Inferential sensors

Inferential sensors, also known as soft sensors, are models that nse readily measurable variables to determine product properties critical to prediction of prodnct/process qnafity. Ideally the soft sensors are continuously monitored and controlled, or moiutored on a relevant time scale. They need to make predictions quickly enough to be used for feedback control to keep process variability to a minimum. 

Property prediction may be done nsing snch rontinely measnred parameters as temperatures, pressures and flow rates when there is snfficient process knowledge to correlate these values to product quality. Sometimes process analyzers snch as spectrometers are used to understand the process chemistry and kinetics, thus providing the ability to nse soft sensors if the tool that helped elucidate the critical process variables is unavailable. 

Model predictive control is concerned with continuous feedback of information with the objective of reducing the variability of product quality by changing the set points (narrowing the range) in a plant control loop. By nsing model predictive control, projections on batch quality can be made and midstream corrections made to keep a batch within the target limits of the process. 

Due to the complexity of bioprocesses, and the lack of direct in-process measurements of critical process variables, much work is being done on development of soft sensors and model predictive control of such systems. Soft sensors have long been used to estimate biomass concentration in fed-batch cultivations. The soft sensors can be integrated into automated control structures to control the biomass growth in the fermentation. 

Several statistics from the models can be used to monitor the performance of the controller. Square prediction error (SPE) gives an indication of the quality of the PLS model. If the correlation of all variables remains the same, the SPE value should be low, and indicate that the model is operating within the limits for which it was developed. Hotelling s 7 provides an indication of where the process is operating relative to the conditions used to develop the PLS model, while the Q statistic is a measure of the variability of a sample s response relative to the model. Thus the use of a multivariate model (PCA or PLS) within a control system can provide information on the status of the control system. 

---

Sharmina R, Sundararaja U, Shah S, Griend LV, Sun YJ. Inferential sensors for estimation of polymer quality parameters industrial application of a PLS-based soft sensor for a LDPE plant. Chem Eng Sci 2006 61 6372-6384. 

The main focus of this chapter, however, is on the tools that can use this sensor information for real-time control. These tools are still largely developmental because inferential control is relatively new to the composites industry and change is dependent on both technical merit and on other changes in the culture of the industry. 

The major problem for control based on material states, however, is the quality control culture that requires that parts be accepted based on adherence to a preset cycle within specified limits. Because state-based inferential control systems could theoretically come up with a new cure cycle every time, this sort of specification cannot be used with such systems. Specifications instead have to be in terms of the process plan used for the cure. The satisfactory completion of a certain cure history without alarm states would be assumed to produce an acceptable part. Once the culture was able to accept that difference for autoclave curing, production costs at the U.S. air force s McClellan AFB Logistic Center were substantially reduced [32], This type of specification could also give material review boards a head start on investigations because they would know that a part did not meet specification as well as what sorts of flaws might result from the deviation. The experience at McClellan is that there are fewer parts to review. It is even conceivable that, with improvements to sensors, much of the current postcure nondestructive evaluation used to verily the quality of parts could be incorporated into the process, building quality in rather than inspecting it in after the fact. 

Some of these challenges are overcome through the use of sophisticated models, complex algorithm application, and inferential measurements or use of soft-sensors. 

Moisture control in drying processes has conventionally been done inferentially by humidity sensors in the discharge air stream, but moisture content sensors at the discharge end of the dryer are preferred. Both amount to feedback control, which responds more slowly than feedforward control. For thick load pieces, the mass transfer time to their surfaces may dictate use of feedforward control by locating sensors within the loads (usually difficult) or earlier in the traverse time within continuous dryers. In view of the dead time of some moisture sensors, locating the control moisture sensor(s) at or nearer the entrance will help improve production, product quality, and energy conservation. 

Instrumentation response times Sensor problems Time-delays Interactions between process states Interactions between process units Cascade strategies New sensors sensor location. Inferential measurement and control Predictive control. Robust controller designs Selection of control loop pairings. Decoupling control Feedforward strategies... 

Seasonal variations, changes in incoming chip quality and other external factors that can affect pulp quality are often beyond the control of the thermo-mechanical pulp (TMP) mill operator. Many internal factors are controllable, however, and could possibly be used to counteract these external forces. The ultimate goal is to model in real time parameters that cannot be measured continuously, in order to apply inferential control C soft sensor ) as reported in Strand et al. (2001) and elsewhere (Kooi, 1994 Kresta et al., 1994). Before proposing any such control strategy, however, it is necessary to understand the correlations and trends which are inherent to the refining operation at the heart of the pulp mill, using historical data. 

There are literally thousands of inferential properties, so called soft sensors , in use today that are ineffective. Indeed many of them are so inaccurate that process profitability would be improved by decommissioning them. Chapter 9 shows how many of the statistical techniques that are used to assess their accuracy are flawed and can lead the engineer into believing that their performance is adequate. It also demonstrates that automatically updating the inferential bias with laboratory results will generally aggravate the problem. 

Although many process variables are easily measured, lack of on-line sensors for key polymer properties renders quality control of polymer plants difficult. Process control schemes based on process variables p, T, flow-rate and feedstock compositions) alone are no longer sufficient, because these cannot reveal all material property variations. Significant efforts are being spent on improvements to process control systems, as exemplified by the numerous attempts to monitor polymer properties during processing, such as composition, density, viscosity and dispersion of a minor phase, etc., all of which are somehow difficult to measure. The development of an on-line inferential system for polymer property is a very active research area of polymerisation reactor control [1]. A schematic of inferential systems is illustrated in Fig. 7.1. For highest quality... 

Additionally, online monitoring methods have been developed to adapt off-line characterization methods into in situ (i.e., in-reactor) probes for determination of kinetics and monomer conversion with optical methods such as mass spectroscopy (MS), ESR, FTIR, near IR, and Raman spectroscopy. However, frequently, due to high turbidity and viscosity of the polymer reaction milieu, the optical surfaces are easily fouled, leading to frequent sensor failure. Furthermore, data acquired with these probes are model dependent the empirical and inferential calibration schemes used can be expensive and time consuming to develop and can drift and become unreliable as reactor conditions change and as sensors become fouled. Another limiting feature of these methods is that they usually measure only one characteristic of the reaction, such as monomer conversion and are not directly sensitive to polymer molecular mass and intrinsic viscosity. More detailed discussion of these techniques can be found in Chapters 6-10 of this book. 

In general, CVs are measured on-line, and the measurements are used for feedback control. But sometimes it is possible to control an unmeasured CV by using a process model (a soft sensor) to estimate it from measurements of other process variables. This strategy is referred to as inferential control (see Chapter 16). 

One solution to this problem is to employ inferential control, where process measurements that can be obtained more rapidly are used with a mathematical model, sometimes called a soft sensor, to infer the value of the controlled variable. For example, if the overhead product stream in a distillation column cannot be analyzed on-line, sometimes measurement of a selected... 

Figure 16.12 Block diagram of a soft sensor used in inferential control.

---