Introducing Neural-Max

Raising The Bar

We have all experienced the situation where one operator or one shift is recognized as the top producers. The crew that continuously out achieves their counterparts with no distinguishable change in the overall operation.

What would this mean to the organization to achieve that level of productivity cross all shifts?
What would it mean if even this level could be exceeded?
Without continuously raising the bar any organization invites entropy and decay. That nagging little reduction or acceptance that things were just against me and or I just had a bad shift.
Without the proper tools, operations cannot achieve these increased levels of production. The ability to implement these tools is grounded in an in-depth knowledge of both the process and the operation and demands on the facility.
By working together, we can achieve greater results and or determine exactly what the limitations are and how they are impacting the overall operation.

Binnington Development Corporation has been in the business of combustion and process control for over 20 years. Since our inception in 1986 we have seen many systems developed and new technologies introduced. Some products such as emissions monitoring have been legislated through the establishment of pollution and emissions standards. While being a necessity for the continued operation of a facility they have not, in most cases, contributed to any significant gains in productivity or reliability. We have also experienced a great increase in the abilities of process control and computer system applications available to the plant floor. Even with this increased power and capabilities, in many cases, we discover that some capabilities are left underutilized and potential gains for productivity left unrealized. Binnington has developed a process for the implementation of advanced process controls to capture these previously hidden capabilities. This new level of control can be added to most existing systems to achieve greater productivity and return while addressing any known problem areas. It is time to capitalize on the abilities of the equipment you already posses. Time to discover this previously hidden and/or underutilized potential with Neural-Max.

The Seven Step Process:

We use a seven step process to implement an adaptive optimization control system. Each step follows as a progression to achieve the maximum results. Typically, we see gains with each step.

The level of these gains are very dependent upon the situation and the individual operation. No two applications have the same set of criteria or circumstances. Every application is unique. The process that we have identified below insures that the constraint for each application is properly identified and considered as we move through this process.

STEP 1 The Study

Our first and possibly the most important step is to implement an initial study. The objective of this study is to determine that the desired results are achievable in a practical manner.

primary controls and field devices are in place and functioning
mechanical or process limitations are not being encountered, i.e. fan limits or circulation limits are already maximized and the cost of upgrading would be prohibitive to the project.
communications between the supervisory system and the existing control system can be achieved reliably

Included in this study is the review of historical data if and whenever possible to gauge the seasonal variability. The study also allows us to gauge the returns that can be expected from this project. If the gains are not sufficient to warrant our system, we may identify other areas where the investment would give a better return.

STEP 2 The Schedule

In cooperation with the operations and maintenance staff a schedule shall be agreed upon for the full implementation of the system. Attention will be given to annual shut downs and any tie ins that may be required. Typically, system shutdowns or outages are not required to tie the supervisory control system to the existing control system. However annual shutdowns and turn arounds may occur during the implementation and learning phase of this project and that will need to be addressed.

STEP 3 Process Boundary Definition

This phase of the project has two distinct areas of review.

The first is to determine how far the parameters that will be taken into consideration will come from. The level of impact of each major system will need to be defined; i.e. the impact of large loads such as paper machines starting and stopping.

The second is the limits of the process and how they are going to be coded into the system; i.e. Optimal O2 range for firing on all combinations of fuel. Co and Nox levels both desired continuous operation and maximum levels allowed. Etc.

This segment is also where the I/O points are determined for the process control system. This includes the input and outputs for the model and the control parameters that will be adjusted.

NOTE: End of Step 3 onward, consistent gains in productivity should be seen on the system.

STEP 4 Model Creation

Once the I/O points and the control definition parameters have been determined, the application engineers install the process hardware and implement the communications protocols that will allow our system to communicate. The model and the operator interface are then created and the model is installed on the system.

STEP 5 Adaptive Control Learning Process

Once the model has been developed and installed, the training of the model is started and the monitoring of real time operating data begins. It should be noted that the system learns during regular operation, no specialized training runs or interruptions to the normal operation of the system are required. See the segment below for a brief discussion on the learning phase of this process.

STEP 6 Adaptive Control Commissioning

Once the model has been trained and the predicted output closely matches the realized output, the final phase is to implement the optimization algorithms and apply their outputs to the control parameters. This is the maximization of the process control.

The algorithms as mentioned previously, continuously search for combinations and permutations to the control parameters to achieve better results.

STEP 7 Reporting

This final stage is the development of reports and the final analysis of the overall system performance. Included in this reporting stage is a review of the initial system goals and the determination of their achievement

The Key Component – Non Linear Modeling

Modeled after the human brain’s neural networks which learn through the observation of the process and determine the relationship between inputs and process output. Neural networks mathematically model the process inputs and outputs and through an interactive process create a non linear relational model.

The accuracy of the model is determined by the number of inputs given and the true relationship between the inputs and the outputs. For example: if a critical input such as combustion air flow was not given as an input then the model would not be able to correlate between a change in air flow and the change in steam flow. Alternatively if an input that does not have a true relationship is entered to the output, then again a change to this parameter would not have any real impact on the output.

As one or more inputs change the resultant change in the output is then known and predictable. Another interesting facet of this learning process is the timing relationship between a change of an input and its subsequent change to the outputs.

Using this capability and staying within operational safety parameters, the model can then be used as a basis for achieving maximum productivity and efficiency.

The model can be left in the training mode to continuously monitor the relationships. This allows for seasonal variations such as moisture content variation or load demand variations to be taken into consideration without input from an outside source.

Multiple models can be implemented at any time. The model that is predicting the output most accurately is in control. The models actually compete for control of the system.