4. Configuring Models

This article contains the following sections:


Model configuration is the core capability of the AI Data Pipeline. It provides an interface for you to define the configurations that will act as a set of instructions for the AI Data Pipeline as it produces your unified data model.

The AI Data Pipeline applies these user instructions to incoming machine and manufacturing system data record streams, and then leverages patent pending algorithms to render data models that represent your production processes.

Sight Machine AI Data Pipeline User Configuration Data Flows

These individual data models serve as the foundation for a unified data model representing the entire production process that enables Sight Machine’s Enterprise Manufacturing Visibility (EMV) and Enterprise Manufacturing Analytics (EMA) applications to deliver unique insights into a manufacturer’s operations.

Currently, we support the configuration of Facilities, Machine Types, and Machines. This is sufficient for the production of a Cycles model, a Downtimes model, and OEE metrics.

About Time Sequencing

In order to model streaming data in near real time, Sight Machine blends, joins, and integrates data streams from multiple sources. To establish relationships between disparate data types and sources, the AI Data Pipeline tags data records with timestamps at various points in the data acquisition process. At this point, each manufacturing data packet contains at least two time fields:

  • data.timestamp: A field within the raw data that indicates the time when the raw data was generated. This field is the critical element used to order, group, and blend data from the various incoming data streams, and is used by the Sight Machine platform for all downstream calculations, like cycle boundaries, downtime boundaries, etc.
  • capturetime: This field indicates the time that the Sight Machine software captured the data from its original source.

NOTE: We expect that you are working with real-time data that is in order. If you are working with high-latency data or a blend of data that is coming in at different times, contact Sight Machine for support and assistance.

How Cycles Affect the Modeling Process

For the AI Data Pipeline to properly model data, each signal or data stream from an individual sensor must have a discrete value or set of values for each cycle. Typically, each signal would be distilled to one value; rarely must all signals be represented in the data to produce a useful model. Process or data engineers with expertise in the production process at each facility are usually best equipped to define the rules for establishing these cycle values. After these rules are defined, the AI Data Pipeline will automate the method of establishing values for each cycle to the streaming data.

For many signals, using an average of the continuous time series values captured during a cycle provides the appropriate level of fidelity for constructing the data model. In the first example, you can see that for continuous temperature data, you could model and visualize the average temperature over a number of cycles.

Example 1: Modeling Continuous Data

In the second example, for a data stream that varies significantly during the duration of a cycle, capturing the start, end, and minimum temperature values generate a more accurate model than a simple average.

Example 2: Modeling Periodic Data

Establishing these rules for determining the discrete value(s) of each signal or data stream from an individual sensor for each cycle is a critical element of user configuration. The AI Data Pipeline provides an easy-to-use interface for defining these rules for each Data Field in the Machine Type configurations.

Adding a Facility

When working with a new Sight Machine instance, start by creating a Facility model.

This model includes the location of the facility, the shift schedule for the facility, and later on, all of the assets located at this facility.

It is important to add shifts for each facility so that the Sight Machine platform can use this information to analyze when production and non-production times are taking place. For instance:

  • Cycles that occur during a shift are labeled with that shift. Cycles that occur outside of any configured shift are labeled as Off-Shift.
  • Downtimes that occur during a shift are labeled as Unplanned. Downtimes that occur outside of any configured shift, or during a break, are labeled as Planned. This is used in the calculation of OEE Availability.

To add a Facility:

  1. On the Models tab, click Facilities.
  2. On the main Facilities screen, click the plus (+) button to add your first Facility.
  3. Type a name for the facility, and then click Configure Facility.
  4. Under Location, in the Address field, type the address of the Facility. The geolocation and time zone automatically populate based on that address.

    NOTE: The interface accepts a partial address, so if you prefer to leave the address private, you can type a city, state, or country instead.
  5. Under Shifts, click 1 Shift.
  6. In the upper-right, click the plus (+) button to add a new shift. You can also click the overflow menu on the existing shift, and then click Edit Shift.
  7. In the New Shift window, do the following:
    • In the Name field, enter a user-friendly reference for the shift. This name appears in Manufacturing Applications.
    • Under Shift Occurs on These Days, click the field to open a drop-down menu, and then specify which days in the week this shift runs.
    • In the Shift Start Time and Shift End Time fields, specify when the shift is scheduled to start and end.
    • To add breaks to this shift, click Add Break, and then specify the start and end time of the break. These are optional.
    • Click Add Shift or Apply to apply the changes.
  8. When finished with the shift setup, in the top-left next to the Facility name, click the back arrow.
  9. Back on the main screen, on the left, click Deploy.
  10. After you deploy the facility, you return to the Facilities screen. You can click the facility to view or edit it. Whenever you make changes, you see a Revert All Changes button on the left, in addition to the Deploy button. Revert All Changes deletes all undeployed changes, and reverts to the most recently deployed version.

Configuring a Machine Type

Machine Types are the core of the AI Data Pipeline, and for each customer machine, define how every bit of data will be processed by the Sight Machine platform. Think of Machine Type as a model for a set of machines that behave in the same way or have the same function .

When creating a new Machine Type, you will need to perform the following tasks:

  • Create a new Machine Type.
  • Define cycle boundaries.
  • Generate downtimes.
  • Create and manage downtime reasons.
  • View and establish rules for the data fields.
  • Deploy the Machine Type.

Creating a New Machine Type

The first step in creating a new Machine Type is to provide a unique Machine Type name and select an Asset generated in FactoryTX. When you click the Configure button, the Sight Machine platform automatically inspects all raw data associated with that Asset and generates the basic structure of a Machine Type. This inspection may take a few seconds, depending on the amount of raw data being analyzed. After the inspection runs, check that the list of Raw Data Fields is what you expect. If not, you might want to revisit your FactoryTX configuration .

To create a new Machine Type:

  1. On the Models tab, click Machine Types.
  2. On the main Machine Types screen, click the plus (+) button to create your first Machine Type.

  1. On the New Machine Type screen, type a name for the new Machine Type. This should be a user-friendly reference for the Machine Type that appears in Manufacturing Applications.
  2. Click the Asset field, and then in the window, select an asset that was uploaded from FactoryTX to use to configure the Machine Type.
  3. Click Configure Machine Type.
  4. A status screen appears as the AI Data Pipeline automatically scans thousands of raw data records, and generates the basic layout of your new Machine Type.
  5. After you create the Machine Type, you should double-check that all of the raw fields were automatically detected. On the Machine Type screen, click DATAFIELDS, and when the page scrolls down to that section, click Raw Data Fields/Cycle Data Fields.
  6. On the Data Fields screen, check the list of all the raw data fields and their associated data streams. If there are any fields missing:
    • First, use the Raw Data Table tool to validate that your raw data includes all fields. If not, return to your FactoryTX configuration to address the issue.
    • If the Raw Data Table contains fields missing from the Data Fields screen, please contact Sight Machine for assistance.

Defining Cycle Boundaries

The second step in creating a Machine Type is to make sure that you correctly define the cycle boundaries, to ensure that each cycle in the model starts and ends at the correct timestamps .

To define cycle boundaries:

  1. On the Machine Type screen, click CYCLES, and then click Cycle Boundary Method to open a drop-down menu with choices.

  1. Select one of the following:
    • Counter Already Exists: Use if there is a cycle counter in the raw data.
    • Machine Event: Use if there is an event log in the raw data that can be mapped to cycle boundaries (e.g., load and unload).
    • One Cycle Per Record: Use to convert each record from each stream into a cycle.
    • Time Interval: Use to make all cycles the same duration.

  1. If you select Counter Already Exists, scroll down to Cycle Counters, and then do the following:
    • Under Cycle Counters, click Add a Steam, and then click the Stream drop-down arrow to select an option (for example, PLC).
    • Click the Counter drop-down arrow to select the field on the stream that represents the cycle counter.
    • Under Set the Boundary Function, click the Boundary Function drop-down arrow to select the boundary for the counter.
    • Under Max Cycle Time, accept the default of 10000 ms or set a new time. When a cycle exceeds this duration, the AI Data Pipeline ends the cycle and starts a downtime, and then waits for the signal to indicate the start of a new cycle.

  1. If you select Machine Event, scroll down to Event Source, and then do the following:
    • Under Event Source, click the Stream drop-down arrow to select an option.
    • Click the Field drop-down arrow to select a field that contains an event stream from the machine.
    • In the Cycle Event Start field, type the string in the event log that represents the start of a cycle (for example, load).
    • Under Max Cycle Time, accept the default of 10000 ms or set a new time.

  1. If you select One Cycle Per Record, to establish the boundaries based on that one record, under Max Cycle Time, accept the default of 10000 ms or set a new time.
  2. If you select Time Interval, on the Cycles screen, under Cycle Duration, accept the default of 60000 Milliseconds or set a new time.

Generating Downtimes

There are two methods for creating downtimes :

  • Maximum Cycle Time: You define Cycle Boundaries, and downtimes are created when Cycles reach the Maximum Cycle Time.
  • Downtimes During Gaps: The AI Data Pipeline automatically creates downtimes in any gaps that exist between Cycles.

Creating and Managing Downtime Reasons

Regardless of how a downtime is generated, you can classify it based on a list of downtime reasons.

To create and manage downtime reasons:

  1. Under Downtime Definition, in the drop-down list, select the Raw Field that contains codes identifying the downtime reason.

  1. Click Downtime Reasons.
  2. In the Add a Downtime Reason window, do the following, and then click Apply:
    • Type a name for the reason.
    • Type the raw data code.
    • In the Type drop-down list, select Planned or Unplanned.
    • Type an optional description.
      For example, you may know that a certain machine returns the error code 0111 for a parts issue.
  3. This will add your Raw Data Code Field to the list below. Planned or Unplanned are determined based on whether the downtimes occur during a specified shift or break, or outside of a shift.

Viewing and Establishing Rules for the Data Fields

After you set up cycle boundaries and create downtime reasons, you can validate the data fields with which you are working.

To view and establish rules for the data fields:

  1. On the Machine Type screen, click DATAFIELDS, and when the page scrolls down to that section, click Raw Data Fields/Cycle Data Fields.

  2. In the Data Fields window, for each Raw Data Field, first check that the Data Type is correct.
  1. If you need to change a type, click the type and select a different type from the drop-down menu. You can select any of the following:
    • Continuous: Containing only numbers, including integers, decimals, negative numbers  (e.g., temperature, pressure).
    • Binary: Containing only 0 and 1.
    • Categorical: All data values are treated as a string.
    • Datetime: Containing only timestamps.
  2. Use the arrows on the right to expand some or all of the data fields.

  3. For each Raw Field, add and remove Functions to get the desired Cycle Data Fields, as follows:
    • For continuous Raw Fields, you can apply arithmetic functions, such as min, max, avg, etc.
    • For categorical and datetime Raw Fields, you can choose to store the First or Last value onto each cycle. You can also choose to store a comma-separated list of unique values.
    • For binary fields, you can apply functions to calculate the duration that the signal was True, the number of rising edges, and the start and end time of the signal.
  4. For each Cycle Data Field, check that the name is descriptive and user-friendly, and update the ones that are not.
  5. In the upper-left, click the back arrow to return to the main screen.

Deploying the Machine Type

On the main screen, on the left, click Deploy. After you deploy the Machine Type, you return to the Machine Types screen. Click a Machine Type to view or edit it. Whenever you make changes, you see a Revert All Changes button on the left, in addition to the Deploy button. Revert All Changes deletes all undeployed changes, and reverts to the most recently deployed version.

Configuring a Machine

After you have defined much of the critical information in Machines Types, a Machine is a fairly straightforward configuration. Simply name your Machine, identify the Asset from FactoryTX that defines this Machine, and select the Machine Type and Facility to which the Machine belongs.

To configure a Machine:

  1. On the Models tab, click Machines.
  2. On the main Machines screen, click the plus (+) button to add your first Machine.
  3. On the New Machine screen, do the following:
    • Type a name for the Machine.
    • Select an Asset to associate.
    • Click Configure Machine.
  4. Select the Machine Type and Facility for this Machine.
  5. Scroll down to Shift Schedule and check that the configuration is correct. To change a Machine’s shift schedule, click the Add a Shift button, or the trashcan to the right of a shift.
  6. Click Deploy.
  7. The Machine is deployed and you return to the Machines screen. Click a Machine to view or edit it. Whenever you make changes, you see a Revert All Changes button on the left, in addition to the Deploy button. Revert All Changes deletes all undeployed changes, and reverts to the most recently deployed version.

Configuring a Part Type

Before creating a Part Type, make sure you already have the following:

  • At least one deployed Facility
  • At least two deployed Machines Types, with a serial field on each
  • At least one deployed Machine per Machine Type

To get started, open your Sight Machine environment, navigate to the AI Data Pipeline, navigate to the “Part Types” page under the “Models” tab, and click the Add Part Type button:

Give your Part Type a descriptive name and category:

Under “Traceability”, add each of your Machine Types, and then select the relevant Serial Field:

Go to “Data Fields” and review and update the information.

  • For each Data Field copied from your Machine Types, check that the Function is set correctly. If each Part is created from one cycle, then First/Last functions will be sufficient to copy that record. If each Part is created from more than one cycle, be sure to pick the appropriate aggregation function.
  • Ensure that each of your Data Fields has readable, user-friendly name.

Return to the previous page and Deploy your Part Type.

Now that you’ve deployed your Part Type, you can navigate to the Compute tab and contextualize your data.

Select the relevant Assets and Start Date and Compute.

When the jobs complete, navigate back to “Manufacturing Applications” and create a new Data Table. Select the “Parts / Lots” model, select your Part Type, select a relevant Date Range, and Update.