Dandelion Plan

Introduction

In the search for interplanetary information, aerospace probes demand a high value for manufacturing and often unforeseen occur such as equipment damage, communication failures, during the collection process.

Goal

To address these issues, Dandelion plan comes with a lower cost design, improved communication, and reduced loss of information being collected and analyzed.

Detailing

Mini probes meet the following specifications:

• 12 cm each mini probe;
• About 30 mini probes are sent to the specific planet;
• All mini probes sent data to the satellite;
• A set of mini probes, contains unique sensors for each objective;
• Data sharing between mini probes;
• Resilience between the mini probes;
• Mapping and coverage of larger areas;
• Prevents data loss between each other.

The probes arrive by means of a parachute that generates the braking, and during this braking the spheres come loosening and separating. By separating the probes they form clusters and begin to communicate.

Each data collected by the probes is shared with each other, preventing total data loss if any probe is damaged. After sharing the collected data occurs sending to the satellite.

Technologies used

• Python;
• Scikit-Learn;
• Numpy;
• Pandas;
• After effects;
• Google slide;
• Adobe Illustrator;
• Remote control motor;
• Machine learning;
• Machine Learning Techniques.

We use K-means to start the cluster for work between the mini probes on the planet that were launched.

In data mining, k-means clustering is a Clustering method that aims to partition n observations among k groups, where each observation belongs to the nearest average group. This results in a division of the data space in a Voronoi Diagram.

The problem is computationally difficult (NP-hard), however, there are efficient heuristic algorithms that are commonly employed and quickly converge to an optimum location. These are generally similar to the expectation maximization algorithm for mixtures of Gaussian distributions, through an iterative refinement approach used by both algorithms. In addition, both use cluster centers to model data, however, k-means clustering tends to find comparable spatial extent clusters while the expectation maximization mechanism allows for different shapes. Reaching 91.37% accuracy.

We also used the meteor dataset granted by NASA to eliminate the missing ones, remembering that the algorithm discussed here will be used internally in the mini probe that is Kohonen's Self-Organized Map.

A typical neural network with competitive learning is a single-layer network (one or two-dimensional) where all neurons receive the same input.

Each neuron computes its activation level by multiplying its weight vector by the input vector in the usual way. The neuron that has the highest activation level is called the winner and only it will have nonzero activity at the network output, ie the input pattern being presented to the network will trigger only one neuron from the neural network. In this case we do not have accuracy values and it is worth remembering that in this case the neuron was taught to control the missing value of the data from each mini probe.

Thus making it difficult to lose data between the mini probes and sending them.

Cost and benefit

According to the survey of the proposed requirements in the mini probes and the reduction is almost 40% of the cost of a current probe. Example Curiosity Rover currently costs 2.5 billion and mini probes set out to lower their cost as put.

Conclusion

With the selected techniques, we can say that we have been able to reduce data loss through rebuilding optimization and using Machine Learning, and of course generate a lower cost for NASA and use a resilience-based technique that does not make you lose the probe. once, and has small probes to explore the site.