Artificial Intelligence Developer

• Apr 2023 - Present

Master Thesis Student

• Apr 2022 - Apr 2023

Really being able to proof that a mathematical claim is correct or that software code is executing according to specification is hard. But there are tools, proof assistants, supporting the building of proofs. I would like to get experienced in using such a tool. In my opinion there is a bright future for this knowledge when designing security and safety systems. To write code that is 100% correct, and this proven mathematically. I have chosen to study and use the Coq tool, because I have read that Airbus used it to analyze critical code. For the thesis I am writing Gallina code for Coq to verify claims and proofs of my professor in papers on hybrid systems. He and others define a mathematical framework for such systems. Hybrid systems are composed of digital and analog modules within an environment that also can be modelled. For instance, airplanes, cars, robots, electrical systems. When everything is captured in mathematical models then the behavior can be calculated. This does not have to be simulation. The system can, for instance, be interrogated on the presence of system deadlocks or other undesirable system behaviors. Using Coq to build proofs is meticulous work. Progress is slow. Lots of stamina is necessary. Also it is not mainstream to incorporate this in a development flow. My expectation is that it will take about 2 years of regular practice to become fluent with this tool. However, in some circumstances its usage can be powerful. When lives get at stake, or lots of money. And with increased fluency it becomes cheaper and can be used to build robust code for less critical systems. Show less

Student Intern

• Sep 2021 - Feb 2022

At EKK Eagle Simrax I have worked on a predictive maintenance problem. This company produces parts for the automotive industry. There is a conveyor belt there with a lapping module that requires regular maintenance. A few times during an operator workshift of 8 hours. The operator himself decides when to do this. At regular time intervals or reacting to an increase in scrap production. During maintenance the entire belt is stopped. Conveyor belt modules store much data about productivity and sensor readings. I started with talking to the stakeholders. Manager innovation, database expert, operators. By analyzing the data I confirmed that the maintenance on the lapping machine costs relatively much time and the scrap detection is relatively high just after the lapping module. Smart scheduling of lapping module maintenance could boost productivity. When to do the maintenance then? – Not too often, but often enough. Obviously. I have investigated if present available data can be used to predict an increase in scrap production later. If this can be done successfully then the operator can stop the belt just before this happens and do the maintenance. For this, I have set up and trained multiple artificial neural networks (ANNs). A variation in architectures in order to search for the best in predicting the future. Pyhton with PyTorch were the tools of choice. Other packages where needed. I have successfully set up a framework that is good at predicting future slack. Evidently predicting the more distant future comes at the cost of loss in accuracy. I have given advice on extra sensor readings that could help to get better results. The predicting ANNs can be used in parallel to the existing metrics. Expert eyes can tune the framework to deliver the best ANN for the job. The database expert and innovation manager were enthusiastic during the entire internship. I have showed then what AI could do and delivered a framework that could be developed further. Show less

Senior Digital ASIC Designer

• Nov 2000 - May 2020

◼ Writing robust, transparent, low maintenance design code ◼ Mitigating project risks by setting up early inspection tooling and tracking reports ◼ Scrutinizing design quality with various verification techniques, such as: directed testing, constrained random testing, combinatorial testing, fault injection, formal testing ◼ Analyzing notorious structures: clock generation, reset strategy, clock domain crossings ◼ Delivering security level CC-EAL 6+ ICs for eBanking and eGovernment products ◼ Identifying security weaknesses using formal verification on NXP's EdgeLock™ SE050 for enhanced IoT security ◼ Assembling and managing design/verification checklist Show less