There is no argument that building and maintaining models in an online environment is a painful process: Download analyzer data, Synchronize analyzer and Lab data, Use a modeling tool to define chemometrics parameters, Remove outliers, Create the model, and Install it back on the analyzer.
Now, do it again for each property, and then repeat the process several times until the models perform as expected.
Can't we use Artificial Intelligence to replace traditional model development techniques?
Can we let a machine detect outliers? Was the spectrum processed correctly? What about synchronizing Lab and spectra time stamps?
Well, we let machines decide when to buy and sell shares, how to navigate drones back to their home base, and we let unmanned vehicles drive and decide when to break.
Artificial Intelligence is everywhere
It's time it enters the online process.
A modeling solution based on Artificial Intelligence can save time and money, it can respond fast to changes and provide similar accuracy to traditional models,.
The user must have the tools to know that the process works as expected, and to control the level of automation.
It must be simple enough to be used by operators with no knowledge in chemometrics, and on the other hand, the administrator must be able to control the modeling algorithm, and (optionally) approve newly created models.
The iModel Solution
iModel developed a software solution (ModelGateway) that automates Chemometrics.
It loads Spectra from any analyzer and gets Lab Data over the network.
The software automatically processes the training set and creates a model for each analyzed property. It then predicts any new spectrum and transmits the prediction to the DCS. The analyzer is not required for any prediction and communication tasks anymore.
All data (Spectra / Lab Data / Predictions) is stored in a database providing an easy to use dashboard for managing the models and predictions
A graphic tool is available for lab users who wish to get a feeling of what the automatic modeling looks like. The tool lets the user define the settings of the modeling algorithm, it displays the progress of the algorithm, and the result model can be saved and used by the software.
The software provides a set of tools for validating the quality of the models by displaying spectra comparison, comparing predicted data to lab data and providing statistic calculations. Models developed manually using Grams/Unscrambler can be integrated into the software and be used as reference for automated models.
Let the machine do the work while you take the final decision.
This can be done by configuring the software in Semi-Automatic mode. It will build the models, prepare a comparison graph between the current and the new model, and let you decide whether to switch to new model (green button) or stay with the old one (red button)