99.33 million
Average Number of Defects Per Quarter
- 10%
Of Medical Device Manufacturer Drop in Share Price After a Single Recall
$457 billion
Global MedTech Industry Revenue in 2020
EXECUTIVE SUMMARY
TensorIoT helped our customer improve their operational efficiency and train their machine learning in order to recognize defects.
GOAL
How can you create a custom AWS IoT framework to properly train computer vision machine learning models to perform quality assurance testing?
RESULTS
Our client has a custom IoT framework to gather the data needed to train computer vision models to automatically screen for any defects in their surgical products such as chips, scratches, and anodization.
Background
Our customer develops medical devices and software systems for surgical planning and monitoring, including spinal surgical implants. While manufacturing and marketing medical devices, our client continuously tries to improve their operational efficiency, from recognizing defects to training computer vision from collected data. Our customer identified their labor-intensive quality control processes as a bottleneck in their manufacturing process, making it the prime target for improvement
The Challenge
The customer wanted to improve their operational efficiency by using computer vision and machine learning to perform automated quality assurance on manufactured parts. In order to do this, they needed to acquire data to generate and train the machine learning models, as well as obtaining a custom IoT framework that could support the machine learning model on the edge in order to facilitate device management and deployment of new models with new features.
The Solution
TensorIoT designed a physical compartment to facilitate image capture as part of a machine learning workflow, allowing highly controlled images to be captured and fed into a datalake. TensorIoT also designed an accompanying ingestion pipeline and workflow to facilitate easy tagging of all metadata (part #, ground truth defect identification, etc) to help the customer consolidate the data needed to train computer vision machine learning models to recognize defects.