Automated closed loop step testing of process units

Details Automated closed loop step testing of process units Automated closed loop step testing of process units

2001-07-12

2004-11-16

Knight, Anthony (Department: 2121)

Data processing: generic control systems or specific application

Generic control system, apparatus or process

Optimization or adaptive control

C700S028000, C700S031000, C700S032000, C700S033000, C700S034000, C700S038000, C700S067000, C700S051000, C700S052000, C700S053000, C700S073000, C700S074000

Reexamination Certificate

active

06819964

ABSTRACT:

BACKGROUND OF THE INVENTION
Multivariable Process Control (MPC) algorithms, e.g., Dynamic Matrix Control (DMC), require sufficiently accurate dynamic models of the process unit to ensure high performance control and maintain closed loop stability. The accuracy of the model places an upper limit on the obtainable closed loop performance of the multivariable control system. However, there is a finite limit imposed on the obtainable model accuracy. This is due to the approximation error introduced by representing the real process, which is often non-linear, with linear models, and the ability to identify the model through a system identification process based on observed process data that is usually corrupted by noise and disturbances.
In general, the most cost-effective way to derive accurate models of a large-scale process unit, is to vigorously perturb the process unit with suitable test signals without exceeding safety or operability constraints. The process perturbations have to cover the full amplitude and frequency range of the unit. Several different types of test signals can be used, including steps, pulses, random white noise sequences, of pseudo-random-binary (PRBS) signals. In the process control industry, step test signals are widely used because it is easy to generate these signals manually, and the procedure is referred to as step testing. For the purposes of this discussion, perturbing a process unit with the intent of identifying an empirical dynamic model, is referred to as step testing, whatever test signals are used.
Step testing consists of making sufficiently large orthogonal and independent step changes in all the manipulated variables (MV's) of the process unit under careful supervision. Manipulated variables are those that are adjusted through actuators coupled to respective control valves, reactors, pumps/compressors, etc. forming the process unit and are for example feed rates, flow rate, temperature of a vessel, and the like. The step test data is then used in system identification algorithms to fit empirical dynamic models to the observed process responses. In order to minimize the duration and consequent cost of the step test, these step changes must be of sufficient amplitude to clearly observe the dynamic behavior of the process and maximize the signal to noise ratio. Correlation (dependence) between the MV's has to be minimized to ensure that accurate models can be identified.
Model accuracy results from using large step changes, ensuring minimal correlation between MV's and minimal feedback correlation, and ensuring that the step test sequence spans the full frequency range from very fast to very slow steps relative to the Time-to-Steady-State (TTSS) of the process. Unwanted feedback correlation results from the need to make frequent correcting moves in the MV's to counteract the effect of large unmeasured disturbances, and can degrade the accuracy of the model.
Control valves must also be prevented from fully opening or closing (valve saturation), and tank levels must be kept within the range of the level measurement devices. The fast high frequency dynamics of the process model are important to ensure high performance closed loop control. The slow low frequency dynamics (or process gains) is important to ensure accurate prediction of the future steady-state operating point of the process. This ensures that the optimizer built into the MPC algorithm will determine the most economically optimum steady state targets for the various process variables, and the MPC control algorithm will maintain the process close to the optimum targets, resulting in substantial economic benefit.
It is also important to introduce enough large steps to ensure that the identification algorithm can average out the effect of unmeasured disturbances. The duration of the test is a direct result of the frequency content of the process output signals resulting from the test signals, relative to the frequency content of the process outputs resulting from unmeasured disturbances. Where the process model matrix has a dominantly diagonal model structure (i.e., several units connected in a series structure), independently perturbing several or even all inputs simultaneously can shorten the test duration. Essentially, the signal to noise ratio of every CV (controlled variable, e.g., temperature, pressure, composition, product properties, etc.) in every frequency range of interest has to be maximized.
A significant part of the cost of implementing MPC on major process units, is the cost associated with using highly trained control engineers to supervise the unit while step testing is in progress. The project team often has to supervise the unit on a 24 hours per day, 7 days per week basis to ensure that the step changes do not cause the process unit to exceed safety or operability constraints. Full supervision greatly increases the cost of implementing MPC on large process units with a large MV count, and/or a long time to steady state. The need for an automated algorithm to conduct the step testing of the process unit while ensuring safe operation and keeping all the products within quality specification, while guaranteeing good identification results, has been recognized for a long time and will provide a substantial competitive advantage to its inventor.
Previous Approaches
Several approaches have been used before and are described in the academic literature. Some are summarized next.
Manual Step Testing: Essentially, two or three highly skilled process control engineers working shifts around the clock introduce manual step changes usually in one independent variable at a time, while supervising the unit around the clock. Any unacceptable deviations in the dependent variable are corrected for by introducing additional steps to move the process back to the safe operating region (correcting moves). If the process control engineers are highly skilled, then this approach can provide acceptable data and sufficiently accurate models, but this is not always easy, and it can be very expensive. However, there is a natural tendency to make changes in a fixed order and to respond to process disturbances by making correcting moves. This inadvertently introduces correlation into the MV sequence and makes the model identification problematic. In practice, it is quite difficult to prevent valve saturation and loss of tank levels, the manual step test sequence does not usually have sufficient high frequency content, and step changes are kept small enough to prevent large deviations in the CV's to reduce the risk of constraint violation. At present, this method is widely used in the process control industry.
Using a Programmed Step Test Sequence: This method relies on a sequence of carefully designed programmed step changes in every independent variable around a pre-defined average value, with the ability to manually adjust the average value and step size, or low and high limit values. Typically, the control engineer will choose a sequence based on process insight and good engineering practice to excite the fill frequency range of the process and ensure independence between the step test sequences. This method requires less intervention from the process control engineers once the sequence has been set up, but it still requires careful supervision, as the control engineer has to monitor the process closely, and move the average values when constraint violation occurs. High frequency content can be improved using this approach, but preventing valve saturation is still difficult. Once again, step change amplitudes are kept small enough to prevent excessive constraint violation, and it is still difficult to preventing loss of levels especially if automatic level controllers have to be disabled. This method can provide some improvements in terms of frequency content and reduced correlation, but does not reduce the cost of the project as full supervision is still required.
Using Pseudo Random Binary Sequences (PRBS): A PRBS sequence is automatically generated for every indepe

Affiliated with

Also associated with

Rating

Automated closed loop step testing of process units does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Automated closed loop step testing of process units, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Automated closed loop step testing of process units will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3299843