that at each stage identifies the “best” two levels to combine for any factor having three or more levels. Thus, at each iteration, every possible model formed by pair-wise combinations of the current factor levels are compared to the current model via a LRT. For each candidate model a non-significant p-value (greater than 0.05) indicates that there appears to be no significant difference between the two combined levels. The “best” of all candidate models is given by the least significant p-value: i.e., the one with the largest p-value above 0.05. The model that yields such value is then chosen for the current model in the next iteration. When no pair-wise combinations yield non-significant p-values, no further collapsing is possible because all remaining levels are significantly different and the algorithm terminates.

[0048] The fourth step in a method in accordance with the present invention comprises the stage of inclusion of interactions/other EVs. In this stage, consideration is given to model refinement by applying stepwise model selection using AIC from the current model. The range of models is bounded by the most complex that includes the current EVs, their pair-wise interactions and the excluded EVs, not just those identified in Stage 1. The inclusion of previously disqualified EVs allows the confirmation that they are indeed not necessary. The inclusion of pair-wise interaction parameters allows the capture of EV effects that are not behaving in an additive way. In other words, interactions allow an EV to have a different influence on survival over the different values provided by considering another EV.

[0049] The fifth step of a process according to the present invention is Model Checking where the proportional hazard assumption of the final model is tested. A global test is employed to see, if overall, the model violates this assumption. If it does, then proportional hazard assumption tests for individual EVs can suggest which of them violate the assumption. Any such variable needs to be declared as a strata variable in a stratified CPH model. Here separate baseline functions are fit for the levels of each violating EV. FIG. 4 illustrates the final model of survival curves for each combination of EVs Motor Series, Cable Bands and Cable Manufacturer for the test data set pertaining to the ESP system referenced in FIG. 1 while FIG. 5 illustrates the final model of FIG. 4 for the specific Motor Series “D/L/I”, Cable Manufacturer “C” and Cable Bands “0”, with 95% confidence interval (bounds) incorporated either side of the expected survival curve.

[0050] A method according to the present invention may also be utilized to apply survival analysis to any oilfield equipment components where time-to-event information has been recorded. The method of the present invention allows comparative reckoning between different components present in the system comprising several, or many, individual components, and allows analysis of these components either individually or simultaneously (in the presence of other components).

[0051] A method of performing a survival analysis on the components of an item of equipment comprises the step of representing the data on said components in a counting process formulation of a Cox model. In one embodiment, the i-th component may be represented by a set of observations: s_ij, t_ij, δ_ij, x_ij, k_i, j=1, . . . n_i, where (s_ij, t_ij] is an interval of risk, open on the left and closed on the right, δ_ij=0 if the component has failed at time t, and 1 if the component has not failed. x_ij is the explanatory variable vector over the interval and k, is the component type stratum variable. In the present method we will thus produce multiple observations of each component in an installation, and multiple observations of individual components across installations. In the data used in the above-described system survival analysis:

• • [0052] s_ij=daystart
  • [0053] t_ij=dayend
  • [0054] δ_ij,=eqstatus
  • [0055] k_i=comppart
  • [0056] x_ij=remaining explanatory variables.

[0057] With the method of the present invention, it is not necessary to have contiguous observations. Observations that are not contiguous may, for example, occur when considering a data set specific to a geographic region and a temporary component is utilized in an installation otherwise outside its valid operation range before returning to an installation within its normal operating region.

[0058] A method according to the present invention next comprises the step of applying an extension of the Cox Proportional Hazard model to the data, and as noted above, in one embodiment the extension that is applied is the Lunn extension. The Lunn extention is utilized because a key assumption in the CPH model is that observed survival or event times are independent. However, when dealing with component data, observations involving components in the same installation are naturally related and multiple observations of an individual component through its reuse are also clearly related.

[0059] A method according to the present invention next comprises the step of performing individual survival analysis on each of the identified components by using the five-stage method described above. The final step of a method according to the present invention is to establish the time-to-event estimate for each of the recorded components either taken individually or in the presence of the other components such that an estimate of competing risks is obtained.

[0060] Table VI below illustrates the results of applying the method of the present invention to data obtained with respect to ESPs. Section A of Table VI illustrates a model for system well failure. Section B of Table VI illustrates a model for component failure not considered individually but treated as a system well failure. Sections C-F of Table VI, illustrate time-to-failure estimates for the protector, motor, intake and pump, respectively, of the ESP, but performed individually. Section G of Table F illustrates analysis of the components in the presence of one another which provides a competing-risks estimate.

[0061]

	TABLE VI
	Hazards	95% Conf. Int.
Variable	Coef	se(coef)	z	p	Ratio	Lower	Upper
Model for All Components
PuMoPr.abrasionTRUE	2.14	0.397	6.62	3.5e−11	8.510	4.516	16.039	G
chkvalveTRUE	1.59	0.338	4.79	1.7e−06	4.923	2.563	9.456
PuInMo.cntrlizrTRUE	−1.53	0.382	−3.84	1.2e−04	0.216	0.099	0.473
PuMo.npull	1.28	0.269	5.28	1.3e−07	3.593	2.235	5.777
PuMoPr.panelmfgREDA	−1.85	0.274	−7.20	6.2e−13	0.157	0.095	0.260
Pr.scaleTRUE	−2.40	1.050	−2.25	2.4e−02	0.091	0.011	0.735
Model for Pump
abrasionTRUE	1.58	0.648	2.78	5.4e−03	4.833	1.593	14.669	F
cntrlizrTRUE	−1.09	0.545	−2.13	3.3e−02	0.335	0.122	0.916
mtcVOLTRON-113	1.27	0.461	3.02	2.5e−03	3.562	1.563	8.118
npull	1.48	0.386	4.07	4.7e−05	4.383	2.151	8.928
panelmfgUnknown	1.43	0.424	3.40	6.8e−04	4.175	1.831	9.521
Model for Intake/GS
chkvalveTRUE	1.96	0.632	3.25	1.1e−03	7.131	2.183	23.294	E
cntrlizrTRUE	−2.32	1.069	−2.28	2.3e−02	0.098	0.013	0.722
Model for Motor
abrasionTRUE	2.357	0.692	4.06	4.9e−05	10.554	3.384	32.915	D
npull	0.878	0.396	2.41	1.6e−02	2.406	1.178	4.913
panelmfgREDA	−1.901	0.443	−4.26	2.1e−05	0.149	0.062	0.359
Model for Protector
abrasionTRUE	2.83	0.719	4.70	2.7e−06	17.008	5.212	55.505	C
panelmfgREDA	−1.58	0.561	−3.03	2.5e−03	0.205	0.074	0.572
scaleTRUE	−2.12	1.044	−2.05	4.0e−02	0.119	0.016	0.912
Model for system: component failure
scaleTRUE	−5.500	2.025	−2.72	6.6e−03	0.004	0.000	0.216	B
mtcVOLTRON-113	1.901	0.461	4.12	3.7e−05	6.694	2.714	16.526
cntrlizrTRUE	−1.323	0.608	−2.18	2.9e−02	0.266	0.081	0.877
panelmfgREDA	−1.315	0.372	−3.53	4.1e−04	0.268	0.129	0.557
motnumb	0.319	0.389	0.82	4.1e−01	1.376	0.642	2.946
scaleTRUE:motnumb	2.464	0.968	2.55	1.1e−02	11.754	1.764	78.370
Model for system: well failure
cablemanufUnknown	Strata	—	—	—	—	—	—	A
corosion	Strata	—	—	—	—	—	—
fieldid	−1.071	0.343	−4.81	1.5e−06	0.343	0.222	0.530
wellevtno	−0.117	0.890	−2.05	4.0e−02	0.890	0.796	0.995
cblbands4	0.531	1.701	2.12	3.4e−02	1.701	1.040	2.783
wellmotorCL-562+	−1.970	0.139	−4.36	1.3e−05	0.139	0.058	0.338
WellmotorOthers	−1.422	0.241	−2.81	5.0e−03	0.241	0.089	0.651