Data Scientist - Intern
Airbus
Bengaluru, Karnataka, India
Job Description: Airbus Innovation Centre - India & South Asia: Airbus Innovation Centre - India & South Asia is responsible for industrializing disruptive technologies by tapping into the strong engineering competencies centre while also leveraging and co-creating with the vibrant external ecosystems such as big Tech Enterprises, mature startups/MSMEs, national labs & universities and strategic partners (customers, suppliers etc.) The technology areas that the Innovation Centre focus on are - Artificial Intelligence, Industrial Automation, Unmanned Air Systems, Connectivity, Space Tech, Autonomy, Decarbonization Technologies etc. among others. Airbus Innovation Centre in India is 1 among 3 Innovation Centres globally for Airbus with a strong focus on A.I. and Digital Engineering. We build products from the ground up with the help of stakeholders from within Engineering and Digital competence centres (in addition to the external stakeholders mentioned above) to deliver operational excellence and contribute to the Innovation & Technology roadmap of the organization. Title: High-Dimensional Constrained Design of Experiments for ML applications Introduction: Surrogate models are used in the area of multidisciplinary analysis and optimization. These surrogate models have the advantage over simulations that they can approximate the effects of parameter variations in real time. This enables savings in terms of time and costs when developing a new aircraft or aircraft variants. In addition, more variations of the parameters can be performed. The optimal point of design can be searched for and the necessary knowledge about the interrelationships of the parameters at the point of design is provided. These surrogate models are (in our use cases) Machine Learning (ML) models. Accordingly, a data set must be available for the training of these surrogate models. The Design of Experiments (DoE) methodology is used to create an optimal data set for this purpose. The goal is to map ‘m’ simulation inputs to ‘n’ simulation outputs. The larger goal is to create an adaptive DoE which, based on the needs, either finds the optimum dataset or does active learning to increase the performance of the subsequently built surrogate model. Since the simulations are performed sequentially, it is possible to use the already calculated data points to determine the position in the design space where data points have the largest amount of information. However, before investigating further we want to focus again more on the DoE. In order to decrease the design space, i.e. the space that optimization algorithms have to search through, a constraint DoE was developed. The aim of this work is to reach practical application of this constraint DoE by adding further input dimensions. A full description of an existing DoE to translate into a constraint DoE is available. It works today with a cubical “base” DoE whose domain is transformed, in a post-processing step, to comply with the underlying constraints. The problem with this methodology though is that the final sample distribution is not homogeneous. This again leads to potential bias and unnecessary large sample sizes for ML applications. Key Responsibilities: Today we have two constraint DoE's: 1. zerofuel_mass, zerofuel_cg, fuel_weight 2. altitude, speed, vertical_loadfactor Learn about the existing DoE libraries (OpenTurns, JohnDoE) by producing unit tests, docstrings, and documentation For the 1st DoE you switch from a temporary Fuel vector implementation to the official implementation Enhance the 1st DoE by increasing the complexity through an additional dimension: fuel density Enhance the 1st DoE by increasing the complexity through splitting dimension fuel_weight into: re_fuel_weight and de_fuel_weight Merge the 1st and 2nd DoE into one DoE add all missing independent further dimensions to reach practical applicability. Qualifications: Strong Python skills. Statistics background Experience with Data Wrangling and Preprocessing. Experience in Design of Experiments for Data Generation. Proficiency in version control systems (e.g., Git) and software development best practices. Machine Learning & Deep Learning Model Development Cycle. EDUCATION: M.Sc . / M.Eng. in Computer Science, Date Engineering, Mathematics, Aerospace
