Devang Roopesh Mehta

• Refactored deep learning based anomaly localization code to deploy with azure function on the cloud and on the edge device Nvidia Jetson Orin Nano Development Kit • Benchmarked inference on the edge by quantizing the PyTorch model using Torch-TensorRT python package, starting with standard docker containers and following up to native deployment on the edge • Prepared use-case examples of applications of data science as demonstrations, exercises and assignments

Defect localization on wooden boards with autoencoder based deep learning • Implemented sliding window, feature extraction with pre-trained VGG16 neural network and Kmeans clustering to classify images • Implemented Kmeans clustering to segment each image of Describable Textures Dataset for artificial anomalies • Developed a pipeline to generate anomalous data to train the neural network based on the autoencoder architecture • Familiarized with Pytorch framework and its remote execution in Docker container

• Worked with common protocols and libraries (Garbled Circuits, Secret Sharing) in the field of Secure Multiparty Computation (SMC) for industrial use in cross-company data exchange. • Developed a software prototype for a distributed system to perform computation on sensitive data from multiple companies using Docker, Python, and MP-SPDZ • Developed APIs and prototype websites using FastAPI, Jinja2, HTML, and Python

• Developed software for an optimised algorithm to extract foreground in a 3D Point Cloud in C++ • Visualized the colour for the 3D Point Cloud for easier user accessibility in the user interface

• Literature study on BRAM read and write access for use with the Red Pitaya hardware • Modified the FPGA design in Vivado Design Suite to measure and save the frequency of an external source

• Developed test setups and necessary software for capturing images with an associated illumination. • Developed software for the optical inspection of component surfaces in an Anaconda environment with functions in Python of the OpenCV and Tensorflow libraries • Generated synthetic images for training in neural network and modified a neural network • Implemented the U-Net model for 3 class image segmentation

• Optimized the algorithms and the software in C++ in terms of execution time and • Modified an emulation program for testing the communication between the software set up as a service and the PLC • Developed a mobile test stand for testing and demonstrating the system

• Selected and integrated functions of the PCL and the Realsense camera library in C++ • Created a prototype test stand for testing and demonstrating the system • Demonstrated the system in front of the industrial partner TGW using real objects

• Explained the practical exercise to the participants

Detection of Rotational Deviation of Objects based on 3D Camera • Developed a pipeline to acquire point clouds from Intel Realsense 3D camera and applied Iterative Closest Point (ICP) algorithm to compute transformation of objects. • Successful and accurate spatial transformation detection of small objects using computer vision techniques. • Software development using C++, Point Cloud Library (PCL), IntelRealsense SDK 2.0.

• Training for participants in various CAD software packages • Demonstration of the course to the prospective candidates • Focus on part design, assembly design, surface design, sheet metal design and drafting