Easy as APC
27 Oct 2004
Today's global markets are extremely competitive, with conditions that continue to grow more intense as new capacity is introduced to the market. The profitability challenge for manufacturers is to reduce operating costs, while maintaining manufacturing flexibility.
To accomplish this, manufacturers must be able to translate their business objectives and goals to the manufacturing level. A key component to achieve this is a properly designed automation system.
Consequently, many companies are enhancing their current control systems with advanced process control (APC) and optimisation technologies.
APC makes plants flexible enough to take advantage of opportunities in the market place, and better aligned with the organisation's business objectives. This increases product yield, improves product quality and minimises operating expense and an enhanced manufacturing flexibility by minimising operation expense.
APC can improve control and optimisation of a plant operation in three main areas:
1. ability to predict and control product quality immediately on site;
2. reducing process variability, allowing operators to push processes to their constraints, thus increasing production rate;
3. improving operation consistency, meaning that grade transitions are performed more consistently. The best and quickest transitions are repeated more frequently, reducing transition losses.
This improved stability and control will also result in operation efficiency, which may be measured in reduction in catalyst use or reduced loss of raw material.
A standard APC application consists of the following components: a virtual online analyser (VOA) that predicts key properties; a reactor controller that controls key process parameters such as pressure, temperature, percent solids, and gas compositions (for gas phase processes this is also called a gas composition controller); a quality controller that controls key properties by sending updated set points to the controller; and a transition control facilitating grade transitions. For more complicated transitions, where complex sequencing is required, a sequence controller will also assist these solutions.
Most refining and petrochemical operations are, to some extent, non-linear processes. However, operating within a very narrow range of process conditions, process models are assumed to be linear.
Control and optimisation is possible using a linear multivariable predictive control (MPC) solution and benefits are achieved mainly through improved quality control and ability to operate closer to process constraints.
Polymer processes, by contrast, are often very non-linear. Most polymer units produce a number of products. As a result, the focus of operations is not only to maintain the quality of the product during in-grade operation, but also to make the transition from one product grade to the next grade. Since polymer plants are operated over such a wide range of conditions, the traditional linear advanced control tools do not work well for them.
It is not surprising, therefore, that the widespread growth of advanced control and optimisation solutions over the last three decades did not spread to polymer producers. These first-generation solutions were simply the wrong tools for the job.
The key for a successful control and optimisation solution is the quality and accuracy of the process models used for prediction and control. This is especially true for polymer APC projects that are complex non-linear processes with limited information available about these processes.
The Pavilion research team has developed and patented an approach that maintains the computational efficiency of the neural networks for online optimisation and ensures a predictable gain profile for the model using Extrapolating Gain-Constrained Neural Networks (EGCN).
EGCNs were developed specifically for prediction, control and optimisation of industrial processes. The EGCN's approach to model development and implementation procedures are safe and efficient mainly because of: accurate gain representation within the models; imposing constraints on the gain behaviour in the extrapolation regions; and external hard constraints on gain sign and values as a failsafe measure.
Figure 1 above provides an overview of an Extrapolating Gain-Constrained Neural Networks (EGCN) model composed of a set of nodes and functions. In this structure, x0 is an affine function of the node inputs hi, and f(x0, r0) is a parametric non-linear mapping from x0 to the output of the node, h0. These nodes are interconnected in an arbitrary flowing network.
The software implementation of Pavilion's Process Perfecter offers various options for the configuration and training of EGCN models used within the NLMPC module. These options allow the user to select the most appropriate architecture and training methodology for a particular application and give rise to a family of EGCN-based techniques.
Some of the options relevant to this discussion include:
Activation function: while the default activation function in the EGCN model is a traditional sigmoid ('e'-shaped) function, user-defined activation functions can be specified to provide additional flexibility in matching the characteristics of the problem.
Network topology: the topology of the interconnected nodes can be customised to support configurations other than the usual cascaded-layer topology. For example, linear nodes can be included in the hidden layer to affect global extrapolation behavior as shown in the simulation experiments.
Synthetic extrapolation data-points: the automatic generation of synthetic data-points can be configured. These points, which have no target output values, are defined within the extended operating region to control interpolation and extrapolation behavior in the presence of sparse data.
Extrapolation logic: outside of the established operating region of the model, the response surface of the model can be complemented by principled logic. For example, the response surface can be linearly extrapolated over the global input-space using gains computed by the model at the boundaries of the operating region.
These solutions have been demonstrated on nearly every process technology and major catalyst type. As a result, manufacturers have an increased level of flexibility that enables them to respond more quickly to changing customer demands and market conditions. By deploying technology that more tightly aligns production performance with dynamic business goals, Pavilion is helping manufacturers exercise greater control over their processes and their bottom line.
Derek Horn is senior business development manager at Pavilion Technologies, which provides APC and environmental compliance solutions.
