Well Production Control System

The present invention relates to methods and apparatus for monitoring the operation of sucker-rod well pumping units, and more particularly to methods and apparatus for detecting problems in wells employing sucker-rod pumping units.

Sucker-rod type pumping units are widely used in the petroleum industry in order to recover fluid from wells extending into subterranean formations. Such units include a sucker-rod string which extends into the well and means at the surface for an up and down movement of the rod string in order to operate a downhole pump. Typical of such units are the so called "beam-type" pumping units having the sucker-rod string suspended at the surface of the well from a structure consisting of a Samson post and a walking beam pivotally mounted on the Samson post. The sucker-rod string normally is connected at one end of the walking beam and the other end of the walking beam is connected to a prime mover such as a motor through a suitable equalizer bar connected to a crank and pitman connection. In this arrangement the walking beam and the sucker-rod string are driven in a reciprocal mode by the prime mover.

A variety of malfunctions such as worn pumps, broken sucker-rods, split tubing, and stuck pump valves can interrupt the pumping of fluid from a well. Such malfunctions can be caused by normal wear and tear on the equipment, by the nature of the fluid being pumped or they could be caused by abnormal pumping conditions.

One abnormal pumping condition which is fairly common is known as "fluid pound". Fluid pound occurs when the well is pumped-off, i.e., when fluid is withdrawn from the well at a rate greater than the rate at which fluid enters the well from the formation. When this occurs, the working well of the downhole pump is only partially filled during an upstroke of the plunger and on the down stroke the plunger strikes or "pounds" the fluid in the working barrel causing severe jarring of the entire pumping unit. This causes damage to the rod string and to the surface equipment and may lead to failure of the pumping unit.

The present invention provides new and improved methods and apparatus for detecting problems in a well pumping unit having a sucker-rod string and a power unit to reciprocate the rod string to produce fluid from a well. A load cell is connected on the equalizer bar between the sucker-rod string and the power unit to develop a signal representative of the load on the rod string, and a transducer is connected to generate a signal representative of the position of the rod string. In the present invention an updating means uses the load signal to establish a selected value of this load signal and uses the rod string position to establish a reference position of the rod string. Means are provided for monitoring the load signal when the rod string reaches the reference position and means are provided for disabling the power unit when an absence of fluid below the pump plunger causes the load signal to exceed the selected value with the rod string at the reference position.

When load cells and rod transducers are used in an outdoor environment their characteristics may vary with changes in temperature and changes in weather conditions. This is especially true when low cost load cells and other transducers are used. These changes in characteristics causes the value of the load signal and the value of the rod position signal to drift. The present invention uses a microprocessor to monitor slow changes in load signal and in rod position signal and to calculate updated selected values of load signal and updated values of reference position signals. The microprocessor uses sudden or significantly large changes in load and/or position signals to determine that trouble is present in a well.

The ability of the present invention to use rod string position signals in establishing a reference position for a particular well allows inexpensive apparatus to be used with a variety of wells and allows the well to be automatically recalibrated so the well equipment can be operated for extended periods of time without human intervention. The establishing means includes a microprocessor which stores programs and certain well parameters in nonvolatile memories so that a loss of power at the establishing means will not cause a loss of programs or well parameters, and so operation and control of the well will resume when power is restored.

FIG. 1 is a diagrammatic illustration of a well equipped with a sucker-rod type pumping unit.

FIG. 1A is a cross-sectional view of the pumping unit taken along the line 1A--1A of FIG. 1.

FIG. 2 is a plot of the position vs. load of the sucker-rod of the pump for one cycle of normal operation and showing a reference point in the plot.

FIG. 3 is a plot of position vs. load of the sucker-rod as the well progresses into fluid pound and showing change in the plot as transducer characteristics change.

FIG. 4 is a plot of position vs. load of the sucker-rod as the well progresses into gas pound.

FIG. 5 is a graph illustrating the process of interpolation of values of sucker-rod position and load values to accurately determine the load value at a reference position.

FIGS. 6A, 6B comprise computer circuitry which can be used in the apparatus of FIG. 1.

FIG. 7 is a matrix diagram illustrating the operation of software state machines used in the present invention.

FIG. 8 is a diagram illustrating symbology of a typical software state machine used in the present invention.

FIG. 9 illustrates a message switched software operating system of the present invention.

FIG. 10 illustrates a software state machine scheduler of the present invention.

FIG. 11 is a flow chart showing the process of dynamically calibrating the well to account for a drift in transducer characteristics and well characteristics.

FIG. 12 is a message flow diagram showing the mode of operation of the apparatus of FIG. 1.

FIG. 13 is a state diagram of a set point fluid pound detector of FIG. 6 used to detect well pump-off.

FIGS. 14 and 15 illustrate the flow of data through the operating system and math utility of the present invention.

FIG. 16 illustrates typical position and position derivative waveforms in the apparatus of the present invention.

FIG. 17 illustrates the relationship between smoothed (filtered) data signals and nosiy (unfiltered) signals and shows signal phase shifts which must be considered in apparatus of the present invention.

FIG. 18 is a message flow diagram of a stroke discriminator of the present invention.

FIG. 19 is a software state diagram of the stroke discriminator of the present invention.

FIG. 20 is a software state diagram of a stroke derivative detector of the present invention.

FIG. 21 is a software state diagram of a stroke extremes detector of the present invention.

FIG. 22 is a software state diagram of a stroke area calculator of the present invention.

FIG. 23 illustrates a procedure used in calculating the area inside a dynagraph curve for a typical well.

Referring to FIG. 1, there is illustrated a wellhead 10 of a well which extends from the earth's surface 11 into a subsurface well producing formation (not shown). The wellhead comprises the upper portions of a casing string 12 with a sucker-rod string 16 extending downward into a down hole pump (not shown) which moves liquid to the surface where it passes into a flow line 17. The sucker-rod string 16 is suspended in the well from a support unit consisting of a support post 18 and a walking beam 22 which is pivotally mounted on the support post by a pin connection 23. A cable section 24 is connected between the upper end of the sucker-rod string 16 and the lower end of a horsehead 28. The cable section 24 is connected to the walking beam 22 by means of the horsehead 28.

The walking beam 22 is reciprocated by a prime mover such as an electric motor 30. The prime mover drives the walking beam through a drive system which includes a drive belt 34, crank 35, crank shaft 36, a pair of crank arms 37 (only one shown), and a pair of pitmans 41a, 41b which are pivotally connected between the crank arm and the walking beam by means of an equalizer bar 42 and an equalizer bearing 43 (FIGS. 1, 1A). The outer end of the crank arms 37 are provided with a counterweight 46 which balances a portion of the load on the sucker-rod string in order to provide a more constant load on the prime mover. A load cell 47 is clamped or otherwise connected to the equalizer bar 42 at a position between the equalizer bearing 43 (FIG. 1A) and the pitman 41a. The load cell 47 develops a signal due to the slight bending of the equalizer bar 42 caused by the load on the sucker-rod string 16. The amount of bending of the equalizer bar 42 is determined by the amount of load on the rod string 16 (FIG. 1).

The load cell 47 provides a DC output signal which is proportional to the load on the sucker-rod string 16, and an analog-to-digital converter 48 provides a corresponding digital signal to a computer 49a. A position measuring means or transducer 53 measures the vertical position of the sucker-rod string 16 by providing a voltage which is proportional to the angle of the walking beam 22 and thus is proportional to the position of the rod string 16. The digital-to-analog converter 48 also converts the signal from the transducer 53 into a digital signal which is used by the computer 49a and by an XY plotter 54. Signals are transferred between the computer 49a and a computer 49b by a pair of wires 55a, 55b. Instructions from a keyboard 60 and from a control and display unit 61 and output signals from the load cell 47 are used by the XY plotter to provide a visual plot of the characteristics of the particular well which the rod string operates. The plotter 54 can be used for observing operation of the well and for setting up the equipment to monitor the well. After setup is completed the plotter can be disconnected, or if desired the plotter can be eliminated altogether and the display unit 61 or other means for setting up the equipment can be used.

A plot of the position versus load of the rod string 16 for a typical cycle of the rod string when the well is filled with fluid is disclosed in the solid line graph of FIG. 2. It can be seen that as the rod string moves on the upstroke from the Xmin position to the Xmax position, the load on the string increases to a maximum value and then returns to approximately the initial value. Of more importance is the variation in the load as the rod string moves downward with the load decreasing to a minimum value at a fairly rapid rate and then moving upward to approximately the original value at the Xmin position.

As the well approaches pump-off (FIG. 3), the load on the rod string changes more rapidly as the rod string moves in a downward direction. When the fluid in the well drops, a pump plunger in the pump falls and strikes the surface of the fluid in the well producing a "fluid pound" which can damage the rod string and other parts of the pumping system. As the fluid level in the well decreases the pump plunger progressively moves a greater distance on the downstroke before contacting the surface of the fluid in the well causing the plotted load curve to progressively change from the full well curve 65 to the dotted curves 66-69 with the curve moving progressively toward the left as the fluid in the well drops lower. This moving trend can be observed and the pump shut down to prevent damage to the equipment.

The present invention provides a method for detecting pump-off by using the apparatus of FIG. 1 to select a set point (Xset, Yset) (FIGS. 2, 3) having a value determined by the characteristics of each individual well and to change the set point when these characteristics change and/or when the characteristics of the load cell 47 and/or of the transducer 53 change. The computer 49a (FIG. 1) compares the fluid pound curves 66-69 with the position of the set point and shuts down the motor 30 when the fluid pound curve moves to the left of the set point shown in FIG. 3.

A human operator uses the keyboard 60 or other input to the computer 49b (FIG. 1) to enter an X percentage value and a Y percentage value into the computer 49b which transfers these values to the computer 49a causing the computer 49a to calculate an Xset value the entered percent of the distance between Xmin and Xmax (FIG. 2), and to calculate a Yset value the entered percent of the distance between Ymin and Ymax thereby obtaining the position of the set point. The value of Xset and Yset can be computed using the following formulae:

Xset=(Xmax-Xmin)(X%.div.100)+Xmin

Yset=(Ymax-Ymin)(Y%.div.100)+Ymin

The values of Xmax, Xmin, Ymax and Ymin which can be used are the maximum and minimum values of the curve of FIGS. 2 and 3. The X% and Y% are the percentage values selected by the human operator using knowledge of the well and of the pumping equipment in choosing these percentage values. Also any two nominal values of X and any two nominal values of Y can be selected instead of using the maximum and minimum values suggested.

A change in temperature can change the characteristics of transducers 47 and 53 and cause the signals from the load cell 47 (FIGS. 1, 6A) and from the position transducer 53 to gradually change from values on the solid line graph of FIG. 2 to values on the dotted line graph of FIG. 2. To compensate for this change and prevent a change in transducer characteristic from producing a false indication of pump-off, the values of Xset and Yset are periodically updated to correspond to the dotted graph (FIG. 2) by calculating the new values X'set and Y'set using maximum and minimum values of X and Y on the dotted graph. Thus, a drift in characteristics of the load cell 47 and of the transducer 53 does not change the relationship of Xset and Yset to the graph being plotted. Several cycles of operation are used in calculating the values of Xset and Yset so that a sudden change in the shape of the graph, due to pump-off, will produce only a small change in the values of Xset and Yset, and pump-off can be detected.

When the set point (Xset, Yset) has been selected the computer continually monitors the X value of the curve (FIG. 3) during the downstroke of the plunger until the curve reaches the value of Xset as the curve moves from Xmax toward Xmin. With the curve at Xset point the computer checks the value of Y. If the value of Y is greater than the value of Yset the computer 49a (FIG. 1) provides a signal which causes the motor 30 to stop and the well is shut down. To insure that the well is really pumped-off at this time, it may be desirable to allow the pump to move through two or more cycles with the curve (FIG. 2) to the left of the set point each time, before the motor 30 is turned off. This prevents shut down of the well due to an erratic signal from the load cell 47 or from the transducer 53 or from other electronic equipment or from the behavior of the well itself.

It is also important to be able to distinguish the difference between fluid pound and "gas pound" in the well being monitored. Gas pound occurs when the well is filled with fluid but gas is present in the fluid being withdrawn from the well, and the gas delays the shift of the fluid load from a valve in the pump in the downstroke because the gas is compressible. However, the gas and fluid mixture offers more resistance to downward movement of the plunger than is offered in a pump-off condition so the plunger drops more slowly than in fluid pound. These differences can be seen by comparing the full well card of FIG. 2 with the fluid pound curve of FIG. 3 and with the gas pound curve of FIG. 4.

The gas content of the fluid being pumped from a well may vary in an unpredictable manner so that the downward stroke of the pump plunger may jump back and forth in a random manner between the downstroke curves 70a-70e of FIG. 4. For example, on the downward stroke the load cell 47 and the stroke transducer 53 (FIG. 1) may provide the curve 70b, while the next downstroke develops the curve 70e and the next downstroke develops the curve 70c.

When a well is being pumped-off the fluid level gradually drops so the pump rod load follows curve 65 (FIG. 3) on one downstroke, then follows curve 66, then 67, etc. toward curve 69 with the output of the load cell 24 (FIG. 1) gradually moving toward the left on subsequent downstrokes, as seen in FIG. 3. This difference between a leftward trend in fluid pound and a random movement in gas pound can be used to aid in distinguishing between these two conditions.

Details of a method and apparatus for automatic calibration of a well and for monitoring operation thereof are disclosed in FIGS. 6A, 6B and 11-23. When FIGS. 6A, 6B are placed side-by-side with leads from the right side of FIG. 6A extending to corresponding leads from the left side of FIG. 6B the two sheets comprise a block diagram of an embodiment of the computers 49a, 49b (FIG. 1).

The portion of the computer system disclosed in FIG. 6A comprises a motor controller 71 for receiving signals from the load cell 47 and from transducer 53 and for using these signals to determine the sequence for controlling the motor 30. The computer 49b disclosed in FIG. 6B comprises a display programmer 72 for using the load cell and transducer signals transmitted from computer 49a to operate the XY plotter 54. Signals are interchanged between the motor controller 71 and the display programmer 72 over the pair of interconnecting wires 55a, 55b.

Each of the controller/programmers 71, 72 includes a microcontroller 73a, 73b, a PROM 74a, 74b, a RAM 75a, 75b and a memory decoder 76a, 76b connected for the interchange of information and instructions over a system bus 80a, 80b. A microcontroller 73a, 73b which can be used in the present invention is the Model 8031 manufactured by Intel Corporation, Santa Clara, Calif. and includes an internal computer and a link (not shown) for sending and receiving messages.

Clock pulses for driving the microcontrollers are stabilized by a pair of crystals 81a, 81b. The controller 73a is connected to a power reset circuit 82 to warn that power to the controller is failing. An indicating device 83a receives visual display information from an input/output interface 84 and the graphic display 61 receives visual display information from a display controller 85. Programs for operating the motor controller 71 and the plotter programmer 72 are stored in the PROMS 74a, 74b and data for use in the system is stored in the RAMS 75a, 75b. A load/stroke conditioner 88 (FIG. 6A) amplifies and filters signals transmitted from the load cell 47 and the transducer 53 and sends the smoothed signals to the bus 80a through a multiplexer 89 and the analog-to-digital converter 48. A buffer 87 (FIGS. 1, 6A) provides signals to operate the XY plotter 54 in response to signals from the multiplexer 89. An analog-to-digital converter which can be used is the model AD574A manufactured by Analog Devices.

A procedure for compensating for a drift in characteristics of the load cell 47 (FIGS. 1, 6A) and the stroke transducer 53 is disclosed in FIGS. 2 and 11. A change in load cell and stroke transducer characteristics due to a change from warm daylight to cool evening can cause the graph or card plotted by the XY plotter 54 to gradually change from the solid line curve of FIG. 2 to the dotted line curve. It is desirable that this change in sensor characteristics not cause the well to shut down because of a change in the location of the curve relative to the XY set point, nor should this change prevent a shut down of the well when pump off occurs. Recalculating the position of the set point from Xset, Yset to X'set, Y'set keeps the set point in proper relationship to the curve. It is also important to ascertain that the area inside the curve stay relatively constant as the curve moves back and forth between the solid line position (FIG. 2) and the dotted line position.

The procedure shown in FIG. 11 checks to see if the pump has been running so the well is stabilized or to see if the pump has just started operation. If the pump has just started more sensor readings are taken. If the well is stabilized the area of the curve is calculated and compared with a predetermined area to see if the area is within acceptable limits. If the area is acceptable the maximum and minimum rod load values are checked against acceptable values. If everything is within acceptable limits the new area and load values are combined with the last 10 previous values to obtain moving averages of the area, upper load limit and lower load limit. If area and load limit values are outside acceptable limits the new values are not accepted as part of the moving average.

The general operation of a method for detecting pump-off using apparatus of the present invention has been described in connection with FIGS. 1-4. A detailed description of the selection of the set point (Xset, Yset) and the method of using the motor controller 71 and the plotter programmer 72 to determined when the well is in fluid pound will be described in connection with FIGS. 5-23 which provide background of the use of software state machines and of their use in operating the apparatus of FIGS. 1, 6A and 6B and provides details of the operation of a computer program in carrying out various operations performed by the computer of FIGS. 6A, 6B.

The program of the present computer is supported by a real time operating system having various routines that are not applications oriented and that are designed specifically to support programs designed with the state machine concept, that is, a state, input driven program. Some of the routines are sub-routines while others form a module that creates a simple real-time environment under which software state machines can operate. The operating system provides equipment in which a collection of software state machines can operate.

A software state machine is a process that is executed on the digital computer each time that a message is sent to the state machine. The process does not execute in exactly the same way each time that a like message is sent to it because the processing to be done for any message depends on the machine's "state", i.e., its memory of all prior processing that it has done in response to the previous messages. The state can be any length, from eight binary digits to several thousand binary digits depending upon the complexity of a given machine. Given the state of the machine and the current message, the machine will do a given set of processing which is totally predictable. A machine can be represented as a matrix of processes, indexed by a state and a message as shown in FIG. 7. For example, if the state machine of FIG. 7 receives message number one in state one, then process A will be done. If process A were to cause the state to be changed to state 2 then a second message number one, coming right after the first message would cause process D to occur which could cause the machine to change to state 3. It is not necessary that a process cause the state to change, although it may do so in many cases.

A software state machine, upon completing its process defined by the state and by the message returns control to the program that called it, the state machine scheduler which will be described below. During the given process, the machine is not interrupted in order to give processing time to another machine of the same system. Thus, processing time appointment between a given machine and any of its contemporaries in the system is on a message-by-message basis, and such an environment is called a message switched operating system (MSOS). None of the machine's processes are ever suspended for the processes of another machine. For example, if message three comes in state one, process C will begin and end before another state machine can have the central processing unit (CPU) in microcontroller 73a (FIG. 6A) to respond to its next message in its given state.

Certain things can cause a state machine process to "suspend". For example, an asynchronous interrupt can be registered and processed. A requirement of the operating environment is that such hardware events are turned into software messages to be processed in order by the responsible state machine. Only that processing that must be done at the exact instant of the interrupt is done and then the interrupt service process will cause a software flag to be raised, ending the interrupt process. When the operating system notes an asynchronous flag (semaphore), it generates the needed software message to be sent to the state machine that will carry out the non-time-critical segment of the interrupt processing. An example of such a process is data collection at precisely timed intervals. When the timer interrupt signals that data must be collected, it is read in the required manner dependent on the type of the data, queued in a storage area for processing at a later time, and a flag is raised. When this raised flag is noted by the operating ssystem, a software message is generated, the data is stored and the state machine that is responsible for the processing of this data receives the messate at a later time.

A state machine is not given access to the processor by the operating system on a regularly timed basis but is connected to the processor only in order for it to process a message. Whenever the processing of a message is completed the state machine must insure that it will get another message at some point in the future. This is done in the following ways:

(1) Another machine sends a message for synchronizing purposes.

(2) A time period elapses signaled by a timer message.

(3) Real-time data becomes available from some queue.

(4) An input which is being polled, achieves the desired state, and initiates the software message.

(5) An interrupt is sensed and a software message is sent to inform the state machine about this event.

The only time that a machine cannot take care of itself is prior to receiving its first message, so the operating system takes the responsibility of initiating the system by sending to all of the software state machines, functioning therein, an initializing message referred to herein as a "power on" message. No matter what the state of the machine it will respond with a predetermined given process when this message is received independent of the state of the machine.

A convenient means of illustrating the operation of a software state machine is shown in the state machine symbology of FIG. 8 using the messages of FIG. 7 to do some of the processes and to move into some of the states shown in FIG. 7. If we assume the machine (FIG. 8) to be initially in state one, the receipt of message one causes process A to be performed as the transition action for message one received in state one and also causes the machine to move into state two. In state two the receipt of message two causes process E, causes a message to be sent out to another state machine and moves this state machine back into state one. In state one the receipt of message three causes process C as the transition action for receiving message three in state one but does not cause any change in the state of the machine. Some of the other states and processes shown in FIG. 7 are not repeated in FIG. 8 in order to simplify the drawing.

A message switched operating system of the type shown in FIG. 9 includes a main procedure which provides signals to initialize the system through a system initializing procedure and includes the initialization of various interrupts, timers, the scheduler, inputs, data acquisition, the RAMs, the math utility and outputs as well as initializing the available message blocks so that all dynamic memory is put into an available space queue for storing data. The procedure then calls the duty cycle procedure which sequentially calls the asynchronous processing, state machine scheduler and synchronous processing over and over again. All interrupt programs communicate with the duty cycle program by way of semaphores. The duty cycle program runs indefinitely with a state machine message delivery, an asynchronous operation and all synchronous operations timed by the real-time clock for each cycle of the loop. Asynchronous operations that can occur are: data input from a real-time data acquisition queue and communication line interrupts to move characters in and out of the system. In the asynchronous operation significant events occurring cause an available message block to be secured and turned into a message to be delivered to whatever state machine is charged with processing the particular interrupt. Since the data is queued at the time of acquisition, the transfer operation is asynchronous. If the data processing falls behind the data input, the system can use the time between synchronous clock ticks to catch up on the required operation. Details of the data flow in the asynchronous processing of the DQ block of FIG. 9 are shown in FIG. 14. Signals from the load cell 47 and the stroke transducer 53 (FIG. 14) are acquired by the GET XY data procedure and are transferred into the XY data Q in RAM 75a (FIG. 6A) by the PUT XY Q procedure in response to a real-time clock interrupt and are removed by the GET XY Q procedure.

Once the data has been acquired it is processed by the math utility (at PM, FIG. 9). The math utility accesses the raw values of stroke (X) and load (Y) and smoothes the values of X and Y. The smoothed value of X (X) (FIG. 15) and the smoothed value of Y (Y) are obtained by using a moving average smoothing technique where the last n values of X (or Y) received are added and divided by the number of values (n) to obtain a first smoothed value. To obtain the next smoothed value, X, the newest value is included in the sum, but the oldest received value is not included.

The first derivative, X' is then computed and X is corrected for the time lag introduced by the computation of the first derivative to obtain the result Xlag. The values of X', Xlag, Y' and Ylag are then sent to all state machines that have signed up for these values using the "send message" procedure (FIG. 10) to place the messages on the queue of messages to be delivered.

The first derivative is computed using a method developed by A. Savitzky and M. Golay and described in detail on pages 1627-1638 of the July 1964 issue of "Analytical Chemistry" magazine. This method uses a least squares quadratic polynominal fit of an odd number of points and a corresponding set of convolution integers to evaluate the central point. The derivative computed corresponds to the value at the midpoint of a window of equally spaced observations. The value obtained is identical to the best fit of the observed values to the quadratic polynominal A.sub.2 X.sup.2 +A.sub.1 X+A.sub.0 =y. A.sub.2, A.sub.1, and A.sub.0 are selected such that when each X (for the number of points in the window) is substituted into this equation, the square of the differences between the computed values, y, and the observed number is a minimum for the total number of observations (window size). Once A.sub.2, A.sub.1 and A.sub.0 are found the central point is evaluated. The Savitzky - Golay method uses a set of convoluting integers and the observed data points to evaluate the central point.

Since the derivative is evaluated at the center of the set of data a lag equal to the (window size -1) divided by 2 is introduced. Details of the math utility for obtaining values of X', Xlag, Y' and Ylag are shown in FIG. 15.

The synchronous processing performs hardware input polling, timer aging and signal delivery. When an input, requested for polling by any state machine, gets to the desired state such as an off condition, an on condition, above a level or below a level, etc. an available message block is sent as a message to the requesting machine indicating that a given input is in the desired state. The input will no longer be polled until another request is made.

The timer process is slightly different in that the timer queue is made up of message blocks serving as receptacles for the machine requesting the marking of the passage of time and the time of day when the time will be completed. When the time is completed the block is removed from the timer queue and placed on the message delivery queue as a message. Thus, all responsibilities placed on the state machine are accomplished in the operating system by transferring software messages and by the use of real-time flags and queues (semaphonres).

The first component of the operating system (FIG. 9) is a program to deliver a message to a state machine (FIGS. 9, 10). A message is a small block of dynamic memory that is queued for delivery to a designated state machine. This program is called a state machine scheduler and shown in detail in FIG. 10 selects the next highest priority message from the queues of messages ready for delivery. The machine looks up the designation state machine code stored in the message and uses that code to select the proper state machine program to be called with a pointer to the message block as an input. Contained in the program is a state memory. With the memory and the state the proper process can be delivered and executed, and the memory block transferred from the delivery queue to the available space queue for subsequent reuse. Two examples of data that is reused are instructions for sending the messages or setting timers. These processes take available blocks and turn them into messages that will be on the message delivery queue at some later time. Programs such as the message sender and the timer starter are service utilities called by the state machine in order to fulfill the responsibilities alluded to earlier. The state machine scheduler program is the lowest form of the hierarchy which forms the main duty cycle of the operating system. In the diagram of FIG. 9 the relationship of the scheduler to the rest of the operating system is shown.

When power is turned on in the computer of FIGS. 6A, 6B, the power reset generator 82 provides signals which reset various hardware in the computer and cause the first instruction of the computer program stored in the PROM 74a to be executed by the central processor in controller 73a. A "power on" message is sent, in the manner previously described, to each of the state machine modules 91-95 (FIG. 12) in the computer and these state machine modules are initialized. The load signal values from the load cell 47 (FIG. 6A) and the stroke signal values from the transducer 53 are obtained by the processor in microcontroller 73a through conditioner 88 and converter 48 and stored in the RAM 75a (FIGS. 6A, 14) for use by the stroke discriminator which uses these signals to detect maximum and minimum values of load and rod position. The maximum and minimum values of load and rod position are available to other state machine modules upon request.

The stroke discriminator 93 (FIG. 12) provides signals to the fluid pound detector 92 at the start of the downstroke. Details of the stroke discriminator 93 (FIG. 12) and its method of operation are disclosed in FIGS. 16-23 where curve 104 (FIG. 17) shows a typical raw derivative of the rod string 16 (FIG. 1) position vs. time, and curve 105 shows the smoothed derivative of the same. An average of several values of the raw derivative from a timed sequence of values are used in obtaining the smoothed derivative thereby causing a lag between the phase of the smoothed derivative and the raw derivative as shown in FIG. 17. The lagged smoothed derivative is used by a stroke derivative detector 109 (FIG. 18) to obtain the maximum and minimum in the stroke value. Once the max and min values are obtained the system stops looking for another extreme value for a predetermined "blackout time" to reduce the average real processing time consumption by the stroke derivative detector. The blackout time also makes the stroke system more immune to noise in the data input from the stroke transducer 53 (FIG. 1).

There are several software messages that are incoming to the stroke discriminator from the pump-off detection system and from other machines that are not neighbors in the state machine hierarchy. These messages include a "power on" message common to all machines, start and stop messages from other machines which ask for a report of the stroke low point, note of the stroke high point, peak reports of X and Y (stroke and load extremes), and area reports. The Xlag, Ylag and X derivative messages are received from the math utility.

The stroke discriminator 93 (FIG. 18) communicates directly with the pump manager 91 and with the subservient stroke derivative detector 109, a stroke area calculator 110, a stroke extremes detector 111 and other state machines 112. The stroke extremes detector 111 uses the raw values of signal from the load cell 47 (FIG. 1) and the position transducer 53 to find the Xmax, Xmin, Ymax and Ymin. The area calculator 110 integrates the area of the dynagraph (FIG. 2), and the stroke discriminator 93 directs the operation of the other state machines 109-112 shown in FIG. 18.

After the pump manager 91 (FIG. 18) turns on the motor 30 (FIG. 1) a motor on message and a start BDC (bottom dead center) report message (i.e., a signup for start of downstroke report) (FIG. 17) are sent to the stroke discriminator 93. The stroke discriminator waits 3 seconds to allow the stroke signal to stabilize and sends a start message to the state machines 109-111 to monitor the well operation. If a fluid pound is detected during the monitoring operation an alarm signal is sent to the pump manager 91 who turns off the motor and provides a motor off signal to the stroke discriminator.

When the stroke discriminator 93 receives a motor on signal from the pump manager 91, it provides a start signal which causes the stroke derivative detector 109 to measure stroke derivative signal noise during a 3 second turn-on delay period. At the end of the 3 second delay the derivative detector 109 uses the measured noise and the stroke signals to provide upstroke and downstroke signals until the stroke discriminator 93 sends a stop message to the derivative detector.

The stroke extremes detector 111 (FIG. 18) provides a min stroke position, load at min stroke, max stroke position, load at max stroke; min load, stroke position at min load, max load, and stroke position at max load each time a status request is received from the stroke discriminator 93. At the time the status request is received a reset occurs and the calculation of a new set of extreme values is started. This process continues until a stop signal is received by the stroke extremes detector 111 from the stroke discriminator 93.

When the stroke area calculator 110 (FIG. 18) receives a start signal from the stroke discriminator 93 the area calculator receives downside and extreme reports which are used to calculate area of the dynagraph (FIG. 2). The calculated value of the area is sent from the area calculator 110 to the stroke discriminator 93 in response to a status-request signal.

When a power on signal is received by the stroke discriminator (at A, FIG. 19) its memory is initialized and mailing lists of the state machines which want to receive reports are prepared. When the motor on signal at B is received from the pump manager the stroke discriminator (FIG. 19) moves from the motor off state to the motor starting state, starts a 3 second timer and sends a start X' noise measure message to the derivative detector to start its measurement of the noise on the stroke derivative during this 3 second period. When the 3 second motor on delay timer has expired (at C) the derivative detector 109 (FIG. 18), stroke area calculator 110 and stroke extremes detector 111 receive start messages and the BDC count is set to zero. The BDC position is the bottom dead center of the left end of the walking beam 22 (FIG. 1) and corresponds to the start of the downstroke of the sucker-rod string 16. A start report signal (at C, FIG. 19) from any of the state machines places the requesting machine on the specified mailing list if it is not already there. A stop report signal (at F) from any of the state machines removes the requesting machine from the specified mailing list.

When an upside signal (at H, FIG. 19) is received from the derivative detector, in the motor on state, if the BDC count is less than 2 the BDC count is incremented. A status request is sent to the extremes detector 111 (FIG. 18) and a BDC report is sent to all machines who have signed up via a start BDC report message as previously noted. When a downside signal (I, FIG. 19) is received from the derivative detector in the motor on state a TDC or top dead center relative to the outer end of the walking beam report is sent to all who have signed up for such a report. A downside message is also sent to the stroke area calculator 110 (FIG. 18). When an extremes message (J, FIG. 19) is received from the stroke extremes detector 111 (FIG. 18) in the motor on state an extremes message is sent to the stroke area calculator, a status request is sent to the stroke area calculator, and a peak report is sent to all of the state machines who have signed up if the BDC count is at least 2. When an area report (at K, FIG. 19) is received from the area calculator in the motor on state an area report is sent to all state machines who have signed up if the BDC count is at least 2.

The stroke derivative detector 109 (FIG. 18) identifies the maximum and minimum stroke positions by using the zero crossing of the first derivative of the stroke signal (FIG. 16) from the stroke transducer 53 (FIG. 1). The first step in the operation is to determine a dead band or noise band about the zero crossing value (X'=0) as seen in FIGS. 16 and 17. A noise value "d" is a maximum difference between X' from the math utility and the X' smoothed by a fifteen point moving average, detected during the 3 second monitor period and corrected for phase shift. The noise band is used to declare that a top dead center (TDC) position has been reached when X' is greater than +d and a bottom dead center (BDC) position has been reached when X' is less than -d. The operation of the stroke derivative detector 109 (FIG. 18) is disclosed in detail in the state diagram of FIG. 20. When the system provides a power on signal (at A, FIG. 20) the derivative detector is initialized and requests a report of X' from the math utility 94 (FIG. 12). The derivative detector also sets a blackout timer to 2 seconds. At this point a subsequent start X' noise measurement signal from the stroke discriminator starts the derivative detector (at B, FIG. 20). A fifteen point moving average smooth of X' is initiated with the last previous value of the derivative used as a starting value and with the maximum noise set to a value of zero.

The start X' noise measurement message signal (at B, FIG. 20) moves the derivative detector into the X' noise monitor state (2). When a X value is received from the math utility it is smoothed. The absolute value of the difference between the smoothed and the raw values of X' is then computed. If this value is greater than the maximum noise value then the maximum noise is set to this value. When a start signal is received from the stroke discriminator (at E, FIG. 20) indicating that the 3 second noise measurement period is over, the X' zero noise band is set (FIGS. 16 and 17). The maximum noise value is then increased by a 10% safety margin and -d is set to -max noise and +d is set to +max noise (FIG. 17).

If the last X' value received is greater than zero then the increasing state is entered. If, however, the last X value is less than zero, then the decreasing state is entered. The derivative detector now monitors the X' values in order to detect the top and bottom of the stroke (FIG. 16).

The operation for the detection of the start of the upstroke (state 3 to 5 to 8 to 4, FIG. 20) is the same (except for the sense of direction) as the operation for the detection of the start of the downstroke which goes from state 4 to 6 to 7 to 3 so only the one detection operation will be discussed herein.

When the stroke derivative detector is in the decreasing state (3, FIG. 20) and a X' value is received from the math utility the X' value is checked against the upper end of the noise band +d. If the X' value is less than +d then no action is taken and the stroke discriminator detector remains in state 3. However, if X' is greater than +d then the signal has gone through the zero X' band in an increasing direction and therefore may have detected the negative position peak (TDC or end of downstroke and start of upstroke). However, it is possible that noise has caused a false detection, therefore a 3 point timer (time needed to acquire 3 data points at the data acquisition rate) is started and state 5 (FIG. 20) is entered. X' values are recorded in this state during the time required to collect the 3 points of data. When this time has expired X' is again compared to +d and if X' is less than +d a noise glitch has occurred. The zero noise band between +d and -d is increased by 10T or by a count of one, whichever is greater, and the stroke discriminator detector returns to state 3. If, however, X' is greater than the value d a negative position peak has been detected. A blackout timer is started, state 8 is entered and a downstroke message is sent to the stroke discriminator 93 (FIG. 18). During the blackout time X' is not checked. Because of the cyclical nature of the pump stroke another peak is not expected until a known minimum time has passed. The use of the blackout time improves the noise immunity of the detector. When the blackout time has expired, X' math flow is started again, the increasing state (4) is entered and the system looks for the positive position peak. The process is the same as above except for the sense of the comparison as noted hereinbefore.

Details of the stroke extremes detector 111 (FIG. 18) which detects Xmax, Xmin, Ymax and Ymin values, is shown in the stroke extremes detector state diagram of FIG. 21. When power is turned on the stroke extremes detector moves into the idle state (1, FIG. 21). In response to a start signal (at B) from the stroke discriminator 93 (FIG. 18) the values Xlag and Ylag math flow are started and the extremes are initialized. In initializing the stroke extremes, Xmin is set to the maximum positive value used in the detector, Y at Xmin is set to the value of zero, Xmax is set to zero and Y at Xmax is set to a value of zero.

The stroke extremes detector (at C, FIG. 21) uses the Xlag signal from the math utility 94 (FIG. 12) to calculate updated values of Xmax and Xmin and uses the Ylag signals (at D, FIG. 21) to calculate the updated values of Ymax and Ymin. The updated values of maximum and minimum for X and Y are calculated as follows. If X received is greater than Xmax then Xmax is set to the X value received and Y at Xmax is set to the corresponding Y value. The same procedure is done for Ymax. If X received is less than Xmin then Xmin is set to the X value received and Y at Xmin is set to the corresponding Y value and the same procedure is followed for Ymin. These values are sent to the stroke discriminator 93 (FIG. 12) in response to a status request (at E, FIG. 21) and the extremes are then initialized.

The stroke area detector 110 (FIG. 18) calculates the total dynagraph card area (FIG. 2) under the direction of the stroke discriminator 93. When a power on message is received (at A, FIG. 22) the status report total curve area is set to a value of zero. When a start message is received from the stroke discriminator the stroke area calculator moves to the "wait for first report state". When a start of upstroke (D) or start of downstroke report (C) is received in the wait for first report state, the appropriate state either 3 or 4 is entered and the parameters are initialized. The buffer index (FIG. 23) and the total area are both set to an initial value of zero and the math flow is started. As the Ylag (load) values are received, these values are processed in the manner determined by the area calculator state (upstroke or downstroke).

Details of the method and apparatus for calculating the total area of the dynagraph are illustrated in FIG. 23 where the load values U1-Un are sampled at regular intervals during the upstroke and stored in memory positions M1-Mn of a load buffer LB1. At the start of each upstroke (FIG. 23) an index I1 is set to zero so it points to memory position M1 of buffer LB1 in the RAM 75a (FIG. 6A) and the total area is set to zero. At regular intervals on the upstroke each of the load values U1-Un are sampled and placed in one of the memory positions M1-Mn of buffer LB1 under the direction of the index I1. The index is then incremented to the next position.

On the downstroke as each of the new values is received, the index I1 is decremented, each of the lower load values Ln-L1 is subtracted from the corresponding upper load values Un-U1, stored in buffer LB1 and the difference values are used to calculate the area of the dynagraph by slicing the dynagraph into small vertical strips, calculating the area of each strip and adding these strip areas to obtain the total area. For example, the lower load value L14 (FIG. 23) is subtracted from the corresponding upper load value U14 and multiplied by the width between boundaries B13 and B14 to obtain the area of the strip A14. Since only the relative areas of the dynagraph between different well conditions are needed the width of each strip can be assumed to have the value of 1, even though the widths of the strips vary from one portion of the dynagraph to another. Each strip, such as strip A14 has substantially the same width each time the load values are sampled.

The area strips (FIG. 23) are shown as being relatively wide to simplify the diagram, but a greater number of load samples, resulting in narrower strips, can be used to increase the accuracy of the calculations. When a strip width of one is assumed it is necessary to merely subtract each load value L1-Ln from the corresponding load value U1-Un to obtain the area of each strip.

The dynamic calibrator 95 (FIG. 12) continuously updates total card area, upstroke average load, maximum load, minimum load, maximum stroke and minimum stroke values to correct for drifting of the load and stroke input signals from load cell 47 (FIG. 1) and stroke transducer 53. These updated values are used by the fluid pound detector, the stroke extremes detector and the rod-part detector to set signal limits, threshold levels and set points to make the system immune to offset signal drift. The calculations are based upon the assumption that offset drifting will not change the shape of the dynagraph curve of FIG. 2, but will only change its position with respect to the zero values. Any drifting up and down and sideways of the dynagraph is used in recalculating the values of card area and the values of Xset, Yset and for recalculating threshold values of rod part detection, maximum load and minimum load detection. The shape of the curve is checked by calculating total area, and values of maximum rod load and minimum rod load. The current area and load values are checked against reference values and if the current values are within a predetermined range the new card is considered to be good and the current values are combined with the reference values to obtain a moving average of updated reference values.

The initial reference values are taken from the first card when the apparatus of FIGS. 6A, 6B is first turned on and the peak load values, area value, etc. are stored in the moving average buffers (not shown) as the initial reference values. Therefore, it is important that the apparatus of the present invention be turned on when the well pump is operating under normal conditions with no pump-off or other problems are present in the well or in the apparatus. At the beginning of each subsequent pumping episode the reference values from the last previous pumping episode are used as initial reference values.

At the beginning of each pumping episode the power on message causes the pump manager software state machine module 91 (FIG. 12) to provide power to the pump motor 30 (FIG. 6A) through an interface 97 and a motor relay 98. A "power on" message to the set point fluid pound detector (FIG. 13) moves this state machine into the "inactive" state. The motor 30 moves the sucker-rod string 16 (FIG. 1) through a predetermined number of start up ignore cycles to allow the fluid level in the well to stabilize, then the pump manager module 91 (FIG. 12) sends a "motor on" message to the fluid pound detector 92 which moves the fluid pound detector (FIG. 13) from the "inactive" state to the "wait for reference" state or the "monitor BDC wait" state. The fluid pound detector uses values of Xmin, Xmax, Ymin and Ymax from the previous pumping episode and the latest operator selected values of X% and Y% to calculate Xset and Yset using the formulae:

Xset=(Xmax=Xmin)(X%.div.100)+Xmin

Yset=(Ymax-Ymin)(Y%.div.100)+Ymin

All values of Xset, Yset, Xmax, Xmin, Ymax, Ymin and the dynagraph calibration (card) area are stored in RAM 75a (FIG. 6A).

On the initial pumping episode there are no previous card values to compute a set point so the fluid pound detector 92 (FIG. 12) moves to the "wait for reference" state (FIG. 13), where it waits for a "references" message from the dynamic calibrator 95 (FIGS. 12, 13). When the fluid pound detector receives the "references message" an initial set point Xset, Yset is calculated from information in the message and fluid pound monitoring begins. After the initial set point is calculated the fluid pound detector (FIG. 13) uses previous episode values of Xmax, Xmin, Ymax, Ymin and the operator selected values of X% and Y% to compute Xset and Yset, and moves into the "monitor BDC wait" state.

If a "references" message is received before a "motor on" message is received, i.e., during the motor start up strokes, a set point is computed when each message is received (FIG. 13) and the fluid pound detector moves to the "wait for motor on" state. When the fluid pound detector is in the monitoring states of monitor BDC wait state, monitor downstroke and Y test, the set point will be adjusted according to each "reference" message from the dynamic calibrator 95 (FIG. 12), thus compensating for offset changes in stroke and load signals. When the values of Xset and Yset have been obtained, the monitor period (FIG. 13) is started on the next downstroke of the pump rod 16 (FIG. 1) because calibration is not recommended when the area of the dynagraph is reduced.

The above calibration technique permits the set point (Xset, Yset) to be updated to compensate for drift in characteristics of transducers and to slowly changing well conditions, such as a change in fluid level due to water flooding, but prevents the set point from changing due to a pump problem or to a high fluid level resulting from a power outage or from workover of the well. Any sudden change in area of the dynagraph curve would probably be due to pump-off or to pump problems which could further damage pump equipment and such sudden changes should be detected as problems. These problems might not be detected if the set point (Xset, Yset) changed positions relative to the dynagraph.

After the set point detector (FIG. 12) has calibrated itself, it begins to monitor the well for fluid pound during the pump downstroke using the stroke (Xlag) and the load (Ylag) values received from the math utility 94. As each current value (Xc, Yc) is received the last previous value Xl, Yl is stored in the RAM 75a (FIG. 6A) and these values Xc, Xl, Yc, Yl are used to interpolate the values between monitored points (FIG. 5) to obtain a true value of Y at Xset. This is necessary as the periodic time sampled checking of the values of X and Y may not obtain a reading exactly at the point Xset. When a current value of X is less than Xset (FIGS. 2-5) the next value of Y (Yc) is used with the previous Y value (Yl) to obtain a value of Y at Xset. If Y at Xset is greater than the value Yset (FIG. 2) a violation count is incremented. When the violation count reaches a predetermined number, a "pump-off detected" signal is sent to the pump manager 91 (FIG. 12).

When the calculated value of Y at Xset is less than or equal to Yset the violation count is set to zero to insure that a specific number of consecutive violations are obtained before the pump-off detected signal is sent to the pump manager 91 (FIG. 12).

Although the best mode contemplated for carrying out the present invention has been herein shown and described, it will be apparent that modification and variation may be made without departing from what is regarded to be the subject matter of the invention.