FALCONEER TECHNOLOGIES, LLC

Patented Process Performance Software

Making Plants Smarter and Safer

<= span title=3D"http://www.falconeertech.com/home/">WWW.FALCONEERTECH.COM=

YOUR VIRTUAL WORKFORCE™

Real-Time Auto-Generating Process Monitoring Suite:

FALCONEER Technologies FALCONEERTM IV Real-Time Process Monitoring Suite (FALCONEER= ) includes sensor and process condition validation, continuous statistical = process control monitoring, and predictive fault analysis. This software suite th= us provides a single source means for complete, high level monitoring and analyzing all manufacturing & environmental process data in real time= to predict future process performance and optimize current process performan= ce to help improve reliability, yield and quality, avoid failures and accide= nts, and reduce costs. A patented innovation designed into this technology is = the ability to be essentially self-creating and self-maintaining. This feature significantly reduces the cost, time, and resource commitment to implement and to maintain.

The Process Monitoring Suite consists of several software modules with display screens. These include the State Identification (State ID<= /strong>) module, the Sensor Validation and Predictive F<= /span>ault A<= /span>nalysis (S= V&PFA) module, the Virtual Statistical Process C<= /span>ontrol (V= SPC) module and a configuration interface. This application can operate as a standalone system or run as Web service with alarm screens and information screens as Web clients with the capability of OPC messaging to notify pag= ers, cell phones, etc. Below is a summary of the purpose and operation of the different modules that add intelligence to the control systems and instrumentation.

<= strong>State ID:

The software program first determines whether the process is operating within standard conditions or not. This is accomplis= hed with the State Identification (State ID) Module. FALCONEER is idle if the process is not operating. The program begins its analysis of sensor measurements once startup is complete. It continues this analysis until t= he process is shutdown and then is idle again until the next startup complet= es. The process condition monitoring suite thus runs continuously and adjusts= its analysis to current process operations accordingly.

State ID variables are any sensor measurements used = by the program to determine if the process is running. These variables tend = to monitor either key process feeds or process configuration states. They generate green alerts (used to denote normal operation) when the process = is operating and each of these variables is either within its associated standard operating conditions (SOC’s) o= r its normal state. If outside their SOC’s but within their interlock limits, these variables generate yellow alerts. In either case, the program status is “In Production Mode” and “Not Expecting Interlocks”. If outside their interlock limits= or normal configuration, these variables generate red alerts and the rest of= FALCONEER’= s analysis is preempted. In this situation the program status becomes “Not in Production Mode” and “Interlock Expected”. It stays in = this mode until the next startup is completed.

<= strong>SV&PFA:= The Sensor Validation and Predictive Fault Analysis (<= span style=3D'font-family:Verdana;mso-bidi-font-family:Arial'>SV&PFA<= /span>) and alarm management:

The software module monitors current process sensor measurements to determine if they are either valid or incorrect. It also determines if certain processing faults (leaks, pump failures, controller malfunctions, etc.) are occurring or not. It does this by evaluating engineering models that describe normal process operation. Briefly, when = the process system is operating normally (i.e., fault free), the engineering models describing normal process operation should all close. When they don’t this is evidence that something is going wrong in the process system. By looking at the patterns of all this evidence it is possible to infer the underlying fault(s) in the process that could cause such behavi= or. These results are then given as alerts to the users. If not found to be faulty, the sensor measurements are considered validated (i.e., trustwort= hy).

It is very common that the SV&PFA module reports more than one plausible fault at the same time, i.e., many alerts occur simultaneously. This is referred to as the diagnostic resolution of the analysis. In those cases all alerts need to be checked out further to identify the actual process problem occurring. In other words, although e= ach alert given is a plausible explanation of the current process situation, = it still needs to be verified as actually occurring or not by the user. Thus, alerts are meant to identify all the possible operating problems that can explain the process behavior currently being encountered. These alerts are continuously updated as process conditions warrant.

Besides determining single fault alerts, the SV&= PFA module also determines many plausible pairs of fault hypotheses, i.e., multiple faults. However, this is not an exhaustive list of all plausible multiple faults. Furthermore, just like single fault alerts, multiple fau= lt alerts also need to be further confirmed by the process operator. Since in reality multiple faults tend to occur much less frequently than single faults, single fault alerts should always be fully checked out prior to checking out the multiple fault alerts.

The evidence determined by evaluating the engineering models is combined together using a Fuzzy Logic calculation. This calcula= tion determines a certainty factor associated with each fault hypothesis. These certainty factors can range from 0 (at least some evidence does not suppo= rt that fault hypothesis) to 1.0 (that fault hypothesis is a highly plausible explanation of the current process behavior). Alarm generation and display components are used to flag alerts to the users. The package has a comprehensive al= arm management system linked to analysis modules for generati= ng appropriate alarms based on the alerts generated. Appropriate alerts are given if these certainty factors exceed the chosen alert thresholds as described below.

Alerts are given priority by this module according to the following color scheme. If the certainty factor associated with the f= ault hypothesis is greater than 0.97, a red alert (highest priority alarm mana= ged by alarm management system) is given. In this case process conditions are definitely abnormal and it is highly likely that this process fault may actually be occurring. If the certainty factor is less than 0.97 but grea= ter than 0.92, a yellow alert (warning or caution) is given. This indicates t= hat process operations are no longer considered to be normal and the identifi= ed fault may be the reason why. Else a green alert (validation) is given. Th= ese alerts indicate that the corresponding variable is valid (i.e., either the indicated sensor measurement is correct or the assumed value for an unmeasured variable is correct). These threshold values, which are configurable, were chosen to be highly certain that those associated aler= ts are real. They make the SV&PFA module results conservative by not alerting until process conditions are well outside their normal operating ranges. = The SV&PFA module should thus give few false alarms.

<= strong>VSPC:

The Virtual Statistical Process Control (VSPC) softw= are module augments the SV PFA module, providing an independent but complemen= tary analysis of sensor measurements. The VSPC module uses Exponentially Weigh= ted Moving Averages (EWMA’s) to determine in real-time if individual process sensors and process conditions are in control, are going out of control, or are definitely out of control. It directly flags out of control sensors and process conditions in real-time rather than after-the-fact. These alerts can also occur at levels that may allow the process operators to take appropriate control actions to reduce= or eliminate disruptions to process operations or product quality. This meth= od is considered virtual because the analysis is done automatically without = the need for the operators to collect and chart any process sensor readings. =

Statistical Process Control (SPC) is a tool used to assess whether a process is currently under or out of control. Various techniques exist for doing this analysis depending upon the nature of the process being monitored. In continuous processes (as opposed to the manufacture of discrete, individual units), process data collected at a particular moment in time is not completely independent of its previous d= ata. This phenomenon is referred to as auto-correlation between the data. SPC techniques for continuous single point samples that handle auto-correlati= on in the data include Cumulative-Sum (CUSUM) as well as EWMA. These techniq= ues can better handle the effects of auto-correlation and allow small but statistically significant shifts in an observed variable’s value to= be readily detected. With this process monitoring suite, the EWMA calculatio= n is the preferred method because of its ability to not only monitor current process operations but to forecast where the process is headed. There are= two classes of sensor variables that can be used by this module: sensors that= are being directly controlled and those that are not.

Since this analysis is based solely on statistical calculations, false al= arms are possible. Measurements may be identified as out of control and then rapidly determined to be in control once again. Such spurious alerts can = be minimized by choosing update times for the analysis that are close to one time constant for the process being monitored. However, they cannot be totally eliminated. Therefore, VSPC alerts also need to be further confir= med by the users before any.

<= span style=3D'font-size:11.0pt;font-family:Verdana;color:#333333'> <= /o:p>