Transform-CV is ADS’ deep learning-enabled computer vision application for quality assurance applications. Transform-CV analyzes images and makes real-time inferences, providing customers with crucial insights into product quality to improve manufacturing processes. Based on ADS’ value-adds to open-source software, Transform-CV is readily tailorable to a range of customer applications, from cable harnesses to microscopic component defects. Transform-CV is based on commercial hardware and support both PCIe and PXIe platforms to address wide range of manufacturing test & measurement environments that may be employed to augment or replace manual defect analysis, dimensional analysis, and component classification.

How the Transform-CV Can Help You

Defect Analysis

With sufficient training, all types of defects can be recognized.

Dimensional Analysis

Measure part dimensions to tight tolerances.

Component Classification

Verify the presence and location of specified components.

Labeling

Verify label orientation and quality.

Features

Performance

Higher throughput with improved accuracy, precision, repeatability vs. manual Q/A.

Easy-to-Use

Interoperable with popular manufacturing software to support existing customer workflows.

Flexible

Optical components, AI model, and scalable GPU support for a wide range of applications with support for customer-enabled training on new features or products.

Cost-Effective

Open architecture solution eliminates proprietary integrated computer vision platforms.

How it Works

IMAGE CAPTURE

Transform-CV employs one or more high-resolution, high frame rate fixture mounted or handheld cameras via USB, Ethernet, FPD-Link, Camera Link or GMSL interfaces, to capture the subject imaged or video, which is then preprocessed by the system's CPU or GPU prior to inference processing.

TRAINING DATA

Transform-CV employs synthetic or customer-provided training data in the form of annotated images or videos, which feature the objects to be analyzed, their characteristics & identify the presence & location of the defects or aspects that the system must detect.

REAL-TIME INFERENCE

Transform-CV references the image or video against the library of training data to make inference & analysis in real-time (infers more accurately through increased quantity & quality of training data supplied).

PASS OR FAIL

The quantity of the requested feature, such as defects, part count, or dimensions, are displayed to the operator, along with a PASS or FAIL verdict (if necessary) to denote whether the identified quantity matches the user-specified desired quantity or parameter. The system also displays confidence level, frames per second, & parts per second supported.

THROUGHPUT SIMULATION

For assembly-line based production applications, Transform-CV supports the use of different means for emulating throughput. The system identifies dimensions & counts parts on the emulator at frame rates, part counts & other factors identified by the customer. By training the models on an emulated line, system may be tailored to fit a wide range of vision-based Q/A production applications.

INTERFACING TO Q/A SYSTEMS

For each unit tested, the system logs the inference results and can communicate them to the customer's manufacturing Q/A software via messaging or API, if requested.

Software & Hardware

Software

  • Python-based
  • OpenCV V4.6.0 Computer Vision Library
  • YoloV5 Deep Learning Algorithm
  • PyTorch V1, 13, 14 Machine Learning Library
  • PyQT V5.15.9 Graphical User Interface (GUI)

HARDWARE

PXIe Transform-CV Systems

  • PXIe 1095 Chassis
  • RADX NVIDIA Catalyst GPU for PXI Module
  • Monitor
  • HD Camera
  • Illuminator
  • Turntable-based Assembly Line Emulator

PC Transform-CV System

  • GPU-embedded Computing Platform
  • Monitor
  • HD Camera
  • Illuminator
  • Turntable-based Assembly Line Emulator

Partner with Acquired Data Solutions