APC PROJECT STEPS

This is a very high level summary of the steps involved in a classical procedure for advanced process control application. 1. Define objectives: Identify control objectives of the controller. This would involve a collaborative effort with planning & scheduling and operations. It is very important to include process economics to understand the business objectives of the unit. 2. Preliminary design: Propose a first pass controller design in terms of manipulated, feedforward and controlled variables based on design guidelines and experience and discussions with operations. It is a good practice to involve an operator at an early stage of the project. 3. Pre-Test: This step is very necessary to identify any instrument problems, tune regulatory PID Loops and identify any regulatory loop changes necessary for the success of the APC application. 4. Step test: Perform plant tests in which manipulated and where possible feedforward variables are stepped sufficiently to obtain good quality response data. This is a very important step as one is sitting at the operator console and TALKING to the operators. It is very important to learn the unit from them. Some of the experienced operators have seen the unit is various conditions of calm and rock and roll. 5. Identification: Identify a linear empirical model based on the step data. Build different cases with different variables as inputs. Depending on the software that you are using to identify models either SISO (Single Input and Single Output), MISO (Multiple Input Single Output) or MIMO (Multiple Input Multiple Output) models should be identified. 6. Linearisation: Improve model fit by applying nonlinear transforms on selected variables and iterating to step 4. 7. Steady state gain analysis: Extract the steady state gains from the empirical model into a matrix, which is then tested for conditioning. In the case of ill conditioning (colinearity problems), you have two choices. Either revise the design and iterate to step 3, or manually condition the gain matrix by fixing gain ratios. However, a well-posed control problem with few colinearity issues will always be easier to commission, it will be more robust and will perform better once in service. 8. Final design: Construct the final controller using the empirical model. It is important to evaluate if the models make sense. Put on your process engineering hat and see if the model should exist at all. Take away models with very little influence to improve robustness. 9. Closed loop simulation: Perform off line testing and tuning of the proposed controller by testing it in closed loop against the empirical process model. Introduce model mismatch to test

the robustness of the control scheme. Also Tune the controller to achieve the tradeoff between speed of response and Robustness. 10. Commissioning: Implement and commission the controller, making tuning adjustments online. Turn the controller ON after implementing any local MOC (Management of Change) requirements. This should include operator training so that the operator understands the application as well as knows the correct way to interact with the application, also now the operators interaction with the process is going to be different, instead of making changes to the Setpoint of the manipulated variable PID loops, he/she will set the objectives by giving controlled variable limits in the APC application. Their interaction is at a higher level. 11. Post commissioning analysis: A very important step that is sometimes missed. Did the application fulfill the business objectives of its original design.