Choosing the Type of Model

Once you've chosen the asset you want to model with the Quick AI addon, it's time to choose what type of model you want to make.

An Instantaneous model isn't built or run instantly - it takes the instantaneous values of your asset (the values at a single moment in time) and uses them to model one or more values.

For example, it can try to calculate your power usage based on your speed, product type and weight sensors.

Structure

These models usually have a large number of inputs powering a small number of outputs.

Benefits

These models can be trained and run fast, and they have a simple structure.

Caveats

These types of model work best when you have little-to-no lag in your process. A change in one property should have an almost immediate effect on the other properties inside the model. If there are long delays, consider a time-series model instead.

An setup model is designed to predict the setup/limits/thresholds/setpoints you should use in a process where your inputs or outputs vary (for example, mining applications deal with a wide variety of input products, while production lines might produce a number of different types and sizes of product).

Structure

These models usually have a small number of inputs (such as product type, size, weight etc.) powering a large number of outputs.

Caveats

It's not unusual to have to add additional data to the training information, such as summary or feedback information. It's likely you'll need to edit the code to do this, but the provided framework will get you off to a strong start.

Note that these models can be influenced by inconsistency. If people are not setting their equipment consistently, the model might have trouble with accuracy. If the model isn't training correctly, we suggest disabling individual output channels and re-training to identify the issue.

Time-series models are used when you have significant amounts of lag in your process - if there is a noticeable amount of time or processing distance between a cause and an effect.

These take significantly more time to train and run compared to instantaneous models.

Structure

These models usually have a large number of inputs powering a small number of outputs.

Benefits

These models can produce astounding results when dealing with interactions over time.

Caveats

These models take significantly longer to train and run than regular models. They are able to extrapolate significantly more information with a small amount of data, but this also makes them more likely to produce 'noisy' or less accurate signals.

Press the Generate Code button to Download and Run the Script.