#hackingforfuture

Machine Learning (ML)-based models are often black-box. To use ML model in industrial setting it should be fair and reliable. Metrics like accuracy-score or r2-score makes it reliable, but one cannot say anything about the fairness of the model. Model fairness and interpretability are critical for data scientists as well as production engineer to explain their models and understand the value and accuracy of their findings. ​

The objective of this task is to interpret a trained ML model using CXPlain and SHAP library.​

You will be given a tabular dataset for a classification task. Using the given dataset, a ML model should be trained to predict the target value. Using this trained model, feature importance for the input features should be calculated with the help of CXPlain and SHAP Model interpretation library and compared quantitatively. ​

Sub-Tasks:

Vibration sensors are frequently used in smart factory applications. Standard thresholds for alarms are usually are configurations derived from the ISO10816 standard. This is a good starting point, but it does not reflect real sensor applications and installations, especially when supervising several sensors at several similar machines.

In reality the mounting position and orientation of sensors at different machines is not identical. In reality “natural” vibrations of two or more “identical” machines are not identical. In reality many vibration influencing variables are ”machine individual”, e.g. on cutting machines there is influence of the actual cutter condition, the “age” of the coolant, the overlay of pick-and-place movements to the “individual” vibration signal of the “individual” machine.

The task therefore is to develop, verify and deploy an algorithm, that is able to individually generate thresholds through learning data from the first week after the installation of the machine. So-to-say delivering a machine individual zero-line for alarm-threshold out of the time-series vibration data.

Sub-Tasks:

Augmented Reality (AR) and Mixed Reality (MR) are increasingly used in manufacturing environments, for example to support manual work with visualizations of work instructions. With the increasing spread and utilization of AR and MR applications in manufacturing, a need arises to create 3D models which accurately depict the characteristics of real life objects. If, for instance, assembly instructions for an engine are to be provided to assembly workers, a precise model of the engine might be useful in order to visualize assembly steps with a high accuracy. However, the creation of accurate 3D models is a time consuming task so that an automation of this work step would add significant value.

The objective of this task is to find a way to automate the creation of a 3D model of a real reference object which in this case is a ordinary PC keyboard.

User input can be photographs, videos, camera scanning process of the reference keyboard. The output should be a textured 3D mesh of the reference object in GLTF format in its correct physical size.

The requirements are: