Method and apparatus for producing a porosity log of a subsurface formation corrected for detector standoff

Abstract

A borehole logging tool is lowered into a borehole traversing a subsurface formation and a neutron detector measures the die-away of nuclear radiation in the formation. A model of the die-away of nuclear radiation is produced using exponential terms varying as the sum of borehole, formation and thermal neutron background components. Exponentially weighted moments of both the die-away measurements and die-away model are determined and equated. The equated moments are solved for the ratio of the borehole to formation amplitude components of the measurements. The formation decay constant is determined from at least the formation and thermal neutron background terms of the weighted measurement and model moments. An epithermal neutron lifetime is determined from the formation decay constant. This epithermal neutron lifetime and the amplitude ratio are used by a trained neural network to determine a lifetime correction and an apparent standoff. A standoff corrected lifetime is determined from the epithermal neutron lifetime and the lifetime correction. A porosity log of the formation is produced which is corrected for detector standoff from the borehole wall as a function of the standoff corrected epithermal neutron lifetime calibrated in borehole models of known porosities.

Description

BACKGROUND OF THE INVENTION

This invention relates to a borehole logging method for measuring the die-away of nuclear radiation of a subsurface formation and for correcting signals representative of the decay constants and amplitudes of the measured radiation for the effects of detector standoff from the borehole wall of the formation where such signals may vary with time, distance, or any other independent variable. A porosity versus depth log is then produced for the subsurface formation as a function of the standoff corrected formation decay constant derived from calibrated measurements in borehole models of known porosities and conditions of detector standoff.

A borehole logging tool is conventionally used to measure the die-away of nuclear radiation in the formations surrounding a borehole. In neutron porosity logging, for example, a neutron source is utilized for bombarding the formations with fast neutrons as the logging tool is advanced through the borehole. In the course of moderation, the fast neutrons reach the epithermal range and thence are further moderated until they reach the thermal neutron range. The populations of neutrons at the various energy levels die-away with time following primary irradiation and thus offer means of characterizing the formations. The rate of die-away of the epithermal neutron population gives a quantitative indication of the amount of hydrogenous material present in the formations which in turn is indicative of the porosities of the formations. Examples of both methods and apparatus for carrying out epithermal die-away porosity logging are set forth in U.S. Pat. No. 4,097,737 to W. R. Mills, Jr., and U.S. Pat. Nos. 4,556,793 and 4,590,370 to L. S. Allen and W. R. Mills, Jr.

An article entitled "Improved Methods of Signal Processing For Pulsed Neutron Capture Logging", SPWLA Twenty Second Annual Logging Symposium, Jun. 23-26, 1981 by R. Randall and E. C. Hopkinson discusses a method of pulsed neutron capture logging to differentiate oil, gas and saline formation water environments through casing. The method applies statistical averaging to a single exponential die-away term in the logged data to determine thermal neutron decay rate. In a still further reference, U.S. Pat. No. 4,600,838 to D. K. Steinman and L. A. Jacobson, there is described a method of thermal neutron die-away logging for overcoming excessive statistical fluctuations in the logged data, particularly in strongly absorbing formations. This method involves the determination of zero and first order moments of time during first and second discrete sequences of time gates respectively and the taking of the ratio of such moments to obtain a thermal neutron decay constant for the formations surrounding the logged borehole. In yet further references, U.S. Pat. No. 4,625,110 to Smith, Jr. and U.S. Pat. No. 4,638,161 to Smith, Jr. and Verbout, there is described a porosity determination utilizing a two-exponential model approach to epithermal neutron die-away. Finally, U.S. Pat. No. 4,972,082 to Loomis et al. discloses use of a modulated exponential function to provide correction of an epithermal neutron die-away measurement for the situation where the logging tool is not in good contact with the borehole wall.

While the foregoing described methods and systems have been utilized in the production of well logs for characterizing the die-away of nuclear radiation of subsurface formations, there is still need for a very sensitive method of improving on the characterizations provided by porosity logs when the neutron detector of the logging tool is not in contact with the formation (i.e., separation or standoff of the logging tool detector from the borehole wall). Hereinafter, such separations will be referred to as standoff, although they can arise from the logging tool pulling away from the surface of the borehole wall or from the borehole enlarging away from the logging tool. In such cases, the formation decay constant determined from the die-away of nuclear radiation measurements will be in error by the degree of effect such standoff has on the measurement. It is therefore an object of the present invention to provide for a porosity log generated from a standoff corrected formation decay constant such that the formation may be accurately characterized as to porosity.

SUMMARY OF THE INVENTION

The present invention is directed to a method and apparatus for measuring the die-away of nuclear radiation of a subsurface formation surrounding a borehole and for correcting such measurements for the effect of logging tool detector standoff from the borehole wall of the formation so that an accurate formation porosity log may be produced.

A logging tool is lowered into a borehole traversing a subsurface formation of interest and measurements are obtained of the die-away of nuclear radiation in the formation. Signals are generated representative of the intensity of the measurements. A model of such signals is generated having three exponential terms varying as the sum of the borehole, formation and thermal neutron background effects on such measurements. This is a complete representation of the principal components of the die-away process and provides a very sensitive parameter for correcting measured lifetimes for logging tool standoff. Three exponentially weighted signal moments and three exponentially weighted model moments are determined. These moments are equated and solved for the ratio of the borehole amplitude to the formation amplitude components of the signals. An initial formation decay constant and its reciprocal, epithermal neutron lifetime, is determined from a solution of the moment equations which involves at least the formation and thermal neutron background terms of the exponentially weighted signal and model moments. This formation epithermal neutron lifetime and amplitude ratio are applied to a trained neural network for determining a lifetime correction and apparent standoff. A standoff corrected epithermal neutron lifetime is derived as a function of the epithermal neutron lifetime and the epithermal neutron lifetime correction. This function is derived from logging measurements in borehole models at several porosities and conditions of logging tool separation from the formation. A porosity versus depth log is produced as a function of the standoff corrected epithermal neutron lifetime derived from calibration measurements in the borehole models at known porosities and conditions of detector standoff from the borehole wall.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 illustrates a borehole logging system for making die-away measurements of nuclear radiation from a subsurface formation surrounding a borehole and for producing a porosity versus depth log of such formation.

FIG. 2 illustrates a specific example of epithermal neutron die-away measurements as might be made by the borehole logging system of FIG. 1.

FIG. 3 illustrates the downhole and uphole electronics units of the borehole logging system of FIG. 1.

FIGS. 4 & 5 are flow charts representing the steps carried out in accordance with the method of the present invention for correcting the die-away of nuclear radiation measurements of FIG. 2 for the effect of detector standoff from the formation.

DESCRIPTION OF THE PREFERRED EMBODIMENT

Referring to FIG. 1, there is illustrated a borehole logging system useful in logging a subsurface formation for porosity in accordance with the present invention.

A borehole tool 10, supported by cable 11, comprises a high energy pulsed neutron source 12 and an articulated radiation detector pad 13. A high-voltage power supply 14 is provided for the source 12 and a downhole electronics unit 15 is provided with circuits for utilization and modification of signals from radiation detector pad 13. Also included are circuits for the control of the high voltage power supply 14. A backup arm 16 with attached pad 18 extending from the logging tool 10 may be included to force the logging tool 10 to follow the average contour of the borehole wall 24. Cable 11 extends to a surface electronics unit 17 where the signals from the radiation detector pad 13 are processed and recorded, along with an indication of depth, as a porosity log of the subsurface formation.

Particularly troublesome, however, to such a radiation measurement is the presence of rugosity in the borehole wall. Any borehole enlargement of depth greater than about one-eighth inch can seriously affect the formation radiation measurement.

To help eliminate the troublesome borehole rugosity effects on the formation radiation measurement, the radiation detector should be housed in a pad 13 which is small enough to track short borehole enlargements. Most such enlargements can be tracked by a radiation detector pad that is on the order of one foot in length. The radiation detector pad 13 includes at least one epithermal neutron detector 19 which is positioned against one side of pad 13, and a shielding material (i.e., a moderating material with or without an appropriate absorbing material) (not shown) which fills the remaining portion of the pad. Such a detector is illustrated in U.S. Pat. No. 4,581,532 to Givens et al. and in U.S. Pat. No. 4,692,617 to Allen et al., the teachings of which are incorporated herein by reference. The arm 22, shown in FIG. 1, azimuthally orients the pad so that the side of the pad against which the neutron detector is positioned is firmly pressed against the surface of the borehole wall. In this manner, the focusing of the directional sensitivity of the assembly consisting of the neutron detector and the shielding material on the formation is maximized, while the directional sensitivity of such assembly to radiation from the borehole fluid is minimized by the moderation and/or absorption of such borehole fluid radiation by the shielding material. Articulating arm 22 is shown in FIG. 1 for purposes only as one example of a means for positioning the radiation detector against the surface of the borehole wall as is illustrated in the aforementioned U.S. Pat. No. 4,692,617 to Allen et al. Other positioning means may also be acceptable such as by locating the radiation detector in the main body of the logging tool and pressing the tool or that portion of the tool housing the radiation detector against the surface of the borehole wall as shown in the aforementioned U.S. Pat. No. 4,581,532 to Givens et al. or in the aforementioned U.S. Pat. No. 4,972,082 to Loomis et al., the teaching of which is also incorporated herein by reference.

Nevertheless, there is still a need for improving on the characterization of the subsurface formation as to porosity in those situations where there is standoff of the neutron detector from the formation, whether due to the pad 13 not being completely in direct contact with the borehole wall or whether due to the entire tool 10 not being in complete contact with the borehole should the articulated pad 13 not be employed and the detector be alternatively housed in the main body of the tool 10. Such improvement, in accordance with a specific feature of the present invention, involves the correcting of the formation decay constant used in producing the porosity log for such detector standoff in accordance with a function relating formation decay constant to the ratio of borehole to formation components of die away amplitude, particularly where the measured die-away is comprised of a plurality of exponential terms, such as represented by the die-away curve illustrated in FIG. 2. For purposes of illustration, the curve in FIG. 2 is assumed to consist of the sum of two epithermal neutron die-away components and one thermal neutron die-away component. A first exponential die-away component A.sub.1 e.sup.-.mu. 1.sup.t dominant from 0 to t.sub.1 microseconds represents predominantly the rate of die-away of epithermal neutrons attributable to borehole effects. A second exponential die-away component A.sub.2 e.sup.-.mu. 2.sup.t dominant from t.sub.1 to t.sub.2 microseconds represents predominantly the rate of die-away of epithermal neutrons attributable to the formation. A third die-away component A.sub.3 e.sup.-.mu. 3.sup.t dominant from t.sub.2 to t.sub.f microseconds represents predominantly the rate of die-away attributable to thermal neutron effects. Each of these components can frequently be observed as a straight line when total intensity is plotted on a semilogarithmic scale during the time domain when the effect is dominant. For a more detailed description of such a plurality of die-away characteristics, as well as a description of a borehole logging apparatus 10, including surface equipment, for making the die-away data measurements, reference may be had to the aforementioned U.S. Pat. Nos. 4,097,737; 4,556,793; and 4,590,370, the teachings of which are incorporated herein by reference.

Referring now to FIG. 3 there is shown the downhole electronics 15 and surface electronics 17 useful for processing the die-away signals from the nuclear detector and producing the desired porosity log corrected for standoff. The die away signal from the neutron detector within pad 13 of FIG. 1 is applied through amplifier 25 and pulse height discriminator 26 to a time analyzer 28.

Time analyzer 28 stores the incoming die-away signal as a multichannel time sequence of counts representing the die-away spectrum. Preferably this is a sequence of contiguous channels each one microsecond in width, but other recording modes could be used. The die-away signal is accumulated over a time frame of approximately one second and corresponds to approximately 5000 firings of the neutron source. The spectrum thus accumulated is then passed to uphole electronics 17 for processing.

The uphole electronics 17 is controlled by a master digital computer (Hewlett Packard Model 1000 for example). A first signal processor (Stage I) 30 solves equations which relate time moments calculated for the accumulated data spectrum with time moments calculated for the mathematical model of the die-away process. This solution produces the parameters A.sub.i and .mu..sub.i (i=1,2,3, previously described.

The parameters A.sub.i and .mu..sub.i are then passed to a second signal processor (Stage II) 31 where an epithermal neutron lifetime .tau..sub.2 is determined from the formation die-away parameter .mu..sub.2 and is used along with the ratio R of the amplitude of the borehole die-away component A.sub.1 to the amplitude of its formation die-away component A.sub.2, as described above, by a neural network to determine an epithermal neutron lifetime correction .DELTA..tau..sub.2 and apparent standoff S. Thereafter a standoff corrected epithermal neutron lifetime .tau..sub.2 * is produced by third signal processor (Stage III) 32 which utilizes calibration measurements taken in borehole models at known porosities and conditions of detector standoff to produce the desired recording, or log, of porosity versus depth. Apparent standoff can also be simultaneously recorded as a function of depth.

A detailed explanation of the operation of the uphole electronic signal processing will now be made in conjunction with the flow charts of FIGS. 4 and 5. The characteristic parameter requiring correction is the lifetime (or decay constant) of the formation and the characteristic parameter which can be used for correction is the ratio of the amplitude of the borehole component to the amplitude of the formation component of the die-away measurement. This amplitude ratio can be used to correct the measured lifetime for situations where the detector is not in direct contact with the borehole wall because the effect of the standoff correlates strongly with the amplitude ratio when the time dependence of the borehole measurement is modeled as the sum of three exponential functions: a borehole component, a formation component and a thermal neutron "background" component, as shown in steps 410-480 of FIG. 4. The measured epithermal neutron signal f(t) can be expressed as:

f(t)=A.sub.1 e.sup.-.mu. 1.sup.t +A.sub.2 e.sup.-.mu. 2.sup.t +A.sub.3 e.sup.-.mu. 3.sup.t (1) where the capital letters A.sub.1 -A.sub.3 denote amplitudes and the Greek lower case letters .mu..sub.1 -.mu..sub.3 denote decay constants of the three components. In this representation the letters A.sub.1 and .mu..sub.1 apply to the borehole component, the letters A.sub.2 and .mu..sub.2 apply to the formation component, and the letters A.sub.3 and .mu..sub.3 apply to the thermal neutron background component.

If the radiation detector is strongly focused on the formation by appropriate shielding and by tool design and positioning, then decay constant .mu..sub.1 can be treated as a constant parameter determined by calibration of the measurement in suitable borehole models of known physical characteristics. Decay constant .mu..sub.3, although varying with formation and borehole properties, can be assumed a constant average value because of the small effect of the thermal neutron background component on the epithermal neutron measurement (A.sub.3 <<A.sub.1 or A.sub.2). This leaves the decay constant .mu..sub.2 and the amplitudes A.sub.1, A.sub.2 and A.sub.3 to be determined from the borehole measurement (steps 410-420 ). Three parameters can be determined from the calculation of three weighted moments (step 430) of the epithermal neutron flux; these are A.sub.1 /A.sub.2, A.sub.3 /A.sub.2 and .mu..sub.2. It is not necessary to solve for the absolute magnitudes of A.sub.1, A.sub.2 or A.sub.3.

If exponential weighting is used to suppress the effect of later data channels with increasingly large statistical uncertainties, the expected values of the three model moments of time are: ##EQU1## and the sums are over the appropriate time intervals where the exponential model is valid. Reasonable choices for the z.sub.j range from about 0.01 inverse microsecond to about 0.5 inverse microsecond. Three intensity moments are expressed: ##EQU2##

where N.sub.i is the intensity of the recorded data.

The three exponentially weighted model and intensity moments are equated and the resulting equations solved for A.sub.1 /A.sub.2, A.sub.3 /A.sub.2 and .mu..sub.2.

The exponentially varying intensity data are measured as a continuous function of time or in discrete time intervals. Firstly, with respect to the continuous case, the intensity data are modeled as follows: ##EQU3## where .mu..sub.k ' are the decay constants for the continuous case and A.sub.k ' are the amplitudes. The expected value of weighted moments of time with exponential weighting are determined from the model as follows: ##EQU4## each z.sub.j.

The weighted moments of data are then computed as: ##EQU5## where y(t) represents the continuous data.

Next the corresponding model and intensity moments are equated:

E[t.sub.j (t.sub.a, t.sub.b)]=t.sub.j (t.sub.a,t.sub.b) (8) for all three weighting functions yielding three transcendental equations. The three equations are solved simultaneously for estimates of the parameters A.sub.1 '/A.sub.2 ', A.sub.3 '/A.sub.2 ' and the formation component decay constant .mu..sub.2 ' with the borehole component decay constant .mu..sub.1 ' and the thermal neutron component decay constant .mu..sub.3 ' being known (step 440). It is not necessary to determine A.sub.2 '. The starting time t.sub.a and ending time t.sub.b are illustrated in FIG. 2. For the continuous case A.sub.k ' has units of counts per unit interval and .mu..sub.k ' has units of inverse unit interval.

Secondly, with respect to the discrete case, the intensity data are modeled as follows: ##EQU6## where n=1, . . . , N. The decay constants are .mu..sub.k " for the discrete case and the amplitudes are A.sub.k ". The expected value of weighted moments of time interval with exponential weighting are determined from the model as follows: ##EQU7## for each z.sub.j ' (different from z.sub.j only by conversion to discrete time channels). The weighted moments of intensity are then computed as follows: ##EQU8## where Y(n) represents the discrete data. Next the corresponding model and intensity moments are equated:

E[n.sub.j (n.sub.a,n.sub.b)]=n.sub.j (n.sub.a,n.sub.b) (12) yielding three transcendental equations which are solved simultaneously for estimates of the three parameters A.sub.1 "/A.sub.2 ", A.sub.3 "/A.sub.2 " and the formation component decay constant .mu..sub.2 " with the borehole component decay constant .mu..sub.1 " and the thermal neutron component decay constant .mu..sub.3 " being known. For the discrete case A.sub.k " has units of counts per unit time and .mu..sub.k " has units of inverse unit time.

Reasonable choices for the z.sub.j ' are 1/64, 1/32 and 1/16 when the time channels are each 1 .mu.s in duration.

The decay constant .mu..sub.2 is now converted at step 445 to epithermal neutron lifetime .tau..sub.2 :

.tau..sub.2 =1/.mu..sub.2. (13)

A decision is then made on the need for employing a double exponential model. The decision can be based on the decay constant .mu..sub.2 (i.e., .mu..sub.2 ' for the continuous case or .mu..sub.2 " for the discrete case), the amplitude ratio A.sub.1 /A.sub.2 (hereinafter termed R) or some combination of both. Illustrated in FIG. 4 is the use solely of R as the decision parameter (step 450).

At step 455 the values of epithermal neutron lifetime .tau..sub.2 and the amplitude ratio R are applied to a trained neural network for the determination of a lifetime correction at .DELTA..tau..sub.2 and apparent standoff S. The neural network is trained using many sets of .DELTA..sub.2, R, .DELTA..tau..sub.2 and S which have been obtained from measurements with the pulsed neutron porosity logging tool in borehole models under a wide variety of known conditions.

Having obtained .DELTA..tau..sub.2 and S from the neural network, a corrected epithermal neutron lifetime .tau..sub.2 * is determined from .tau..sub.2 and .DELTA..tau..sub.2 at step 460:

.tau..sub.2 *=f.sub.1 (.tau..sub.2,.DELTA..tau..sub.2), (14) or .tau..sub.2 *=.tau..sub.2 +.DELTA..tau..sub.2 (15)

Having obtained a corrected value for .tau..sub.2, conversion to a porosity P versus depth log is carried out at step 470:

P=f.sub.2 (.tau..sub.2 *) (16) A useful form is: P=a+bx+cx.sup.2 +dx.sup.3 (17) where x again stands for .mu..sub.2 ' or .mu..sub.2 " and a-d are empirical constants. The empirical constants a-d are obtained from logging tool measurements made in borehole models under known conditions.

Clearly, to obtain an accurate empirical equation (17), it is necessary to have measurements made in borehole models at several porosities for a variety of standoff conditions. A practical minimum is probably four porosities and four different conditions of detector-formation separation. Many more combinations would be helpful. The experimental measurements can obviously be augmented by accurate theoretical calculations, and these should be used when practical. It should be apparent to those skilled in the art that such empirical functions can have many acceptable forms other than those set forth above in equations (14), (15), (16) and (17). The best form is usually selected on some goodness of fit criterion.

It should be noted that the Stage I process of solving the three transcendental equations for A.sub.i and .mu..sub.i can also be carried out by a trained neural network. This approach has the advantage of dramatically speeding up the determination of A.sub.i and .mu..sub.i, but the disadvantage of introducing small residual training errors into the determination of these parameters. The training errors are quite acceptable and are of the order of one to two percent. Thus neutral networks can be used to handle the entire process: A first neural network produces physical parameters from the data which are exceedingly sensitive to porosity and standoff. This network is used because it is exceedingly fast computationally. A second neural network combines two parameters from the first stage of analysis to obtain a lifetime correction and apparent standoff. This network is used because the kernel driving the transformation is analytically very difficult to describe. The lifetime correction is appropriately combined with the measured lifetime and converted to porosity by a simple analytical transformation.

The foregoing transformations by two neural networks can clearly be combined into a single network. However, overall accuracy of the training exercise appears to be better in the two-stage approach because the physical model of the die-away process extracts from the input data two parameters which strongly characterize porosity (.mu..sub.2) and apparent standoff (R).

One additional step, alluded to earlier, should be taken to enhance results. When standoff is large or when porosity is high the number of counts recorded in a time analyzer is reduced and the three-exponential model for determining the initial formation decay constant .mu..sub.2 will produce results having larger statistical fluctuations than desired. In this case R, for example, is greater than a threshold level and the analysis for the decay constant .sub.2 is switched at step 450 to a two-exponential model of a second formation decay constant .mu..sub.2, as shown in steps 500-530 of FIG. 5, representing only the formation and the thermal neutron background. The general equation for f(t) symbolically representing both the continuous and discrete cases is then ##EQU9##

The decay constant .mu..sub.2 is converted at step 505 to epithermal neutron lifetime .tau..sub.2 :

.tau..sub.2 =1/.mu..sub.2 (20)

At step 507 the values of epithermal neutron lifetime .tau..sub.2 and the amplitude ratio R are applied to a trained neural network for the determination of a lifetime correction .DELTA..tau..sub.2 and apparent standoff S. The neural network is trained using many sets of .tau..sub.2, R, .DELTA..tau..sub.2 and S which have been obtained from measurements with the pulsed neutron porosity logging tool in borehole models under a wide variety of known conditions.

Having obtained .DELTA..tau..sub.2 and S from the neural network, a corrected epithermal neutron lifetime .tau..sub.2 * is determined from .tau..sub.2 and .DELTA..tau..sub.2 at step 508:

.tau..sub.2 *=f.sub.3 (.tau..sub.2, .DELTA..tau..sub.2), (21) .tau..sub.2 *=.tau..sub.2 +.DELTA..tau..sub.2. (22)

Conversion to the porosity versus depth log is then carried out at step 520:

P=f.sub.4 (.tau..sub.2 *) (23)

This use of a two-exponential model may require that the analysis be started a few channels later in time, but it still produces superior results because the three-exponential model has become so dominated by the borehole component that measurement sensitivity to the desired formation component is reduced. The switch from three exponentials to two exponentials should be based on relative accuracy considerations. Note that the amplitude ratio R from the three-exponential model is always used to determine the effect of detector-formation separation, even when the two-exponential model is used to determine .tau..sub.2. It will normally be desirable to display on the log the magnitude of the porosity correction (in porosity units) as well as the corrected porosity and apparent standoff.

It should be clear to those skilled in the art that the data moment calculations used in the three-exponential model can be used in the two-exponential model without recalculation and that the well-known equation for the sum of a finite geometric series should be used to simplify certain sums in the model representations.

It should be clear to those skilled in the art that the training of the neural network and the resulting determinations used to produce values of logging tool standoff S and an epithermal neutron lifetime correction factor .DELTA..tau..sub.2 from the epithermal neutron lifetime .tau..sub.2 and amplitude ratio R are well known and need not be reproduced herein. They are readily found, for example, in Chapter 8 of the text Parallel Distributed Processing, Volume 1, by D. Rumelhart and J. McClelland, MIT Press. Briefly, however, a neural network consists of a collection of processing elements called neurons arranged in layers. A neuron has multiple inputs, a summing device, and a non-linear thresholding operation on the summed output. Each neuron in a layer is connected with every neuron in an adjacent layer by weight. The first layer is called the input layer and the last layer is called the output layer. Layers of neurons in between are called hidden layers. The network is trained by supplying a training set of data comprising a particular input and its associated desired output pattern to the network. The input is applied to the network and an output computed. An error is found by subtracting this output from the ideal output and the weights are then adjusted in the network through a process known as back-propagation to make this error smaller. In operation, the input is fed through the network iteratively and each time the weights are adjusted to make the error smaller until the error reaches some threshold level. Additional data from the training set is fed through the network in the same way and the errors reduced each time. Finally, the network learns to associate a particular input pattern with a particular output pattern.

Neural networks have been used in pattern and voice recognition and in digital signal processing. In using a neural network in the pulsed neutron porosity logging operation of the present invention, the network is trained using back-propagation to take input values of epithermal neutron lifetime .tau..sub.2 and amplitude ratio R and provide output values of logging tool standoff S and a correction factor .DELTA..tau..sub.2 which can be added to .tau..sub.2 to get standoff-corrected values .tau..sub.2 *. Because this calculation of standoff and standoff-corrected lifetime .tau..sub.2 * is very fast using a neural network and the conversion of corrected lifetime to porosity involves only a single function, an accurate determination of both porosity and standoff can be made easily in real time as the well log is being recorded.

One example of training data used for a three exponential model of an epithermal neutron die-away process is illustrated below.

Table I illustrates example training data for the neural network; Table II illustrates the network parameters; and Table III illustrates output data from the trained neural network for the example. In Table I the training data are as follows: Column 1 lists porosity of the calibration models, Column 2 lists measured lifetimes at zero standoff, Column 3 lists the amplitude ratio R (here identified as Alpha), Column 4 lists the measured lifetimes at the standoff conditions given in Column 6, Column 5 lists the lifetime correction term. Columns 3 and 4 are thus inputs of the training data and Columns 5 and 6 are the desired outputs. Versions of Columns 3 through 6 suitably scaled for network training appear in Columns 7 through 10. Table II is self-explanatory. Table III lists the output obtained from the trained neural network when the calibration data used in its training is the input: Column 5 lists the lifetime correction term and Column 6 the standoff-corrected lifetime. The standoff-corrected lifetime, Column 6, is compared to the lifetime measured at zero standoff, Column 7, by computing their ratio in Column 8. Standoff determined by the neural network is shown in Column 10 and can be compared with the figures appearing in Column 2.

Having now described the present invention in conjunction with a preferred embodiment, it is to be understood that various modifications and alterations may be apparent from the spirit and scope of the invention as set forth in the appended claims. For example, if amplitude A.sub.3 were made to be very small (i.e., A.sub.3 .apprxeq.0) through use of a thick thermal neutron shield, then only two weighted moments are needed to solve for decay constant .mu..sub.2 and amplitude ratio A.sub.1 /A.sub.2 provided decay constant .mu..sub.1 is fixed. If .mu..sub.1 is not fixed, then three weighted moments must be determined to obtain .mu..sub.1, .mu..sub.2 and A.sub.1 /A.sub.2.

Further, a modulated exponential form may be used to represent the data. One of the simplest forms is: ##EQU10## where the .alpha..sub.i are constants. The .alpha..sub.i will normally be determined from measurements in borehole models. This avoids the calculation of additional moments in the real-time data analysis algorithm. Such modulated exponential forms may fit the somewhat scattered data that must normally be worked with satisfactorily. However, they generally result in expressions involving the gamma function and are thus not as convenient to use as expressions involving the standard exponential function.

Still further, the calibrated output could be hydrogen index instead of porosity. The lifetime measurement is most generally calibrated in terms of hydrogen concentration and could be presented this way. It is assumed that the pore space is completely saturated with fresh water in order to present a porosity curve. False (i.e., low) porosity values are recorded in low-pressure gas zones and in partially-saturated zones which are above the water table.

TABLE I__________________________________________________________________________ STAND- STAND-POROSITY TAU3E(0/4) ALPHA TAU3E DELTAU3E OFF ALPHA* TAU3E* DELTAU3E* OFF*__________________________________________________________________________1.8 18.4932 0.8822 18.4932 0 0 -0.17356 0.37466 -0.3 -0.2518.7 6.8584 0.3877 6.8584 0 0 -0.27246 -0.20708 -0.3 -0.2526.5 5.8255 0.1869 5.8255 0 0 -0.31262 -0.25873 -0.3 -0.2540.3 4.3838 0 4.3838 0 0 -0.35 -0.33081 -0.3 -0.251.8 18.4932 1.226 14.7188 3.7744 0.25 -0.1048 0.18594 -0.048373 -0.12518.7 6.8584 0.6791 6.0352 0.8232 0.25 -0.21418 -0.24824 -0.24512 -0.12526.5 5.8255 0.4606 5.2539 0.5716 0.25 -0.25788 -0.28731 -0.261893 -0.12540.3 4.3838 0.177 4.1368 0.247 0.25 -0.3146 -0.34316 -0.283533 -0.1251.8 18.4932 1.6586 12.2188 6.2744 0.5 -0.01828 0.06094 0.118293 018.7 6.8584 1.1946 5.584 1.2744 0.5 -0.11108 -0.2708 -0.21504 026.5 5.8255 0.8063 4.8438 0.9817 0.5 -0.18874 -0.30781 -0.234553 040.3 4.3838 0.7232 4.1875 0.1963 0.5 - 0.20536 -0.34063 -0.286913 01.8 18.4932 2.615 9.1973 9.2959 1 0.173 -0.09014 0.319727 0.2518.7 6.8584 2.8451 5.5078 1.3506 1 0.21902 -0.27461 -0.20996 0.2526.5 5.8255 2.8789 5.2188 0.6067 1 0.22578 -0.28906 -0.259553 0.2540.3 4.3838 3.3038 4.8867 -0.5029 1 0.31076 -0.30567 -0.333527 0.25__________________________________________________________________________ TABLE II__________________________________________________________________________Network name: at.sub.-- dts8.netNework layers: 4Layer dimensions:layer 1: 2layer 2: 6layer 3: 4layer 4: 2Cycles trained: 389328Weights initialized from -0.300000 to 0.300000Noise: 0.000000. Decay: 0.000000Learning rates:layer 2: 2.000000layer 3: 2.000000layer 4: 2.000000Momentum terms:layer 2: 0.800000layer 3: 0.800000layer 4: 0.800000Layer 2 biases:-2.517196 -0.414389 -2.684155 1.372130 -2.945600 3.109718Layer 3 biases:-4.794534 0.418255 -1.748318 0.690356Layer 4 biases:-4.596799 -1.594935Layer 1 to layer 2 weights: 0.916544 -6.208319 6.427018 -4.098271 10.101479 -5.903040 8.857375 -7.860291 -2.030396 21.321926 -1.369920 4.651538Layer 2 to layer 3 weights: 5.031387 -1.240643 4.144804 3.924167 2.057790 -0.975195-7.875674 3.657824 0.026516 -6.422636 6.654104 4.918244 3.928713 -2.622469 2.040396 6.355948 -2.524712 -3.536021 6.635888 3.497980 -1.244142 -2.684376 0.042426 0.273035Layer 3 to layer 4 weights: 11.235277 4.946550 -2.434103 -1.889393 4.572870 2.783114 6.063582 -6.635078__________________________________________________________________________ TABLE III__________________________________________________________________________ STAND- TAU3E TAU3E STAND- STAND-POROSITY OFF TAU3E DELTAU* DELTAU (COR) (0/4) RATIO OFF* OFF__________________________________________________________________________1.8 0 18.4932 -0.30026 -0.003945 18.489255 18.4932 0.99978668 -0.249742 0.00051618.7 0 6.8584 -0.2984 0.024015 6.882415 6.8584 1.00350155 -0.245668 0.00866426.5 0 5.8255 -0.30174 -0.026055 5.799445 5.8255 0.99552742 -0.261719 -0.02343840.3 0 4.3838 -0.29535 0.06981 4.45361 4.3838 1.01592454 -0.238218 0.0235641.8 0.25 14.7188 -0.04813 3.77799 18.49679 18.4932 1.00019413 -0.125226 0.24954818.7 0.25 6.0352 -0.24572 0.81417 6.84937 6.8584 0.99868337 -0.117879 0.26424226.5 0.25 5.2539 -0.26548 0.517785 5.771685 5.8255 0.99076217 -0.129312 0.24137640.3 0.25 4.1368 -0.28618 0.20736 4.34416 4.3838 0.99095762 -0.132637 0.2347261.8 0.5 12.2188 0.118076 6.27114 18.48994 18.4932 0.99982372 -0.000026 0.49994818.7 0.5 5.584 -0.21167 1.324935 6.908935 6.8584 1.00736834 -0.002845 0.4943126.5 0.5 4.8438 -0.2357 0.964515 5.808315 5.8255 0.99705004 0.0002 0.500440.3 0.5 4.1875 -0.28679 0.19815 4.38565 4.3838 1.00042201 0.003209 0.5064181.8 1 9.1973 0.319218 9.28827 18.48557 18.4932 0.99958742 0.249814 0.99962818.7 1 5.5078 -0.21014 1.34784 6.85564 6.8584 0.99959757 0.254361 1.00872226.5 1 5.2188 -0.25852 0.62214 5.84094 5.8255 1.00265042 0.245244 0.99048840.3 1 4.8867 -0.3341 -0.511485 4.375215 4.3838 0.99804165 0.2506699 1.001338__________________________________________________________________________

Claims

1. Apparatus for converting die-away signals representing the die-away of nuclear radiation in a subsurface formation surrounding a borehole into a log representing porosity versus depth within said formation corrected for the effect of detector standoff from the borehole wall, comprising:

a) a borehole logging tool,

b) means for traversing said logging tool along said borehole,

c) a pulsed source of fast neutrons within said logging tool for irradiating the formation surrounding a borehole with fast neutrons,

d) at least one directionally sensitive radiation detector assembly having a neutron detector for measuring the die-away of secondary radiation produced by the fast neutrons which return to the borehole from the irradiated formation,

e) means for positioning said radiation detector assembly so that said radiation detector assembly contacts the surface of the borehole wall and is oriented to position the neutron detector to maximize directional sensitivity of the neutron detector to the die-away of secondary radiation from the formation at the point of contact of the radiation detector assembly with the borehole wall and to minimize directional sensitivity to the die-away of secondary radiation from the borehole fluid,

f) means for producing intensity signals representative of the die-away of secondary radiation measured by said detector as said logging tool traverses said borehole,

g) means for producing an initial formation decay constant from at least formation and background components of said intensity signals,

h) means for producing a ratio of the borehole amplitude to formation amplitude components of said intensity signals,

i) means for determining an epithermal neutron lifetime from said initial formation decay constant,

j) a trained neural network to determine an epithermal neutron lifetime correction and apparent standoff from said epithermal neutron lifetime and said amplitude ratio,

k) means for producing a standoff corrected epithermal neutron lifetime in accordance with a function of said epithermal neutron lifetime and said epithermal neutron lifetime correction to correct for any standoff from the borehole wall encountered by said detector as said logging tool traverses said borehole, and

l) means for generating a log of porosity versus depth within said formation corrected for detector standoff from the borehole wall as a function of said standoff corrected epithermal neutron lifetime derived from measurements in borehole models at known porosities and conditions of detector standoff from the borehole wall.

2. The apparatus of claim 1 wherein said porosity log is standoff-corrected for any borehole enlargement that occurs in juxtaposition with said radiation detector assembly as said logging tool traverses along said borehole.

3. The apparatus of claim 1 wherein said porosity log is standoff-corrected for pulling away of the radiation detector assembly from contact with the surface of the borehole wall as said logging tool traverses along said borehole.

4. A method for converting signals representing the die-away of nuclear radiation in a subsurface formation surrounding a borehole into a log representing porosity versus depth within said subsurface formation corrected for the effect of detector standoff from the borehole wall, comprising the steps of:

a) lowering a logging tool having a source of fast neutrons and a spaced-apart neutron detector into a borehole traversing a subsurface formation whose porosity is to be determined,

b) irradiating said subsurface formation at an initial depth position within said borehole with fast neutrons from said neutron source,

c) recording intensity signals representing the die-away of nuclear radiation with said spaced-apart neutron detector after said fast neutrons from said neutron source have passed through and been moderated by said subsurface formation,

d) generating a model of the die-away of nuclear radiation in a subsurface formation having amplitudes and decay constants varying exponentially in response to borehole, formation and background effects on said intensity signals,

e) producing weighted moments of said model and of said intensity signals,

f) producing a ratio of the borehole amplitude to formation amplitude components of said intensity signals derived from said weighted moments,

g) producing an initial formation decay constant signal from at least formation and thermal neutron background terms of said model and said intensity signals,

h) producing an epithermal neutron lifetime signal of said formation,

i) producing an epithermal neutron lifetime correction signal and apparent standoff value S from said epithermal neutron lifetime signal and borehole-to-formation amplitude ratio R using a trained neural network,

j) producing a standoff corrected epithermal neutron lifetime signal in accordance with a function of said epithermal neutron lifetime signal and said epithermal neutron lifetime correction derived from die-away signals recorded in borehole models at known porosities and conditions of detector standoff from the borehole wall,

k) traversing said logging tool along said borehole from said initial depth position,

l) repeating steps b) to j) as said logging tool traverses along said borehole, and

m) generating a log of porosity versus depth within said subsurface formation corrected for detector standoff from the borehole wall as a function of said standoff corrected epithermal neutron lifetime signal derived from calibrated measurements in borehole models at known porosities and conditions of detector standoff from the borehole wall.

5. The method of claim 4 wherein said model of the die-away of nuclear radiation is produced from a sum of exponential terms as follows:

f(t)=A.sub.1 e.sup.-.mu. 1.sup.t +A.sub.2 e.sup.-.mu. 2.sup.t +A.sub.3 e.sup.-.mu. 3.sup.t

where

A.sub.1 =borehole amplitude

A.sub.2 =formation amplitude

A.sub.3 =background amplitude,

.mu..sub.1 =borehole decay constant,

.mu..sub.2 =formation decay constant,

.mu..sub.3 =background decay constant, and

t=time.

6. The method of claim 4 wherein:

a) said weighted moments of said model are produced as follows:

E[t.sub.j ]=.SIGMA.t.sub.i w.sub.ji f(t.sub.i)/.SIGMA.w.sub.ji f(t.sub.i),

where

w.sub.ji =e.sup.-z j.sup.t i,

z.sub.j =weighting function decay constant

j=1,2,3, and

b) said weighted moments of said intensity signals are produced as follows:

t.sub.j =.SIGMA..sub.i t.sub.i w.sub.ji N.sub.i /.SIGMA..sub.i w.sub.ji N.sub.i,

where

N.sub.i =intensity of the recorded data.

7. A method for converting signals representing the die-away of nuclear radiation in a subsurface formation surrounding a borehole into a log representing porosity versus depth within said subsurface formation corrected for the effect of detector standoff from the borehole wall, comprising the steps of:

a) lowering a logging tool having a source of fast neutrons and a spaced-apart neutron detector into a borehole traversing a subsurface formation whose porosity is to be determined,

b) irradiating said subsurface formation at an initial depth position within said borehole with fast neutrons from said neutron source,

c) recording intensity signals representing the die-away of nuclear radiation with said spaced-apart neutron detector after said fast neutrons from said neutron source have passed through and been moderated by said subsurface formation.

d) generating a model of the die-away of nuclear radiation in said subsurface formation having three exponential terms varying in response to borehole, formation and thermal neutron background effects on said intensity signals as follows: ##EQU11## where .mu..sub. = borehole decay constant,

.mu..sub.2 =formation decay constant,

.mu..sub.3 =thermal neutron background decay constant

A.sub.1 =borehole amplitude,

A.sub.2 =formation amplitude,

A.sub.3 =thermal neutron background amplitude,

t=time,

e) determining the expected value of weighted model moments with exponential weighting in accordance with the three-exponential model of step d) as follows: ##EQU12## where z.sub.j =weighting function decay constant

t.sub.a =starting time,

t.sub.b =ending time,

f) determining weighted moments of said intensity signals with exponential weighting in accordance with the three-exponential model of step c) as follows: ##EQU13## where y(t)=intensity

g) equating said weighted model moments with said weighted intensity moments;

E[t.sub.j (t.sub.a,t.sub.b)]=t.sub.j (t.sub.a,t.sub.b)

h) utilizing said equated weighted model and intensity signal moments from step g) to determine a ratio signal R of the borehole amplitude A.sub.1 to the formation amplitude A.sub.2 components of said intensity signal:

R=A.sub.1 /A.sub.2,

i) determining an initial formation decay constant .mu..sub.2 utilizing at least two-exponential terms of said equated weighted model and intensity signal moments from step g) attributable to the formation and thermal neutron background effects,

j) determining an epithermal neutron lifetime .tau.:

.tau.=1/.mu..sub.2,

k) determining an epithermal neutron lifetime correction .DELTA..tau..sub.2 and apparent standoff S from said epithermal neutron lifetime .tau..sub.2 and amplitude ratio R using a trained neural network,

l) determining a standoff corrected epithermal neutron lifetime .tau..sub.2 * for detector standoff from the borehole wall in accordance with a function f(.tau..sub.2,.DELTA..tau..sub.2), derived from calibration measurements taken in borehole models at known porosities and conditions of detector standoff from the formation:

.tau..sub.2 *=f(.tau..sub.2, .DELTA..tau..sub.2),

and

m) traversing said logging tool along said borehole from said initial depth position,

n) repeating steps b) to l) as said logging tool traverses along said borehole, and

o) producing a porosity versus depth log of said subsurface formation corrected for detector standoff from the borehole wall as a function of said standoff corrected epithermal neutron lifetime .tau..sub.2 * derived from calibrated measurements in said borehole models at said known porosities and conditions of detector standoff from the borehole wall.

8. The method of claim 7 wherein step i) of determining the initial formation decay constant .mu..sub.2 is carried out utilizing all three exponential terms of said equated weighted model and intensity moments attributable to borehole, formation and thermal neutron background effects.

9. The method of claim 7 further comprising the steps of:

p) comparing said ratio signal R determined in step h) to a threshold level, and

q) utilizing in said neural network the initial formation decay constant .mu..sub.2 determined in step i) for all three of said exponential terms attributable to borehole, formation and thermal neutron background effects if R is less than or equal to said threshold level and repeating step i) to determine a second formation decay constant .mu..sub.2 from only the two of said exponential terms attributable to formation and thermal neutron background effects if R is greater than said threshold level.

10. The method of claim 7 further comprising the steps of:

p) comparing said initial formation decay constant .mu..sub.2 determined in step i) for all three of said exponential terms to a threshold level, and

q) utilizing in said neural network said initial decay constant for all three of said exponential terms if said initial decay constant is less than or equal to said threshold level, and repeating step i) to determine a second formation decay constant .mu..sub.2 for use in said neural network from only the two of said exponential terms attributable to formation and thermal neutron background effects if said initial formation decay constant .mu..sub.2 is greater than said threshold level.

11. A method for converting signals representing the die-away of nuclear radiation in a subsurface formation surrounding a borehole into a log representing porosity versus depth within said subsurface formation corrected for the effect of detector standoff from the borehole wall, comprising the steps of:

a) lowering a logging tool having a source of fast neutrons and a spaced-apart neutron detector into a borehole traversing a subsurface formation whose porosity is to be determined,

b) irradiating said subsurface formation at an initial depth position within said borehole with fast neutrons,

c) recording intensity signals representing the die-away of nuclear radiation with said spaced-apart neutron detector after said fast neutrons from said neutron source have passed through and been moderated by said subsurface formation,

d) generating a model of the die-away of nuclear radiation in said subsurface formation having three exponential terms varying in response to borehole, formation and thermal neutron background effects on said intensity signals as follows: ##EQU14## where .mu..sub. = borehole decay constant,

.mu..sub.2 =formation decay constant,

.mu..sub.3 =thermal neutron background decay constant

A.sub.1 =borehole amplitude,

A.sub.2 =formation amplitude,

A.sub.3 =thermal neutron background amplitude, and

n=1, 2, , . . . , N (discrete intervals of time)

e) determining the expected value of weighted model moments with exponential weighting in accordance with the three-exponential model of step d) as follows: ##EQU15## where z.sub.j =weighting function decay constant

n.sub.a =discrete interval starting time,

n.sub.b =discrete interval ending time,

f) determining weighted moments of said intensity signals with exponential weighting in accordance with the three-exponential model of step c) as follows: ##EQU16## where Y(n) represents discrete data, g) equating said weighted model moments with said weighted intensity moments:

E[n.sub.j (n.sub.a,n.sub.b)]=n.sub.j (n.sub.a,n.sub.b),

h) utilizing said equated weighted model and intensity moments from step g) to determine a ratio signal R of the borehole amplitude A.sub.1 to the formation amplitude A.sub.2 components of said intensity signal:

R=A.sub.1 /A.sub.2,

i) determining an initial formation decay constant .mu..sub.2 utilizing at least two-exponential terms of said equated weighted model and intensity signal moments from step g) attributable to the formation and thermal neutron background effects, and

j) determining an epithermal neutron lifetime .tau.:

.tau.=1/.mu..sub.2,

k)determining an epithermal neutron lifetime correction .DELTA..tau..sub.2, and apparent standoff S from said epithermal neutron lifetime .tau..sub.2 and borehole-to-formation amplitude ratio R using a neural network,

l) determining a standoff corrected epithermal neutron lifetime .tau..sub.2 * for detector standoff from the borehole wall in accordance with a function f(.tau..sub.2,.DELTA..tau..sub.2), derived from calibration measurements taken in borehole models at known porosities and conditions of detector standoff from the formation:

.tau..sub.2 *=f.sub.1 (.tau..sub.2,.DELTA..tau..sub.2),

and

m) traversing said logging tool along said borehole from said initial depth position,

n) repeating steps b) to l) as said logging tool traverses along said borehole, and

o) producing a porosity versus depth log of said subsurface formation corrected for detector standoff from the borehole wall as a function of said standoff corrected epithermal neutron lifetime .tau..sub.2 * derived from calibrated measurements in said borehole models at said known porosities and conditions of detector standoff from the borehole wall.

12. The method of claim 11 wherein step i) of determining the initial formation decay constant .mu..sub.2 is carried out utilizing all three exponential terms of said equated weighted model and intensity signal moments attributable to borehole, formation and thermal neutron background effects.

13. The method of claim 12 further comprising the steps of:

p) comparing said ratio signal R determined in step h) to a threshold level, and

q) utilizing in said neural network the initial formation decay constant .mu..sub.2 determined in step i) for all three of said exponential terms attributable to borehole, formation and thermal neutron background effects if R is less than or equal to said threshold level and repeating step i) to determine a second formation decay constant .mu..sub.2 from only the two of said exponential terms attributable to formation and thermal neutron background effects if R is greater than said threshold level.

14. The method of claim 12 further comprising the steps of:

p) comparing said initial formation decay constant .mu..sub.2 determined in step i) for all three of said exponential terms to a threshold level, and

q) utilizing in said trained neural network said initial decay constant for all three of said exponential terms if it is less than or equal to said threshold level, and repeating step i) to determine a second formation decay constant .mu..sub.2 for use in said neural network from only the two of said exponential terms attributable to formation and thermal neutron background effects if said initial formation decay constant .mu..sub.2 is greater than said threshold level.

15. The apparatus of claim 1 wherein a second trained neural network is trained for the producing of said initial formation decay constant, for the producing of said ratio of borehole amplitude to formation amplitude components, and for the determining of said epithermal neutron lifetime correction and apparent standoff.

16. The method of claim 4 wherein steps (f)-(h) are carried out with a second trained neural network.

17. The method of claim 7 wherein steps (h)-(j) are carried out with a second trained neural network.

18. The method of claim 11 wherein steps (h)-(j) are carried out with a second trained neural network.