Pirouette can detect the presence of meaningful information in data and create multivariate models able to make predictions about future samples. Automating the characterization of these future samples is the goal of InStep. It combines a decision tree approach with Pirouette’s multivariate data processing to facilitate routine prediction for both laboratory-based and in-line process monitoring and control functions.
Complex applications benefit from the increased prediction accuracy and sensitivity to outliers (e.g., unusual samples, process upsets) associated with a multivariate approach. Generating a decision tree around a classification model (predicting a sample category), a regression model (predicting a continuous property or concentration), or any combination of both model types allows predefined rules to be applied to complex problems. When coupled to a given instrument system, the result is a turn-key custom analyzer.
The key aspects of InStep are:
- its behind the scene use of Pirouette to make predictions
- a form-based interface for creating decision trees using multivariate models
- a “watched folder” approach which allows it to run concurrently with instrument data acquisition programs
- generation of custom reports which can be displayed on-screen and/or saved to a file
- presentation of control charts for visual monitoring of results and diagnostics
Users may also invoke InStep via a command line call which specifies a method, format, and target data file. In this mode InStep processes the target data and then quits. The Watched Folder parameters in the method are ignored in this case.
InStep has been developed as an ActiveX client, which communicates directly with the IPAK server, the same engine that controls Pirouette processing. The software operates in two basic modes: design and sample processing. Design consists of configuring a method and constructing a template for the prediction output to be used during sample processing.
An InStep method is a set of parameters which specify how to process samples, e.g., which models to use, where and how often to look for data files. Methods are created and edited usinga form as shown here. Example methods are supplied as part of the InStep installation to demonstrate how to set up the software for different purposes. You get decision points, hierarchical models, and no need to specify the preprocessing or transformations (all processing information is contained in the Pirouette model).
A method is run to produce prediction results. An InStep format is a set of parameters which specify how to display these results in a tabular report. Example formats in the InStep installation accompany each of the methods. The reports can be saved for later reference. You can save prediction results to a file automatically or simply save the on-screen report when you have stopped processing. There is enough flexibility in the formatting options that the report can be built for display in a browser using HTML.
In addition to tabular output, InStep can also prepare control charts to display results. Items which can be plotted include the predicted values and the same assortment of outlier diagnostics available to the method. Charting options are available for any predicted value as well as for diagnostics, as shown in this example plot.
Utilities and descriptions are available for a variety of instruments, including those based on Agilent ChemStation and EZChrom, Waters Millennium/Empower, and PerkinElmer Totalchrom. These examples can be used as guides to connect to a variety of analytical systems.