Analytical Instrumentation


STREAMLINING THE USE OF AI/MACHINE LEARNING IN THE CHEMICAL INDUSTRY: CHEMOMETRICS


Abstract Artifi cial intelligence and machine learning are inevitable results of the work driven by the consumer side of our economy. The question is not whether it will impact refi ning and chemical plant operation, but how soon and how long it will take for the benefi ts to outstrip the costs. The goal is to distinguish between vision and hallucination and to provide some practical guidance for making progress in this complicated set of fi elds. There are three categories of measurements that provide us with the data that will form the basis for any interpretation system: single-purpose sensors, chromatographs, and spectrometers. Chemometrics can be used in all three categories and, in fact, is critical to interpreting output from any type of spectrometer. We can easily demonstrate that the use of multivariate analysis for each of the three data sources, taken individually or assembled together, gives faster response, improved fl ow of information derived from these data, and a signifi cant leg up for process understanding. This information is available


and is nearly cost-free. Introduction


As computers came on-line four decades ago, automated quality control became practical, but we still don’t take advantage of that processing capability, instead sticking with engineering heuristics and rendering most of the available data impotent for control. This applies equally to the monitoring of process variables and to the analyzer population. In the process sensor world, ALL measurements could be combined to form a virtual instrument of a unit or even the entire plant, isolating the process-relevant signal from the noise. The advantage is that previously-unseen process upsets can be identifi ed and managed in close-to-real time. Spectroscopic and chromatographic analyzers have enjoyed a more intimate relationship with the computer, but we have not yet tapped into the potential of processing raw data into actionable process information. The key to success is to NOT focus solely on extracting the information content of a stream of data using multivariate analytics. More critical is the blending of these results with application-specifi c product knowledge. Even that is not enough; all is lost if we do not focus on the effective delivery of that blended information content. Tools abound from multiple sources that can assist in achieving a proper blend, but the successful engineer needs to tackle the task with a systems engineering approach. Any integration must be both useful and used.


In the literature and in the media, there is terminology confusion that pervades any discussion of the use multivariate analysis to manage information gathering and performance prediction. AI, machine learning and chemometrics represent overlapping fi elds of study and do not in and of themselves dictate a path to follow in getting to understand what your data is telling you, let alone facilitate the building of useful process models. I choose the term chemometrics in that it blends the thought of chemistry understanding with the math we use, distilling our focus on industries like those tied to hydrocarbon processing.


Regardless of the label in use, the goal is to use mathematical tools to assess quality in products in a cost effective and reliable manner. In these examples, I use Infometrix software, but there are myriad options, particularly for process variables where there is signifi cant overlap with the statistics community. Let’s look at three areas where nearly instant fi nancial benefi ts are available through a simple application of chemometrics.


Process variables


Multivariate data analysis, as powerful as it may be, can be daunting to the uninitiated. It has its own jargon and incorporates concepts which at fi rst glance seem strange. Typically, a novice user becomes comfortable with the multivariate approach only after a period of confusion and frustration. Why bother? The rationale behind multivariate data analysis is simple: univariate methods, while well–understood and proven for many applications, can sometimes produce misleading results and


Chromatography


In the hydrocarbon processing industry, chromatography is the traditional, go-to technology for monitoring chemical composition. It is the most direct way to measure the concentration of individual molecules. With the current generation of fast, highly capable GCs and the advent of process ultra-high- performance liquid chromatographs, analyses can be completed


Figure 2: Process variables turn into inferentials and here clearly show process conditions that are normal (right) and show upset conditions (left). Principal Component Analysis distills all data into a single plot showing when the process is in control. Each datapoint represents the signature of the process for a specifi c time; the closer the points, the more similar the process conditions.


WWW.PETRO-ONLINE.COM at a rate commensurate with optical spectroscopy.


For a variety of reasons, processing chromatographic data using multivariate techniques has been largely neglected over the past few decades. There are two multivariate technologies necessary for automated processing of any-but-the-simplest chromatographic applications: a signal processing step to align the chromatographic traces and, as demonstrated for the process variables approach described in the previous section, a mining of the chromatographic trace (or tabular results) for full information content.


Spectroscopy


Figure 1: The terms used to label discussions in multivariate data processing are not distinct; overlap abounds.


overlook meaningful information in complex data sets.


Univariate methods were developed for univariate data. Applying univariate methods to process measurements may be useful but it is also tantamount to discarding all but one of the measurements. While some problems may yield to a thorough statistical analysis of a single variable, this approach has several drawbacks when applied to multivariate data. Basically, it is incomplete if multivariable relationships are important. If the measurements are numerous and/or correlated, processing them as a unit is advised. Chemometric techniques are natural garbage collectors as the data processing does not easily get hijacked by erratic sources of data. Applying fi rst principals to guide the multivariate process models is critical to long-term success.


Most of the sensors in play in a process setting are pressure, temperature, fl ow, and level monitors that can be considered together, melding them into a custom process instrument. We can use one of two techniques to tease the information content from the mix: Principal Component Analysis helps us classify the state of the process (e.g., is there a process upset?); and Partial Least Squares allows us to predict performance measures (such as yield). These are the tools to use for tuning the results to suit the process at hand and for automating the routine evaluation of the data stream.


Chemometrics has seen much use in the management of optical spectroscopy data collected for monitoring a chemical process, and is routinely integrated into the analytical workfl ow both to improve the signal and to process the signature into useful quantitative and qualitative information. But, the typical employment in process systems is in need of an upgrade.


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46  |  Page 47  |  Page 48  |  Page 49  |  Page 50  |  Page 51  |  Page 52  |  Page 53  |  Page 54  |  Page 55  |  Page 56  |  Page 57  |  Page 58  |  Page 59  |  Page 60  |  Page 61  |  Page 62  |  Page 63  |  Page 64  |  Page 65  |  Page 66  |  Page 67  |  Page 68  |  Page 69  |  Page 70  |  Page 71  |  Page 72  |  Page 73  |  Page 74  |  Page 75  |  Page 76