RELEASE NOTES
RELEASE NOTES
2022. 7. 5.
How and when critical information on bottlenecks, variability and quality is delivered should never be a mystery. Fast, reliable results from investments in Drishti require transparency and consistency, so we’ve enabled new administration features that provide an easy way to request new AI models and changes to existing ones.
A Line AI Video Set is the vehicle for setting up Drishti AI. It is a collection of data samples for cycle and action times at each station on your line. Drishti AI models are trained with these data samples. The result of creating a Line AI Video Set and submitting a setup request is the continuous measurement of cycles and actions that feed Drishti Analytics.
Here’s how a Line AI Video Set works:
Video is used to collect the data samples at each station via annotation tools
The video and data samples at each station are used for training AI models
Only one line AI video set is used at a time to train the AI model(s)
When assembly processes at a station change significantly, a request for a model update is made via a new AI video set
To make a request for a line AI setup you must have administrative permissions. You also must have standardized work instructions, or what Drishti calls, a Work Definition. This is the set of actions that either define the process at each station or at least provide some critical actions that would define a cycle. The Work Definition helps with training the models and provides action and work times within a cycle.
You’ll be ready to make an AI Setup request with the correct permissions and standardized work instructions.
When you select Line AI Video Sets from the Setup dropdown menu you can create a Line AI Video set and all the station reference videos for submission. There are a set of screens that will guide you through the process.
The first is the list of Line AI Video Sets for each line. Here you’ll see the line name, products, number of stations, relevant dates and status for each video set record. You can have multiple Line AI Video Sets for each line, however only one will be “Activated”, since only one video set at a time is used for AI model training, retraining and deployment.
You navigate to the second screen via the Details link on any Line AI Video Set in the list. There you will see a pre-populated list of station records, one for each station on the line as configured in the Lines & Devices Setup. This is where you create video reference records for both cycle and action times.
The video references for each station and product are based on the Work Definition for the line. Our annotation tools are used to record the times for actions and cycles at a station and each product assembled there. If you need a refresher on how to use Drishti Self Annotate, please see our tutorial.
Note: If you know you’ll only be requesting cycle data at a station, you will only be required to submit video references for cycles, however a Work Definition is still required. For the best results, consider creating some critical actions that will help define the cycle.
Once your station reference videos are recorded you will be able to navigate to the third and final set of screens to request AI model training and deployment.
First you will request the type of detection needed at the station. When requesting Action or Work Time detection at a station, cycles will also be detected, so cycles will remain the default selection on this screen. Work Time Detection is a variant of cycle detection. It will detect the time a person works at the station and can be configured to your specifications.
Next, you will define the AI delivery specification. This tells Drishti when you want your AI-based analytics. They can be delivered minute by minute in near real time or at the end of every day. We call this the data availability schedule. And, if there are specific days and times you want us to analyze (or don’t want us to analyze) then the video processing schedule is defined. The result of specifying both the data availability schedule and the video processing schedule gives you the AI delivery specification.
In the final summary screen you will be able to see the requested AI detection specifications and AI delivery specification, so you can make any adjustments as needed before submitting the AI Setup request.
Once submitted, you can monitor the status of your request at any time. There are five updates you can receive on your request, however you may not see all five for a given request. Here’s what they mean:
Request submitted - the AI request is recorded in the Drishti Portal
Request accepted - the AI request has been accepted by the Drishti AI team and can be trained and deployed per your request
Live - the AI has been trained and deployed per your request
Paused - the AI is temporarily paused and data is not available for the station
Deactivated - the AI request has been been deactivated and is no longer running for the station requested
Only the Paused and Deactivated status can be set with administration permissions. The other status updates are managed by Drishti.
A Line AI Setup request is complete once all station reference videos and AI requests have been submitted for the line. You will see another AI status on the AI Video Set list page where you can monitor the entire request for the line.
When the AI models for the line are trained and deployed you will see the “Activated” status. This means that all of the stations have AI-based analytics running, they're available in the Drishti Portal and you have completed the AI Setup process.
