<aside> 💡 After several months of observation and experimentation, we are proud to present the new version of Spark.
The main goal is to enable you to optimize the performance of your inspection applications autonomously.
In addition, you will be able to increase your inspection throughput up to 3 parts per second.
Finally, this new version includes numerous visual improvements designed to streamline the use of your Spark station.
</aside>
To evaluate the performance of your applications, compare the quality assessments provided by the system with your internal expertise, directly in Spark.
You can now indicate a manual quality assessment of the inspected parts after the fact, and in a reversible manner.
Inconsistencies between the assessments provided by the system and the user are highlighted in the history, in order to understand the key success factors.
Finally, this information is accessible in the Quality Center for future review.
To find the right compromise on the sensitivity of the system more quickly, the choice of the threshold, at which a part is considered defective, is greatly facilitated.
Indeed, within the calibration function, you can now indicate more flexibly the assessment of part quality by the user. For example, the parts represented in red in the calibration graph are the parts considered defective by the user.
In order to improve the performance of your Spark system, you may need to retrain an application. This operation no longer erases calibration data.
You can calculate the new quality scores resulting from the retraining of the application and save valuable time.
When you want to improve your application by adding a training part, you now store this example in a queue.
You make your application more robust by systematically reviewing candidate parts for the next training.
When you want to add multiple parts from the inspection, it is no longer necessary to retrain the model for each addition. You can wait until the queue is sufficiently filled to launch a single retraining.
Your iterations are thus more relevant and less time-consuming, in order to achieve the desired performance.
