Process data interpretation

By definition, the exemplar patterns used by these algorithms must be representative of the various pattern classes. Performance is tied directly to the choice and distribution of these exemplar patterns. In light of the high dimensionality of the process data interpretation problem, these approaches leave in question how reasonable it is to accurately partition a space such as R6+ (six-dimensional representation space) using a finite set of pattern exemplars. This degradation of interpretation performance as the number of possible labels (classes) increases is an issue of output dimensionality. 

Increased trust in pattern recognition The active user involvement in the data mining process can lead to a deeper understanding of the data and increases the trust in the resulting patterns. In contrast, "black box" systems often lead to a higher uncertainty, because the user usually does not know, in detail, what happened during the data analysis process. This may lead to a more difficult data interpretation and/or model prediction. 

In the chemical engineering domain, neural nets have been appHed to a variety of problems. Examples include diagnosis (66,67), process modeling (68,69), process control (70,71), and data interpretation (72,73). Industrial appHcation areas include distillation column operation (74), fluidized-bed combustion (75), petroleum refining (76), and composites manufacture (77). 

Analysts The above is a formidable barrier. Analysts must use limited and uncertain measurements to operate and control the plant and understand the internal process. Multiple interpretations can result from analyzing hmited, sparse, suboptimal data. Both intuitive and complex algorithmic analysis methods add bias. Expert and artificial iutefligence systems may ultimately be developed to recognize and handle all of these hmitations during the model development. However, the current state-of-the-art requires the intervention of skilled analysts to draw accurate conclusions about plant operation. 

Measurement Selection The identification of which measurements to make is an often overlooked aspect of plant-performance analysis. The end use of the data interpretation must be understood (i.e., the purpose for which the data, the parameters, or the resultant model will be used). For example, building a mathematical model of the process to explore other regions of operation is an end use. Another is to use the data to troubleshoot an operating problem. The level of data accuracy, the amount of data, and the sophistication of the interpretation depends upon the accuracy with which the result of the analysis needs to oe known. Daily measurements to a great extent and special plant measurements to a lesser extent are rarelv planned with the end use in mind. The result is typically too little data of too low accuracy or an inordinate amount with the resultant misuse in resources. 

The overall process of data interpretation and the development of suitable remedial strategies once a set of causes has been identified, is set out in Figure 6.4. The two-stage process of confirming the initial causal hypothesis is recommended to overcome the tendency to jump to a premature conclusion and to interpret all subsequent information on the basis of this conclusion. 

These include identification of process equipment and instruments, interpretation of the meaning of their values and trends, navigation through different VDU pages by means of a selection menu, etc. The common feature of these tasks is handling the display system to search and locate relevant process data. In this respect, "classical" ergonomics checklists (see Chapter 4) are very useful in facilitating performance of such tasks. 

The correct interpretation of measured process data is essential for the satisfactory execution of many computer-aided, intelligent decision support systems that modern processing plants require. In supervisory control, detection and diagnosis of faults, adaptive control, product quality control, and recovery from large operational deviations, determining the mapping from process trends to operational conditions is the pivotal task. Plant operators skilled in the extraction of real-time patterns of process data and the identification of distinguishing features in process trends, can form a mental model on the operational status and its anticipated evolution in time. 

The ideas presented in Section III are used to develop a concise and efficient methodology for the compression of process data, which is presented in Section IV. Of particular importance here is the conceptual foundation of the data compression algorithm instead of seeking noninterpretable, numerical compaction of data, it strives for an explicit retention of distinguished features in a signal. It is shown that this approach is both numerically efficient and amenable to explicit interpretations of historical process trends. 

Labels are distinguished based on whether they are context dependent or context-free. Context-dependent labels require simultaneous consideration of time records from more than one process variable context-free labels do not. Thus, generating context-free trend, landmark, and fault descriptions is considerably more simple than generating context-dependent descriptions. Context-free situations can take advantage of numerous methods for common, yet useful, interpretations. Context-dependent situations, however, require the application of considerable process knowledge to get a useful interpretation. In these situations, performance is dependent on the availability, coverage, and distribution of labeled process data from... 

Because the techniques for data analysis and interpretation are targeted to address different process characteristics, care must be taken in choosing the most appropriate set of techniques. For example, some techniques work best with abundant process data others, with limited process data. Some can handle highly correlated data, while others cannot. In selecting appropriate methods, two practical considerations stand out ... 

---