0 valutazioniIl 0% ha trovato utile questo documento (0 voti)

3 visualizzazioni6 paginetertu

Aug 08, 2018

© © All Rights Reserved

PDF, TXT o leggi online da Scribd

tertu

© All Rights Reserved

0 valutazioniIl 0% ha trovato utile questo documento (0 voti)

3 visualizzazioni6 paginetertu

© All Rights Reserved

Sei sulla pagina 1di 6

Online English edition of the Chinese language journal

Cite this article as: PETROL. EXPLOR. DEVELOP., 2015, 42(1): 123–128. RESEARCH PAPER

bumping wells

JIA Deli1,2,*, LIU He1, PEI Xiaohan1, WANG Guoqing3, YANG Qinghai1, WANG Yao2

1. PetroChina Research Institute of Petroleum Exploration & Development, Beijing 100083, China;

2. Harbin University of Science and Technology, Harbin 150080, China;

3. Research Institute of Production Engineering, PetroChina Daqing Oilfield Company, Daqing 163453, China

Abstract: A direct test approach of downhole pump diagram for beam bumping wells was presented using acceleration transducer to

measure the pump stroke, and it was verified through laboratory experiments and field tests. The feature of the low-frequency and weak

signal exported from the acceleration transducer with impact interference was analyzed. The periodic signal and its drift were modeled to

work out the accumulative error of long-term integration and its influence on measurement results. Based on elimination principles, the

DC-blocking and denoising filter was designed and its mathematical model was established. Through the filter, the drift component can

be separated from the acquired signal. The laboratory experiments and field tests indicate that, the displacement signal without drift can

be obtained from the initial acceleration data through filtering and quadratic integral, thus the downhole pump diagram can be obtained;

the data measurement error is within ±2.0%, which will be further reduced with the increase of stroke times; the test approach of down-

hole pump diagram can realize the calculation of pump effectiveness and the direct measurement of stroke loss, thereby providing criteria

for the process analysis of beam bumping well.

Key words: pump unit; downhole pump diagram; stroke; acceleration; filter

re-integrate the pump stroke. In this way, the working condi-

The downhole pump diagram of beam pumping well, which

tions of pump are accurately identified.

can directly reflect the working conditions of the pump, is an

In recent years, although MEMS sensor technology is de-

important reference for the analysis of well failure and pa-

veloped and improved continuously, there are still problems

rameters optimization of lifting techniques. Currently, the

such as random noise and DC component interference[7]. In

downhole pump diagram is usually obtained through calcu-

order to remove random drift noise of MEMS sensor, some

lating with the elastic mathematical model of pump rod after

researchers put forward many solutions such as Kalman fil-

the test of polished rod at the surface to acquire indicator dia-

gram of polished rod (also referred to indicator diagram at the tering approach and elimination approach[8−10]. These ap-

point of suspension or surface indicator diagram)[1], which proaches are merely applicable to the conditions where the

indirectly shows the working conditions of the pump[2]. How- required measurement precision is not high or where there are

ever, due to strong Coulomb friction between pump rod and supporting technologies for compensation[11−12], but not ap-

oil tube, which is a complicated non-linear process of me- propriate for the test of pump diagram which demands a high

chanical electrical and hydraulic coupling[3−6], as well as com- precision. When the acceleration transducer is used to meas-

plexity and unpredictability of downhole working conditions, ure the downhole pump stroke of beam pumping well, the

it is very difficult to establish an accurate mathematical model, signals acquired are of low frequency (0.05 Hz minimum) and

resulting in errors between the indirect surface measurement weak. In addition, the Coulomb friction between pump rod

results based on the mathematical model and the real pump and oil tube results in strong interference signal, which be-

conditions. So, the pump diagram derived is only approximate. longs to high-frequency impact interference. The interference

This article presents a new test approach of downhole pump intensity can be dozens of times of the peak value of the true

diagram for beam pumping well. The approach uses the Micro signal. In real test, the drift of transducer changes along with

Electro Mechanical System (MEMS) sensor to measure the the strong impact. As ultra-low bandwidth hardware filter

* Corresponding author. E-mail: jiadeli422@petrochina.com.cn

Foundation item: Supported by the National High Technology Research and Development Program (863 Program), China (2012AA061300).

Copyright © 2015, Research Institute of Petroleum Exploration and Development, PetroChina. Published by Elsevier BV. All rights reserved.

JIA Deli et al. / Petroleum Exploration and Development, 2015, 42(1): 123–128

discharge of capacitance in filter will lead to more serious

infidelity when the signals are influenced by strong impact.

The article makes a systematic analysis and modeling of the

low-frequency and weak signals interfered by strong impact Fig. 1 Mechanism of DC-block and denoising filter

and influenced by drift. Moreover, a DC-block and

1.2 Design of DC-block and denoising filter

drift-removal filter is designed according to the elimination

principle. This filter can isolate the drift of signal, so that The mechanism of DC-block and denoising filtering or

long-time integration will not lead to drift accumulation. After separation of drift component from acquired signal is shown

re-integration, accurate pump stroke can be calculated so as to in Fig. 1. The output signal Y can be obtained by separating

measure pump diagram at the upper part of the pump. drift component A from acquired signal X.

Assuming,

1 Numerical analysis and filter design i + lTR −1

ai = k =i

(k=n, n+1, ..., n+lTR-1) (7)

The pump displacement (stroke) can be calculated through lTR

the re-integration of output of acceleration transducer. The Actually, the nature of Equation (7) is to conduct average

initial discrete signal acquired is assumed to be: filtering to X. As the window length of filtering is the sam-

X = R+ A (1) pling point number of each period of R, the length of A is

There are many factors leading to signal drift and noise. In different from that of X. Assuming the length of X is lX, then

addition to white noise, they include drift generated by trans- the length of A is lA=lX - lTR. Compared with sequence X, A

ducer manufacturing error, signal infidelity generated from generates the time delay of TR/2. It can be obtained through

circuits such as integrator, lateral vibrating signal due to the adjusting corresponding relations of sequence that:

friction of pump rod, (random or systematic) errors due to i + lTR −1

for systems. But all systems have minor DC component or yi = xq − k =i

(8)

lTR

random interference of long period. Theoretically, they can be

reduced to a very low level. lTR

where, q=i+

Usually, AR model (autoregressive model) is used to model 2

A and the typical p-step AR model is: Equation (8) is the mathematical model of DC-block and

p denoising filter. The following equation can be obtained by

ai = c + ∑ϕ j ai − j + ε i (i=1, 2, ..., n) (2) combing Equations (5) and (8):

j =1

i + lTR −1

For acquired data within one period, Equation (2) can be ∑ (r k + dk + ε k + c )

simplified to: yi = rq + d q + ε q + c − k =i

(9)

ai = c + di + ε i (i=1, 2, ..., n) (3) lTR

Equation (1) can be expressed as: The following equation can be obtained through the elimi-

X = R+C + D+ E (4) nation of constant item in Equation (9):

i + lTR −1 i + lTR −1 i + lTR −1

In general, the period of E is far smaller than that of R and

the average value of elements is 0; D is far smaller than effec- ∑ rk ∑ dk ∑ εk

yi = rq + d q + ε q − k =i

− k =i

− k =i

(10)

tive value of R and the period of D is equivalent to that of R lTR lTR lTR

or D does not have any periodicity. i + lTR −1

k =i

rk =0. As the period of E is far less than

xi = ri + c + di + ε i (5) i + lTR −1

It can be deduced by integrating X that:

n n n n

that of R, it is believed that ∑

k =i

ε k =0 and Equation (10) can

∑ x Δt = ∑ r Δt + ∑ d Δt + ∑ ε Δt + ncΔt

i =1

i

i =1

i

i =1

i

i =1

i (6) be simplified to:

i + lTR −1

∑

n n

dk

In Equation (6), ∑ r Δt

i =1

i converges and ∑ ε Δt

i =1

i can be

yi = rq + d q + ε q − k =i

(11)

n

lTR

approximate to 0. However, ∑ d Δt + ncΔt

i =1

i is divergent. By integrating Y, the following equation can be obtained:

⎛ n + lTR lT

n+ R

lT

n+ R ⎞

Along with the prolongation of integration time or the in- n

⎜ 2 2 2

1 n TR

i + l −1

⎟

crease of n, this item will increase and annihilate the integra- ∑ yi Δt = ⎜ ∑ q r + ∑ ε q + ∑ d q −

l

∑ ∑ d k ⎟ Δt

i =1 ⎜ q =1+ TR

l l

q =1+ R

T l

q =1+ R

T TR i =1 k = i ⎟

tion of the tested signal; therefore, a DC-block and denoising ⎝ 2 2 2 ⎠

filter shall be designed to eliminate the influence of this item. (12)

− 124 −

JIA Deli et al. / Petroleum Exploration and Development, 2015, 42(1): 123–128

lTR

n+

2

In Equation (12), it is also believed that ∑ lTR

ε q =0. Addi-

q =1+

2

tionally:

n+

lTR

n i + lTR −1

⎛ lTR −1 lTR

−1 n+

lTR

n+

lTR

⎞

2

1 1 ⎜ 2 2 2 2

⎟ Fig. 3 Data processing

∑l d q − l ∑∑ dk = ⎜ ∑

lTR ⎜ f =1

∑ de + ∑ ∑ dh ⎟

TR

q =1+

TR i =1 k =i e= f s=n h=s ⎟ Fig. 4 shows the acquired initial acceleration data, of which

2 ⎝ ⎠

(13) the waveform is similar to general periodic signal. The sam-

Let: pling frequency is 200Hz. Velocity signals can be obtained

⎛ lTR −1 lTR −1 lT

n+ R n+

lTR

⎞ through long-time integration of acceleration (Fig. 5). As there

1 ⎜ 2 2 2 2

⎟

Er = ⎜ ∑ ∑ d e + ∑ ∑ dh ⎟ (14) are weak DC signals in initial signals, severe overall drift

lTR ⎜ f =1 e = f s =n h=s ⎟ (which is random drift) is generated after integration (upward

⎝ ⎠

drift in Fig. 5). The displacement can be obtained through

Then Equation (12) is expressed as:

lTR

re-integration of post-drift velocity data (Fig. 6). Its waveform

n+

n 2 is periodic to some extent but the drift is very severe.

∑ y Δt = ∑

i =1

i

lTR

rq Δt + Er Δt (15) The DC-block and denoising filter proposed in this article

q =1+

2 is used to re-process the initial acceleration data. For the ex-

In Equation (15), the error item Er will not increase along periment objects, the parameter lTR of DC-block and denoising

with the prolongation of integration time. Therefore, its value filter shall be configured to be the product of sampling fre-

is still within a very small range. The integration to Y is equal quency and period, that is, 200×8.65=1 730. Fig. 7 is the drift

to integration to R. In general, n is greater or equal to lTR so as component in acceleration signal after filtering. It can be seen

to eliminate the long-time accumulative error. that, in addition to DC component, the waveform shows ob-

vious step shape, with similar step width to that of initial sig-

2 Experiment verification and field test

nal. In the process of filtering, signal A generates a time delay

In order to measure downhole pump diagram of beam

pumping well, a downhole memory indicator is developed

(Fig. 2). The indicator can realize the measurement of load

and its correction and compensation with environment pres-

sure and temperature. The measurement scope of MEMS ac-

celeration transducer is ±1.7 g. It can measure both static and

dynamic acceleration with accuracy of 1000/g; within the

scope of −40–125 °C, the temperature sensitivity is ±0.3%

and the migration precision of zero point is ±0.025 g.

The indicator adopts the DC-block and denoising filter to Fig. 4 Initial acceleration data

test pump diagram of beam pumping well. The rated stroke of

pump is 2.7 m and the stroke frequency is 6.94 times/min, that

is, every stroke needs 8.65s. Fig. 3 shows the data processing,

which requires twice DC-block and denoising filtering.

− 125 −

JIA Deli et al. / Petroleum Exploration and Development, 2015, 42(1): 123–128

Fig. 8 Acceleration data after filtering Fig. 12 Displacement curve after filtering

Peak Upper Lower Measurement

Stroke/m Error/m

No. peak/m peak/m error/%

a 1.328 −1.388 2.716 0.016 0.59

b 1.334 −1.376 2.710 0.010 0.37

c 1.329 −1.378 2.707 0.007 0.26

d 1.323 −1.376 2.699 −0.001 −0.04

e 1.313 −1.368 2.681 −0.019 −0.70

Fig. 9 Velocity curve after filtering

f 1.298 −1.357 2.655 −0.045 −1.67

of a half period. After filtering, acceleration data of a half g 1.349 −1.363 2.712 0.012 0.44

period are lost in the front and in the end respectively, but it h 1.320 −1.379 2.699 −0.001 −0.04

has no influence on technique test (Fig. 8). Long-time integra-

tion of acceleration signal is made after the elimination of can be seen that the long-time integration didn’t lead to the

drift component and the acquired velocity signal (Fig. 9) is error accumulation by using the filter presented in this article.

compared with Fig. 5. The result shows that the velocity sig- If the rated stroke of 2.7 m is taken as the criteria, the meas-

nal (Fig. 9) has no drift. urement error of this data set is within ±2%.

The drift component can be obtained by filtering velocity When the acceleration is measured by downhole pump

signal (Fig. 10), which also shows periodicity and the data stroke, the stroke frequency is an important influential factor.

loss of a half period. Re-integration of velocity data after fil- In order to verify the adaptability of the approach for different

tering is made (Fig. 11) and the downhole pump displacement stroke frequencies, the simulation platform for rod pump lift-

curve of beam pumping well can be obtained (Fig. 12), which ing developed by Research Institute of Production Engineer-

well reflects the movement of the pump. ing, Daqing Oilfield is utilized, with the same stroke but dif-

Through further processing of data in Fig. 12 (Table 1), it ferent frequencies. A small stroke of 1.98m (the stroke meas-

urement error of experiment platform is ±0.1%) is set, and the

frequency is set to be 3 times/min, 4 times/min and 6

times/min respectively. Fig. 13 shows the post-processing

experiment data. In order to verify the approach which has the

capability of correcting accumulative errors of long-time inte-

gration, 50 data samples are analyzed. On Fig. 13, with the

increase of frequency, the period shortens, the error tends to

converge and the measurement accuracy increases.

Test is conducted for downhole pump diagram of field

Fig. 10 Velocity filtering error pumping well and the pump stroke is adjusted from 2.0 m to

− 126 −

JIA Deli et al. / Petroleum Exploration and Development, 2015, 42(1): 123–128

Fig. 14 Comparison of pump diagrams at different strokes Fig. 15 Comparison of pump diagram at the point of suspension

and downhole

2.8 m. For the basic parameters of tested well, the model of

pump unit is CYJ10-3-37HB, the medium is water, the well odic signal and its drift. The accumulative error of long-time

does not contain gas, the suspension height of pump is 1 000 integration is calculated and its influences on measurement

m, the dynamic liquid level is 320m and the stroke frequency result are identified. Based on the elimination principle, the

is 6 times/min. In order to verify the consistency of instru- DC-block and denoising filter is designed to separate the drift

ments, two test instruments of the same precision are installed component from the acquired signal. The experiment results

at the bottom-hole. It is configured that data are acquired once indicate that velocity signal can be obtained through the fil-

every 2 hours and the acquisition duration each time is 5 min- tering of initial velocity data by means of DC-block and de-

utes for the test instruments. The pump stroke tested increases noising and integration; after re-filtering and re-integration of

from 2.0 m to 2.8 m, which matches the real adjustment. Fig. velocity signal, the displacement signal without drift is ob-

14 shows the pump diagrams at different strokes. It can be tained. The long-time integration does not lead to error accu-

seen that with the stroke increase, the pump load does not mulation. The data measurement error is controlled within

increase very much. However, the pump diagram is integrated ±2%. With the increase of stroke frequency, the measurement

and the pump effectiveness calculated from the equation in- error decreases. The field test results indicate that the test ap-

creases from 66.0% to 69.4%. proach of downhole pump diagram for beam pumping well

Fig. 15 shows the comparison of pump diagrams at the proposed in this article can directly identify the working con-

point of suspension and downhole pump diagram when the ditions of downhole pump.

point of suspension is adjusted to 2m under above test condi-

tions. It can be seen that, due to the elastic deformation of Nomenclature

pump rod, the stroke of point of suspension is different from

that of pump. The calculation of pump rod effectiveness and X—measured signal sequence or acquired value of transducer;

the design of technique parameters shall take full considera- R—measured real value sequence;

tion of elastic deformation of pump rod. A— overall error sequence;

ai—the ith value of A, m/s2;

3 Conclusions c—constant, m/s2;

Acceleration transducer can be used to directly measure the εi— the ith value of error sequence of white noise, m/s2;

stroke of pump rod, so as to obtain the pump diagram. Rele- φ—model parameter;

vant research is made on high-frequency interference and drift. di—the ith value of error sequence of drift, m/s2;

The low-frequency and weak signals with strong impact in- C—constant sequence;

terference are analyzed. Modeling analysis is made on peri- D—drift error sequence;

− 127 −

JIA Deli et al. / Petroleum Exploration and Development, 2015, 42(1): 123–128

E—error sequence of white noise, the average value is 0; rod pumping system. Petroleum Science and Engineering,

xi— the ith value of X, m/s2; 2007, 58: 43–48.

ri—the ith value of R, m/s2; [6] Miska S, Sharaki A, Rajtar J M. A simple model for computer

Δt —sampling period, s; aided optimization and design of sucker rod pumping systems.

Y—input signal sequence of filter; Journal of Petroleum Science and Engineering, 1997, 17:

lA—length of A, piece; 303–312.

lTR—Sampling points of each period of R (must be a even number), [7] He Pengju, Feng Liang. Study on the real-time elimination

points; method of random noise and trend terms in acceleration sig-

lX—length of X, piece; nal. Electronic Design Engineering, 2013, 21(14): 1701–1705.

TR—true signal period, s; [8] Pan Yue, Song Ping, Li Kejie, et al. A filtering method of gy-

yi—the ith value of Y, m/s2; roscope random drift for miniature unmanned helicopter. In:

n—sampling point number; Proceedings of the 2011 International Conference on Com-

Er—denoising integration error; puter Science and Network Technology. Harbin: IEEE, 2011:

g—acceleration of gravity, 9.8 m/s2. 730–734.

[9] Wu Xingming, Duan Li, Chen Weihai. A Kalman filter ap-

References proach based on random drift data of fiber optic gyro. In: Pro-

ceedings of the 6th IEEE Conference on Industrial Electronics

[1] Zhou Jide. The method of judging the state of beam pump and and Applications. Beijing: IEEE, 2011: 1933–1937.

treating the trouble. Beijing: Petroleum Industry Press, 2005. [10] Chen Weicao, Gao Guowei, Wang Juan, et al. The study of the

[2] Peng Jian, Xu Zhiqiang, Zhou Zhiquan. Design of dyna- MEMS gyro zero drift signal based on the adaptive Kalman

mometer based on sensor technology. Chinese Journal of filter. Key Engineering Materials, 2012, 500: 635–639.

Sensors and Actuators, 2012, 25(12): 1701–1705. [11] Cao Dianguo, Liu Changchun, Wang Peng. Removing base-

[3] Liang Hua, Li Xunming. Accurate extraction of valve opening line drift in pulse waveforms by a wavelet adaptive filter. In:

and closing points based on the physical meaning of surface Proceedings of the 2nd International Conference on Bioinfor-

dynamometer card. Petroleum Exploration and Development, matics and Biomedical Engineering. Shanghai: iCBBE, 2008:

2011, 38(1): 109–115. 2135–2137.

[4] Nolen K B, Gibbs S G, Nabla C. Quantitative determination of [12] Chen Y Y, Chang S J, Huang C Y, et al. An elimination design

rod pump leakage using dynamometer techniques. SPE 18185, for drift rate effects of MEMS-based inertial devices. In: Pro-

1990. ceedings of the 3rd International Microsystems, Packaging,

[5] Xu Peng, Xu Shijin, Yin Hongwei. Application of self-or- Assembly & Circuits Technology Conference. Taipei: IEEE,

ganizing competitive neural network in fault diagnosis of suck 2008: 153–155.

− 128 −

## Molto più che documenti.

Scopri tutto ciò che Scribd ha da offrire, inclusi libri e audiolibri dei maggiori editori.

Annulla in qualsiasi momento.