After downloading an MPC function block to a DeltaV controller, verify that the transmitters and valves that provide the inputs and utilize the outputs of the MPC function block are working correctly. Also, make sure the DeltaV Historian is enabled and the area containing the module has been assigned to the DeltaV Historian. Once you have verified all I/Os to the MPC function block, you can use the Predict application to commission the MPC control. You can select the Predict application by right-clicking the MPC function block when the module is online. In response, the Predict application is launched and the following interface appears.
The control (CNTRL) parameters are trended automatically. Up to six of the MPC inputs and outputs can be trended at one time. To add or delete a parameter from the trend, double-click the parameter in the Operate Trend pane. When you right-click a selected parameter, options are provided that allow you to modify the trend view (that is, change the trend range). You can use the toolbar controls or the slider bar to adjust the time window that is shown on the trend.
The process dynamics may be determined by the Predict application using an automated testing feature. Before such testing can be done, the blocks wired to the manipulated outputs of the MPC block must be in the correct mode. By selecting the MPC button in the Control pane, you can force all the downstream blocks to the right mode. After doing this, note that the actual mode of the MPC function block changes from IMAN to MAN.
indicates no change should be made. Enter the longest time (in seconds) that it takes for the process to respond to changes in inputs as the Time to Steady State. With the downstream blocks in their remote mode and the Time to Steady State specified, select Test to initiate automatic testing of the process. In response, the manipulated parameters will automatically change in a pseudo-random manner over a time approximately equal to six times the time to steady state.
If any process output exhibits integrating response, use the following guidelines to estimate the Time to Steady State (Tss):
1. For integrating responses with insignificant lag and/or processes where the non-integrating response is significantly slower than integrating, limit the Tss to the approximate time it takes for the integrating process variable to reach one half of its operational limit value(s), when a normal change is applied to the process input.
2. For processes with significant lag, the recommended Tss is 120 * approximate lag, provided the limiting condition of guideline 1 is not violated.
3. If the identified response is still not satisfactory or the slow dynamics of non-integrating responses necessitate violating guideline 1, estimate the response based on the physical (for example, volume of a tank) and operational process parameters (for example, flow rates, low/high tank level limits, and so on). Then enter the response parameters (Gain per second, dead time, and first order lag, if any) in the corresponding fields of the Step Response Design dialog.
Whenever feasible, change disturbance variables manually during testing with a variable period (approximately Tss) squarewave.
While testing is active, the actual mode of the MPC function block changes to LO. During testing, a manipulated parameter can be changed using the Operate Trend panel. The portion of testing that is complete is indicated by a progression bar. The estimated time remaining to complete the test is shown below the progression bar.
Upon completion of the test, a green area automatically covers portions of the trend associated with the time that testing was active. You can adjust the time covered by the green area either by dragging the start and end divider bars or positioning the bars by right-clicking the chart. The area of test data that reflects normal process operating conditions is indicated by dragging the green area over the data on the trend window. You can exclude data within the green bar that does not represent normal operation by right-clicking within that area and stretching the resulting red bar over the bad data. After you select the test data, indicate if the control or constraint parameters are integrating and then choose Autogenerate to create a step response model for the process and the associated control.
Note The module that contains the MPC block should not be open in Control Studio when you request Autogenerate since the MPC function
block is updated during autogenerate. To utilize the generated control, you must first download the associated module. The process model that is created is shown automatically in the DeltaV Predict application screen. Double-clicking gives you a more detailed view from which you can modify your step response, as shown in the following figure.
Note It is recommended that system time adjustments are not made over the data collection period. If it is necessary to make system time
adjustments, it is best to exclude the data over the time range.
Any disturbances included in the control should be closely examined. The associated step response might be inaccurate if an input did not change significantly in the data used for autogenerate. If this occurs, a warning message will be displayed when the model is generated. In addition, this will be indicated in some cases by the step response that is shown as a straight line (that is, no response). One means of
determining if the step response is accurate is to compare the ARX model to the FIR response. This can be done by identifying yourself as an expert by selecting Options | Expert and choosing the FIR check box. Both responses will be displayed, as shown in the following figure.
Note Only the first 40 points of the FIR response will be displayed.
In some cases, the ARX and FIR responses might not agree. In such a case, you might need to use other data that includes times when the input changed. Then, you will need to regenerate the model and the controller. Alternatively, if you are familiar with the process dynamics,
you can identify yourself as an expert by selecting Options | Expert and then modify the step response to reflect your knowledge of the process.
There are three methods you can use to modify the step response. The first method entails selecting the step response to be modified, right- clicking the response area, and then adding points. Using your mouse, click the new points that define the desired step response, as shown in the following figure. Click Plot to display the new step response. Select Save to save the new step response.
If you are familiar with the process delay and time constants associated with inputs, you can use the second method of modifying the step response. For this method, enter the response directly by selecting Design Response (as shown in the following figure) and select Apply when you are finished to save the new step response.
Also, by selecting Edit Coefficients, you can change individual points in the step response. Select Save when you are finished to save the new step response.
Make sure that you clean up any residual step responses with gains that are ten times smaller than the highest gains with irregular shapes. Once you have modified the step response, you can examine the accuracy of the model by selecting Verify from the model overview. In response, you can examine the calculated output based on the model (as opposed to the actual process output) either for the time frame defined by the green area of the trend plot or for the original data used in generating the model. In response, you can examine the actual process output in direct comparison (as opposed to the output based on the calculated model) either for the time frame defined by the green area of the trend plot or for the original data used in generating the model. You can also examine the XY plot of the actual process output against the calculated output and the corresponding best fit line. In addition, you can use statistics that are provided to indicate how closely
the model matches the plant data. The Model Verification window is shown in the following figure.
Note Only the first 65000 samples will be displayed.
Once you are satisfied that the model accurately reflects the process dynamics, select Generate Control to update the MPC configuration for the new model. When the control is generated from a selected model, control parameters used to generate the control are displayed.
In most cases, it is recommended that you leave the control generation parameters at their default values and select Start (in the Controller Generation window) to generate the control. If you have expert MPC knowledge, you can change the default values to address special requirements. For example, you can increase the Penalty on Move to make the control more robust. Decreasing the Penalty on Move makes
the controller more responsive. Doing so may also be advantageous for unmeasured disturbance compensation. Setting the Penalty on Move to less than the default value would require you to take extra precaution when commissioning. Decreasing the PM might be necessary if the controller is primarily intended for disturbance compensation. To get the proper speed of response, you may need to set the PM to ten times less than the default values. This may cause oscillatory control for some CVs. If an oscillatory CV does not require tight control, decreasing the Penalty on Error for that CV should alleviate the problem. If not, you must increase the PM until stability is achieved.
Qualitatively, increasing the PE decreases the sum of error-squared for the associated CV; increasing the PM decreases the sum of the square of MV moves. Generally, increasing the PE decreases the stability margin, and increasing the PM increases the stability margin. The specific effect achieved is very much dependent on the step responses of the process; so, caution must be exercised. If the objective is tight regulation or load response for a particular CV, and the PE on another CV, which is affected by MVs that affect the target CV, is decreased, then the stability margin for both CVs will increase. The additional margin makes it possible to decrease the PM on one or more MVs that affect the target CV, thereby improving its control response. The relationship of the PE to CVs can be viewed as a relative weighing of error
significance. Since error is evaluated on the basis of percent of span, the same PE value can have different meanings, depending on the span of the CV. For example, the PE applied to a pressure with a span 600 psi differs in units of psi than the same PE applied to a pressure with a span of 300 psi.
For the new control definition to be used after completing control generation, you must download the associated module. All MPC parameters have assigned defaults. You can verify and refine the MPC configuration and control generation during simulation. During online operation of the MPC function block in Control Studio, you can adjust only those parameters of the block that do not impact the controller. Also, you can only change parameters that are used during Control Studio controller generation in offline mode. Any changes in these parameters require that you regenerate the controller in order for the changes to be used in the MPC control.