Diagnostic method and device for monitoring the functioning of a control loop

The invention relates to a diagnostic device for monitoring the operation of a control loop of an automation system according to the preamble of claim 1 and a corresponding diagnostic method according to the preamble of claim 5.

Maintenance and servicing of automation systems can be improved if the correct function of subsystems or components is monitored. If the functionality deteriorates, measures for maintenance, repair or troubleshooting can be taken at the correct point in the system.

From the <patcit id="pcit0001" dnum="DE10007972A1"><text>DE 100 07 972 A1</text></patcit> a diagnostic tool for use in a process control system is known that detects problems or identifies malfunctioning facilities and circuits. A process control system is described which contains a centrally arranged process control device which is communicatively connected to at least one operating work station and one or more plant devices via analog, digital or combined analog digital bus lines. The system devices, which can be, for example, valves, valve positioning devices, switches and sensors (e.g. temperature, pressure or flow transducers) perform functions within the process, such as B. Opening or closing valves or measuring process parameters. The process control device receives signals which indicate the process measurement values ​​obtained from the plant devices and / or other information which are related to the plant devices. It uses these signals to implement a control routine and generates control signals that are sent to the equipment to control the flow of the process. Information from the plant equipment and the control device is made available to one or more application programs that are executed by the operator work station in order to enable an operator to perform a desired function with regard to the process, for example viewing the current process status, Modify the operational flow of the process, etc. To improve the structuring of the control task, function blocks are used, each of which receives input from other function blocks and / or outputs output signals to other function blocks. They each carry out a specific part of process control operation, for example measuring or recording a process parameter, controlling a device or the task of a PID controller. The well-known diagnostic tool automatically collects and stores data that belongs to the various functional blocks of devices and circuits within the process control system, and processes this data to determine which components of the process control system have problems that can lead to reduced performance of the system. To diagnose problems, the diagnostic tool uses a spread indication that is, for example, based on the differences between a historical statistical state of a device over a predetermined period of time, for example the average position of the valve of a valve device, and the current state of the device, in this example the current one Position of the valve can be calculated. As a further example, an application for a control function block describes a stochastic feature that is based on a deviation between a process parameter input of the function block and a setpoint value that is specified for the function block for this parameter. A disadvantage of the known diagnostic tool with evaluation of a stochastic feature of process parameters is that a process that is stationary in the statistical sense is required. It is not possible to diagnose the transition behavior, for example during adjustment processes. Because here every setpoint jump of a controller violates the requirements of a stationary process to a considerable extent and leads to incorrect statements when evaluating such stochastic features.

From the <patcit id="pcit0002" dnum="US6192321B1"><text>US 6 192 321 B1</text></patcit> A diagnostic method for a process controller is known which regulates the course of a process in a control loop. During process operation, the diagnostic device monitors whether the process is in a steady state. If this is the case, the diagnostic device replaces an actuating signal from a controller with a test signal, which is thus fed to the actuator instead of the actuating signal. The test signal stimulates the actuator to carry out a deterministic sequence of operations that are designed in such a way that they only minimally influence the process. The diagnostic device detects the response signals that arise in the process after being excited by the test signal. With a suitable choice of the test signal, the width of an insensitivity zone, a response time, a gain or an overshoot can be calculated by analyzing the response signals. However, this diagnostic method has the disadvantage that the normal course of the process is disturbed, even if this disturbance is minor or only of short duration. In addition, the entire control loop is not monitored by the diagnosis, but only the part to which the control signal of the controller is given. Another disadvantage is the fact that no diagnosis of a control loop can be carried out in the steady state without disturbing the process in any way

From the <patcit id="pcit0003" dnum="US20020040284A1"><text>US 2002/0040284 A1</text></patcit> a method and a device for determining and assessing instabilities in control loops are known. For this purpose, the device has a memory in which time sequences of different sizes of the control loop are stored for later evaluation. To determine instability, for example, the variance of a controlled variable is calculated as a stochastic characteristic and compared with a reference value. For the subsequent investigation of the cause of the error, an evaluation device calculates, for example, the phase shift caused in the individual elements of the controlled system as a deterministic feature.

The invention has for its object to provide a diagnostic device and a diagnostic method with which the operation of control loops in a process engineering system can be virtually permanently and automatically monitored.

To solve this problem, the new diagnostic device of the type mentioned has the features specified in the characterizing part of claim 1. Further developments of the invention and a diagnostic method are described in the dependent claims.

The invention has the advantage that it enables monitoring of control loops and all components of a process engineering system involved in the respective control loop to be quasi permanently enabled, that is to say both in the steady state and during compensation processes. It is therefore possible to intervene more quickly with the elimination of the performance of individual control loops at the right point in the system with measures to rectify or improve the situation. In addition to the diagnosis of a control loop in the steady state, a diagnosis of the transition behavior in the event of setpoint jumps is advantageously possible. In the case of an excitation of the control loop essentially in accordance with a setpoint step change, in particular in the case of an exact setpoint step change, the relative overshoot range and / or the settling ratio, ie the quotient of the rise time and settling time, are determined as deterministic features. This has the advantage that both features allow very direct and diagnostically evaluable conclusions to be drawn about the functioning of the control loop and the control quality. Both features are dimensionless and independent of the application in which the control loop is operated. Therefore, no application-specific calculations or empirical measurements are required in advance for the predetermination of a reference value with which a value of the respective feature determined in a diagnosis during normal operation is compared. Incorrect diagnoses, which could occur in previous diagnostic devices with evaluation of the variance of different process parameters in the event of setpoint changes, are also largely avoided. Another advantage can be seen in the fact that, in the sense of a non-invasive diagnosis, only the data generated in regular process operation are used. The process is therefore in no way influenced by the diagnosis. The control loop in an automation system represents a functional level on which impending errors can be detected particularly quickly and reliably. The control loop has a clearly defined task and a defined behavior after commissioning. Deviations from this target behavior can be recognized by signal processing without additional sensors and usually allow the localization of error sources within this control loop down to the individual components of the transmitter, controller, actuator and controlled system.

In the case of an essentially steady state, it is advantageous to determine the variance of the sequence of actual value data as a stochastic characteristic and to compare it with an associated reference value which has been predetermined and stored in a well working state on the same control loop. This leads to an easy detection of faults in system components that are in the control loop. In addition, a better meaningfulness of the diagnosis is achieved, since a reference value measured on the real control loop is used for the comparison.

Advantageously, an increase in variant due to decay processes of the variables of the control loop after prior excitation by a change in the setpoint value is ruled out as the cause of a misdiagnosis if the evaluation device for determining a stochastic feature is only reactivated by the selection device if after the start of the setpoint step change at substantially constant values Setpoint data a predeterminable multiple of the settling time of the control loop has passed. The variance of the process variables then again corresponds to that which prevails in the steady state.

The diagnostic device can advantageously be designed as a software function block, which can be interconnected in a graphical user interface of an engineering system with function blocks of the control loop and can be loaded into an automation device for operating the diagnostic device. This function block can be implemented, for example, as a PCS7 function block for SIMATIC controllers from Siemens AG and can be provided with a faceplate for implementing a human-machine interface on an operator control and monitoring device in the automation system.

Based on the drawings, in which an embodiment of the invention is shown, the invention and embodiments and advantages are explained in more detail below.

Show it:<dl id="dl0001"><dt>Figure 1</dt><dd>a block diagram of an automation system,</dd><dt>Figure 2</dt><dd>a schematic diagram of a diagnostic function block and</dd><dt>Figure 3</dt><dd>a timing diagram of a step response to clarify deterministic features.</dd></dl>

In <figref idref="f0001">Figure 1</figref> the basic structure of an automation system is shown. Process variables are measured on a process 1 and read in by an analog input 2. To control process 1, an analog output 3 outputs control signals to process 1 as a function of the measured process variables. The analog input 2 and the analog output 3 are components of a decentralized periphery 4. Via a fieldbus 5, the decentralized peripherals 4 and others, for the sake of clarity, are in the <figref idref="f0001">Figure 1</figref> Field devices, not shown, are connected to a programmable logic controller 6. During the operation of the automation system, a control program, which contains, for example, the function of a controller and a diagnosis of a control circuit, runs in the programmable logic controller 6 on a processor (not shown). The control program is carried out with the help of an engineering system 7, e.g. B. SIMATIC PCS7 designed. A personal computer, for example, can serve as the hardware basis for this. In<figref idref="f0001">Figure 1</figref> A graphic function block diagram 8 is also shown, which is intended to illustrate the design of a control loop with a monitoring device by way of example. Such a function block plan can be graphically edited on the engineering system 7. A controller 11 is arranged in a control circuit 9 behind a subtractor 10 and determines a manipulated variable u from the deviation between a setpoint SP and a controlled variable y, that is to say as a function of a controlled deviation e. This is passed to an analog output module 12, the output variable of which acts on a process 13. The controlled variable y is measured with an analog input module 14 and fed back to the subtractor 10. A control module 15 is assigned to the control circuit 9 and evaluates the setpoint SP, the manipulated variable u and the controlled variable y. In addition to the process variables of the control circuit 9 shown, the diagnostic module 15 can monitor further control circuits, not shown in the figure, by evaluating further process variables - as symbolized by 3 points on the left input side of the module 15. The diagnostic results are output via a faceplate, which can be arranged in an operator control and observation work station 16, which is connected to the same data transmission network 17 on which the engineering system 7 and the programmable logic controller 6 are also located. Alternatively, each individual controller function block can be given its own diagnostic block as an add-on. Overviews of the status of all control loops are then created using the means of the process control system, in which the information from all diagnostic function blocks is brought together. After the control software has been designed, it is loaded into the programmable logic controller 6 and runs there during operation of the automation system in accordance with the function modules used and their connections.

<figref idref="f0002">Figure 2</figref> shows the basic structure of a diagnostic function block. Input signals are the process variables to be monitored, in the exemplary embodiment shown the setpoint SP, the controlled variable y and the manipulated variable u. In addition, further process variables can be monitored and fed to the diagnostic function block 15. A sequence of setpoint and actual value data of the control loop, that is to say a section of the time profile of the setpoint SP and a section of the time profile of the control variable y, is stored in a data memory 20. A selection device 21 analyzes the state of the associated control loop on the basis of the sequence of the setpoint data. In the case of a substantially stationary state, the selection device 21 activates an evaluation device 22 for determining at least one stochastic feature, e.g. B. the variance, the monitored process variables. The evaluation device 22 compares values ​​of the variance determined during operation of the automation system with respectively associated, predetermined reference values. If the respective reference value is exceeded, this can indicate an error in the control loop. If the selection device 21 detects an excitation of the control loop essentially in accordance with a predetermined, in particular step-like, course of the setpoint, it activates an evaluation device 23 for determining at least one deterministic feature and compares this particular value of the feature with an associated, predetermined reference value. When a setpoint jump occurs, the transient response of the step response and / or the relative overshoot range are preferably evaluated as deterministic features. In addition to the two states of the control loop mentioned, there may be other states in which the process variables are not evaluated. The state of the control loop determined by the selection device 21 and the value of a stochastic feature determined in the evaluation device 22 and / or the value of a deterministic feature determined in the evaluation device 23 are stored in a result memory 24 of the function block 15. The output of the values ​​and graphic processing takes place with the help of a faceplate on the display unit of an operator control and monitoring device, which also carries out a trend analysis and displays its results.

With the diagnostic function block 15, both stochastic features of process variables are evaluated and deterministic features are used for diagnosis, which are relevant for assessing the control quality of the control loop. Depending on the operating state of the control loop, the appropriate features are automatically selected. If a setpoint jump is detected on the control loop by evaluating the course of the setpoint SP, two deterministic features are determined on the basis of the time course of the step response of the controlled variable y of the closed control loop: the relative overshoot range and the transient ratio. The relative overshoot is related to the level of the setpoint step. The settling ratio is calculated as the quotient of the rise time and the settling time. Both features allow very direct and diagnostically evaluable conclusions to be drawn about the control quality. When a predefinable multiple of the rise time, for example four times the rise time, has passed and the step response has settled, the control loop is again in a steady state and the evaluation can be switched back to stochastic characteristics mean and variance. Instead of the theoretical minimum variance of the control loop, a variance actually measured in good condition on this individual control loop is preferably used as the reference variance in order to improve the meaningfulness of a so-called control performance index.

Functions for monitoring control loops can be located at different levels within the system architecture of an automation system or a process control system. Aspects for such decisions are on the one hand the need for storage space and computing time and on the other hand the relevant time horizon as well as aspects such as engineering effort as well as operator control and monitoring. The functions which access current data or data from a limited time window of a length n with n &lt;100 can advantageously be accommodated in a diagnostic function block, which can be regarded as an additional block for a controller function block. As far as possible, a recursive calculation for evaluating the data should be preferred, since it is practically possible to dispense with data storage.

When evaluating in function block 15, a distinction is made between deterministic and stochastic features of the process variables. In normal operation, stochastic features are continuously evaluated. However, as soon as, for example, a setpoint jump is detected with the selection device 21, the evaluation device 23 is activated, which evaluates the setpoint jump with regard to deterministic features. A corresponding display bit is set on the diagnostic function block 15. As soon as the settling process has been completed and a certain time has passed, the deterministic features are output and the display bit is reset.

On the basis of the time profiles of the setpoint SP, the controlled variable y and the manipulated variable u, a control deviation e and temporal averages of these variables are calculated. A recursive calculation formula can be used. A control deviation e that remains on average with a constant setpoint is already an indication of problems in the control loop if the controller used is a controller with an I component. If a stationary working point (u₀, y₀) is known, the mean current process gain k can also be calculated from the calculated values_G estimate if you assume that only interference-free interferences intervene: <maths id="math0001" num=""><math display="block"><msub><mi mathvariant="normal">k</mi><mi mathvariant="normal">G</mi></msub><mo mathvariant="normal">=</mo><mfrac><mrow><mover><mi mathvariant="normal">y</mi><mo mathvariant="normal">‾</mo></mover><mo mathvariant="normal">-</mo><msub><mi mathvariant="normal">y</mi><mn mathvariant="normal">0</mn></msub></mrow><mrow><mover><mi mathvariant="normal">u</mi><mo mathvariant="normal">‾</mo></mover><mo mathvariant="normal">-</mo><msub><mi mathvariant="normal">u</mi><mn mathvariant="normal">0</mn></msub></mrow></mfrac></math><img file="EP1528447B1_D0001.tif" /></maths>With <o ostyle="single">y</o> - Average of the controlled variable y and<o ostyle="single">u</o> - Average value of the manipulated variable u.

Depending on the application, typical stationary working points are often known in advance, e.g. B. with a flow control: y₀= 0 for u₀ = 0, which means the valve is closed, or for temperature control: y₀ = y_amb for u₀ = 0, that means ambient temperature y_amb with heating switched off. Stationary working points can often also be determined during initial commissioning. From the approach, the variance as a centered moment requires the calculation of differences between each current value of a process variable and its mean:<maths id="math0002" num=""><math display="block"><msubsup><mi mathvariant="normal">σ</mi><mi mathvariant="normal">y</mi><mn mathvariant="normal">2</mn></msubsup><mo mathvariant="normal">=</mo><mfrac><mn mathvariant="normal">1</mn><mrow><mi mathvariant="normal">n</mi><mo mathvariant="normal">-</mo><mn mathvariant="normal">1</mn></mrow></mfrac><mstyle displaystyle="true"><munderover><mo mathvariant="normal">∑</mo><mrow><mi mathvariant="normal">i</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">1</mn></mrow><mi mathvariant="normal">n</mi></munderover></mstyle><msup><mfenced separators=""><mi mathvariant="normal">y</mi><mfenced><mi mathvariant="normal">i</mi></mfenced><mo mathvariant="normal">-</mo><mover><mi mathvariant="normal">y</mi><mo mathvariant="normal">‾</mo></mover></mfenced><mn mathvariant="normal">2</mn></msup><mn mathvariant="normal">.</mn></math><img file="EP1528447B1_D0002.tif" /></maths>

The scatter <maths id="math0003" num=""><math display="block"><msub><mi mathvariant="normal">σ</mi><mi mathvariant="normal">y</mi></msub><mo mathvariant="normal">=</mo><msqrt><msubsup><mi mathvariant="normal">σ</mi><mi mathvariant="normal">y</mi><mn mathvariant="normal">2</mn></msubsup></msqrt></math><img file="EP1528447B1_D0003.tif" /></maths>The root of the variance can be interpreted more clearly, since it has the same physical unit as the measured value.

A Control Performance Index CPI is calculated as the quotient of a reference variance of the controlled variable and its current variance: <maths id="math0004" num=""><math display="block"><mi>CPI</mi><mo mathvariant="normal">=</mo><mfrac><msub><msup><mi mathvariant="normal">σ</mi><mn mathvariant="normal">2</mn></msup><mi>ref</mi></msub><msubsup><mi mathvariant="normal">σ</mi><mi mathvariant="normal">y</mi><mn mathvariant="normal">2</mn></msubsup></mfrac><mn mathvariant="normal">.</mn></math><img file="EP1528447B1_D0004.tif" /></maths>

The Control Performance Index normally ranges between 0 and 1. If the current spread corresponds to the reference value, the CPI reaches the value 1. If the current spread, however, is much larger, the CPI delivers very small values. Various values ​​can be considered as reference values. For example, the variance in a defined good state can serve as a reference value. After commissioning the controller, this can be logged with a PID tuner. As long as the controller and route parameters remain unchanged, i.e. the good condition is not left, the CPI moves in an order of magnitude of approximately 100% corresponding to value 1. A significant reduction in the CPI value indicates an inappropriate controller setting, which may be B. can be caused by changes in the process with constant controller parameters. With a reference variance defined in this way, it is temporarily possible that the CPI will be greater than 100%.

Another example of a reference value of the variance is the variance that arises with a theoretically ideal PID controller, an explicit process model and an explicit interference filter model being required for the calculation. In practice, however, these requirements are rarely met.

The variance obtained with a theoretical minimum variance controller could also serve as a reference value. This only depends on the process dead time and the fault model. This form of the CPI is known as the Harris index and represents a generally unattainable lower limit for a PID controller, which is why the CPI assumes values ​​between 0% for poorly set controllers and close to 100% for well-set controllers. Low CPI values ​​provide an initial indication that the controller setting can still be improved. However, it should be borne in mind that the minimum variance is only a theoretically achievable value and the minimum variance controller has properties that are not desirable in practical applications, for example extremely large manipulated variables. With a CPI based on the minimum variance, it is therefore not desirable to bring it as close as possible to 100% at any price.

Based <figref idref="f0002">Figure 3</figref> Some deterministic features that are particularly suitable for evaluating the control quality are to be clarified below. A time diagram is shown with the qualitative courses of some process variables of the control loop. The time t is plotted on the abscissa, the setpoint SP, the controlled variable y and the manipulated variable u are plotted on the ordinate. It is a schematic diagram in which units have not been specified, since these depend on the respective application. A setpoint jump from an initial value SP0 to an end value SP1 at a point in time t0 is considered to be an essentially predetermined profile of the setpoint SP. Thereupon, the depicted course of the controlled variable y arises with an overshoot over the stationary end value. The absolute level of the overshoot Ovs_max is related to the level of the setpoint step SP_height and is given in percent. With a good controller setting, the relative overshoot is between 0 and 15%. Overshoots greater than 30% indicate a weak damping of the control loop, i.e. a tendency to oscillate behavior.

Another characteristic for assessing the control loop is the transient response in the event of a setpoint step change. This is calculated as the quotient of a rise time T_rise by a settling time T_settling, the definition of which<figref idref="f0002">Figure 3</figref> can be easily derived. The settling time T_settling indicates the period of time after the jump before the step response finally enters a zone with a maximum distance BW from the setpoint SP1. The settling ratio ranges from 30% to 100% with well-adjusted controllers. Lower values ​​can be caused by a too long reset time of the controller. By evaluating archive data, overall statements can be made for longer periods of time and slow trends or deteriorations can be identified.

In order to clarify the results of the diagnosis to an operator, the current CPI, the standard deviation of the controlled variable y, the mean values ​​of the controlled variable y and the are displayed in a main view of a faceplate on an operator control and monitoring station, for example a SIMATIC-PCS7-OS Manipulated variable u, the stationary gain as well as the relative time shares during which the controller is in manual mode and the relative time shares, in which the manipulated variable u hits the upper or lower limit. In another "setpoint" view of the faceplate, the maximum and relative overshoot range, the settling time and the settling ratio are displayed after a settling process as a result of a setpoint jump. These values ​​are retained after the evaluation and are not deleted until the next jump. The window length required for averaging can be specified in a "Parameters" view. In this view, an initialization of the reference values ​​for stochastic characteristics can also be triggered in this view. The reference values ​​of variance and standard deviation are displayed in the fields provided. Another "Limits" view is used to set warning and alarm limits for the Control Performance Index CPI and the relative overshoot. Corresponding messages are only generated if the CPI falls below a predetermined value or the overshoot range is higher than the predetermined upper limit. It is therefore preferred that only lower limits are specified for the CPI and only upper limits for the overshoot range. In a "Trend" view, calculated temporal profiles of the Control Performance Index CPI, the mean value of the controlled variable y and the standard deviation are output for easier assessment by an operator. According to the SIMATIC PCS7 standard, the "Messages" view lists the messages generated by the diagnostic function block with their date, time, class, status and event description.