Tech mahindra
Tech mahindra

Latest Computer Vision applications in Engineering

Posted by: Rana Banerjee On May 31, 2018 12:08 PM facebook linked in twitter

The last few years have seen great advancements in the field of computer vision that has made automation of many manual engineering tasks and processes feasible.

Thanks to deep learning and particularly Convolutional Neural Networks (CNNs), by 2015, we had computers doing a better job of image classification than humans.

In 2016, with algorithms and solutions like faster RCNN, YOLO and SSD, we have seen various light weight implementations of real-time object detection and localization which opens up great applicability in areas like driverless cars, surveillance, etc.

We can now automate tasks like image captioning, classify actions and much more, with slightly more complex architectures.

We also see a transfer of technology from basic science research to real world application at a very fast pace. For e.g. in 2006, Fuji films in their digital cameras implemented the “voila-jones face-detection algorithm” (published in 2001) in a span of just 5 years.

We foresee similar pace of applicability of the latest computer vision technology to the real world, particularly for engineering and inspections. Few such implementations of computer vision across by Tech Mahindra are discussed below:

Implementation # 1: Automated creation of Bill of Materials from Design Specs

Problem Statement: The current Bill of Materials (BOM) creation process involves a person to manually scrutinize the elaborate designs and collate details. The person needs to comprehend the BOM parts and their quantities based on scattered and complex annotations in the designs. Manual effort required for collating such details is huge and prone to errors.

Solution: With the Implementation of a lightweight computer vision solution using various image analysis algorithms like HoughCircles and HoughLines along with OCR to detect numbers and text, the detection of the part balloons and deriving quantities can now be automated quickly and efficiently.

Business Benefits

  • Implementation of the solution helps in achieving up to 80% effort savings
  • Reduced time for processing along with higher accuracy

Implementation # 2: Automated air flow testing for aircraft wings

Problem Statement: One of the final steps involved in airflow visualization testing for aircraft wings requires checking for changes in the direction of flow cones attached at various positions in an aircraft wing during test flights. A video of the wing during this test flight is recorded which is later manually inspected for deviations as various parameters are altered. This requires a significant amount of manual intervention and delay in processing of the test results. In addition, there is an unavoidable possibility of manual error as this is a very mundane task.

Solution: A solution using edge detection algorithms and filters helps identify the flow cones for each of the video frames. The next step involves detecting these flow cones to be processed by a CNN based solution to classify if it is deviated or not.

Tying these components with other video processing components enables us to automatically detect deviations in the flow cones and mark it accurately along a time series. Data is then pushed to an Analytics engine to generate dashboards that provide detailed insights.

Business Benefits

  • More than 50% reduction in manual intervention along with an increase in accuracy.
  • Deeper insights, which can help with RCAs in future.

Real time computer vision solution for aircraft inspections

Problem Statement: Airline inspectors face difficulty in inspecting the aircraft for damages and repairs. A typical structure inspection may take over 2 hours. There is a possibility of the inspector, missing damages at the initial stage due to inaccessibility or oversight, which may lead to higher cost of repairs in the future.

Solution: A computer vision solution using CNN to identify and locate damages like corrosion dents etc., on streaming video of the aircraft exterior inspection area.

An HD camera installed on a drone or a robotic arm to capture high-resolution images and videos of the entire aircraft. During the image capture process, real time AI cognition helps the inspector to identify damages like corrosion etc. The same cognition can be done offline and semi-automatic assessment reports generated.

Business Benefits

  • Decreases the overall time for routine inspection and increases the accuracy of inspections
  • Reduction of the inspection time by up to 90%

Detection of early signs of damages like, corrosions, dents etc., to reduce overall repair cost.

About Author

Rana Banerjee, Principal Consultant – AI

Rana is an AI evangelist and deep learning enthusiast with over 15 years of diverse experience around automation. During his career, he has created many, ahead of the curve, business value driven solutions using emerging technologies. Rana is one of the core Principal Consultants in Tech Mahindra’s AI competency and supports multiple verticals and clients around all regions. He has specialization in implementing deep learning based solutions for various problem areas including Computer Vision.

References

1. https://www.theguardian.com/global/2015/may/13/baidu-minwa-supercomputer-better-than-humans-recognising-images

Tags: Connected Platforms & Solutions
 
(*) symbol is mandatory field
Name:
* Email Address:

Comments

(*) symbol is mandatory field

Post a Comment

* First Name:
* Email Address:
URL:
Comments:
 
Image Code
* Enter Image Code



Contact Us

Archives

For further information please write to connect@techmahindra.com

For further information please write to connect@techmahindra.com