<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Chloe Wang on Medium]]></title>
        <description><![CDATA[Stories by Chloe Wang on Medium]]></description>
        <link>https://medium.com/@chloeewang?source=rss-25cb921bfe24------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*HmsHH963P3Mt71slq3CbtA.jpeg</url>
            <title>Stories by Chloe Wang on Medium</title>
            <link>https://medium.com/@chloeewang?source=rss-25cb921bfe24------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 00:38:25 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@chloeewang/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Building A Mask R-CNN To Detect and Segment Breast Tumors]]></title>
            <link>https://chloeewang.medium.com/building-a-mask-r-cnn-to-detect-and-segment-breast-tumors-8f262eab16ad?source=rss-25cb921bfe24------2</link>
            <guid isPermaLink="false">https://medium.com/p/8f262eab16ad</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[breast-cancer]]></category>
            <dc:creator><![CDATA[Chloe Wang]]></dc:creator>
            <pubDate>Tue, 21 Dec 2021 06:05:49 GMT</pubDate>
            <atom:updated>2021-12-21T06:10:56.947Z</atom:updated>
            <content:encoded><![CDATA[<p><em>I’d recommend reading my </em><a href="https://chloeewang.medium.com/a-look-into-neural-networks-and-deep-reinforcement-learning-2d5a9baef3e3"><em>article on neural networks</em></a><em> to understand what convolutional neural networks are before reading this one.</em></p><p>Breast cancer is one of the most fatal and common cancers in women, killing about<strong> </strong><a href="https://www.cancer.org/cancer/breast-cancer/about/how-common-is-breast-cancer.html"><strong>1 in every 39 </strong>women per year</a>. However, early breast cancer detection can drastically increase the patient’s chance of survival. Detecting breast cancer at an early stage can bring the survival rate up to<strong> </strong><a href="https://www.carolmilgardbreastcenter.org/early-detection"><strong>93%</strong></a> within the first 5 years.</p><p>Thousands of people work to analyze breast sonograms to try to detect breast cancer early — but what if we could automate it? To demonstrate this concept, I built a mask R-CNN to detect and segment breast tumors from ultrasound images.</p><h3>Object Detection and Image Segmentation</h3><p>Before going over R-CNNs, it’s a good idea to run down two different computer vision tasks — <strong>object detection and image segmentation.</strong></p><p>Object detection combines<strong> image classification</strong> (labeling images) and <strong>object localization</strong> (drawing a bounding box around objects in an image).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/519/0*9jmf-svTnN_866pH" /></figure><p>However, object detection doesn’t tell us anything about the shape of an object. Often times, just knowing the class and bounding box coordinates is too vague. Thus, image segmentation is the process of <strong>creating a pixel-wise mask</strong> for each object in the image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*frbTGBou7UNhdQEz" /><figcaption>From <a href="http://cs231n.stanford.edu/"><em>cs231n.stanford.edu</em></a></figcaption></figure><h3>What Are R-CNNs?</h3><p>R-CNNs, or <strong>region-based convolutional neural networks</strong>, are one of the first successful applications of CNNs in object localization, detection, and segmentation. Standard R-CNNs have three parts:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/994/0*BQ_MTER9t8U3KvR1" /><figcaption>From <a href="https://arxiv.org/abs/1311.2524">Rich feature hierarchies for accurate object detection and semantic segmentation</a></figcaption></figure><ol><li>Region proposals (possible bounding boxes for objects) are generated from the input using computer vision techniques</li><li>Features are extracted from each candidate region using a CNN</li><li>Features are classified based on known classes</li></ol><p>However, this original architecture is<strong> slow and expensive</strong>. A model may operate on about 2,000 proposed regions at a time, and since the CNN-based feature extraction is required to pass on every candidate region generated from the region proposal algorithm, it can become very slow.</p><p>Consequently, other forms of R-CNNs were developed to speed up the process.</p><h4>Fast R-CNN</h4><p>A fast R-CNN improves on the original architecture by acting as a <strong>single model</strong> instead of multiple. In a fast R-CNN:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*lbV8ZOrd9DpNIvKW" /><figcaption>From <a href="https://arxiv.org/abs/1311.2524">Rich feature hierarchies for accurate object detection and semantic segmentation</a></figcaption></figure><ol><li>A set of region proposals are passed through a CNN, which is pre-trained for feature extraction</li><li>At the end of the CNN is a custom layer called the region-of-interest pooling layer (ROI pooling) that extracts features specific to a given input candidate region</li><li>Output of CNN is interpreted by a fully-connected layer</li><li>This model then divides into two outputs, one for class prediction through a Softmax layer and the other with a linear output for a bounding box</li></ol><p>This model is much faster than the standard R-CNN architecture because it is only one model instead of three, but it still requires many candidate regions to be proposed with each input image.</p><h4>Faster R-CNN</h4><p>The faster R-CNN was designed to <strong>speed up both training and detection</strong> from previous R-CNN architectures. It proposes and refines region proposals as a part of the training process, which are then used in par with the fast R-CNN model in a single model design. Compared to previous architectures, it <strong>reduces the number of proposed regions and accelerates the operation time to near real-time</strong>. It works by:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/948/0*K0ZIIm0rhKR93_w3" /><figcaption>From <a href="https://arxiv.org/abs/1506.01497">Faster R-CNN: Towards Real-Time Object Detection With Region Proposal Networks</a></figcaption></figure><ol><li>Using a region proposal CNN</li><li>Using a fast R-CNN to extract features from the proposed regions and create bounding boxes and class labels</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*9WgRm8m9Lw9QVVQ1" /><figcaption>From <a href="https://arxiv.org/abs/1506.02640">You Only Look Once: Unified, Real-Time Object Detection</a></figcaption></figure><p>Both of these components use the same output of a CNN and are trained at the same time. The region proposal CNN essentially acts as an attention mechanism for the fast R-CNN, which tells the second network where to pay attention.</p><h4>Mask R-CNN</h4><p>Mask R-CNNs builds on top of the faster R-CNN object detection architecture by not only predicting the<strong> class and bounding box coordinates</strong> of an object, but the <strong>object mask</strong> as well. The object mask can be imagined as a silhouette of an image; they’re a much finer spatial layout of an object. Mask R-CNNs work by:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yt6PKETBU9eGvt4O" /><figcaption>From <a href="https://arxiv.org/abs/1703.06870v3">Mask R-CNN</a></figcaption></figure><ol><li>An image is passed through a CNN that returns a feature map for the image</li><li>A region proposal network is applied on the feature maps, returning object proposals with their confidence score</li><li>A ROI pooling layer is applied on the proposals to make all the proposals the same size. Mask R-CNN ROIs work by computing the intersection between the predicted boxes and ground truth boxes. If the computation is greater than or equal to 0.5, that area is considered a region of interest.</li><li>The proposals are passed to a fully-connected layer to classify and output bounding boxes for objects, as well as returning a mask for each proposal</li></ol><p>Mask R-CNNs are so far one of the most successful R-CNN architectures because they’re<strong> simpler to train, out-perform all previous architectures, are very efficient, and are easy to generalize to other tasks.</strong></p><p>The key element of a mask R-CNNs efficiency is the pixel-to-pixel alignment, which isn’t found in fast or faster R-CNNs. The mask R-CNN has the same two stage procedure as the faster R-CNN. However, in the second stage, the mask R-CNN outputs a binary mask for each ROI in addition to the class and box offset. This contrasts other systems that where classification is usually dependent on masking systems.</p><h3>Building A Mask R-CNN</h3><h4>At a high level</h4><ul><li>I created a mask R-CNN, which combines computer vision and deep learning, to detect and segment breast cancer tumors from ultrasound images.</li><li>I used a model built on <a href="https://github.com/matterport/Mask_RCNN">FPN and ResNet 101 from <strong>matterport</strong></a>. I also started with <strong>pre-trained weights from MS COCO</strong>, though they were not my final weights.</li><li>I set the confidence level so that the model skipped region proposals with &lt; 90% confidence. Then I trained the model for 30 epochs, with 51 steps per epoch. Training took about <strong>1.5 days</strong> in total.</li><li>I sampled my data from <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6906728/">here</a>, but I didn’t use all of the images (I used about 200 benign and malignant tumor images). I annotated my images and turned them into json files using<strong> </strong><a href="https://www.robots.ox.ac.uk/~vgg/software/via/"><strong>VGG Image Annotator</strong></a><strong>.</strong></li><li>You can access the code in <a href="https://github.com/chloeewangg/breast_cancer_mrcnn/blob/main/README.md">this repository.</a></li><li>This project was based off a <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6531264/"><strong>paper from 2019</strong></a> that performed a similar task, using a mask R-CNN on sonograms and classifying benign, malignant, and normal tissues.</li></ul><h4>Project Walkthrough</h4><p>Since my project is built on a model from <em>matterport</em>, most of my code is actually overwriting some of what they defined in their source code so that I could implement a custom dataset. For context, I used Jupyter Notebook with tensorflow 1.5.0 and keras 2.8.0 in an Anaconda environment.</p><p>These are the modules you’ll need to import. Here I also assigned the root directory of my project and the weights path.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rhWpMUwJIdx3QNWyz3EIdA.png" /></figure><p>Then I defined a class that basically allowed me to override some of the values defined in the source code. Notice that I have 3 classes, which includes the benign tumors, malignant tumors, and the background. Also notice that I basically have the model skip region proposals that have a confidence of &lt; 90%.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/1*GN-KNqu4CPfbLWseJDmCCg.png" /></figure><p>Here I defined another class to let me implement my custom dataset. In the <em>load_custom </em>function I defined my classes and the annotations for the ground truth. I also added in images and defined variables based on the annotations I made.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xUF_ygOna3bjambxNl7r8A.png" /></figure><p>To train my model, I loaded the custom training and validation datasets. I set the number of epochs to 10, but I repeated this process 3 times so that the model ultimately trained for 30 epochs (~ 33 hours). I called pre-trained weights from MS COCO to start the training process, although they weren’t my final weights. In the end, the model performed decently, but for optimal performance I would recommend training for at least 50 epochs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XJTlWw7w7DIlkNXpeeBUHw.png" /></figure><p>After training, I tested the model. I redefined some values like the root directory and the custom dataset directory. I also loaded in the configuration, model, and newly trained weights. Here, I select a random image from my validation dataset and run the model on it. Then I displayed the results of the mask R-CNN as it ran through the different layers of the CNN. The final result is a bounding box and mask for the tumor in the image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*D_FdPvfEMvOa7N9lvWNn-w.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NZ2Fa7CYKFEGkdgz5wWKCA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/884/1*Lykmk8EdXuhfIu_QPy-c_A.png" /><figcaption>Prediction of a benign tumor</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/608/1*s5-J5l6l4lyr2o7YVwJewA.png" /><figcaption>Ground truth mask</figcaption></figure><p>There are a few drawbacks to this project, including:</p><ul><li>A fairly small sample size (only about 100 images for the training and validation sets, or 200 images in total)</li><li>Only classifying benign and malignant tumors and not normal tissue</li></ul><p>But overall, it’s a solid demonstration of concept and would have beneficial applications if further developed. You can access the GitHub repository of this project <a href="https://github.com/chloeewangg/breast_cancer_mrcnn/blob/main/README.md">here</a>.</p><p><em>Have feedback or questions? Send me an email at chloewang.lv@gmail.com and I’ll be happy to respond!</em></p><p><em>You can check out more of what I’m up to in my quarterly newsletters. Sign up </em><a href="https://docs.google.com/forms/d/e/1FAIpQLSeAEWbVJplZ_njSIDfs9CWegCagrhYfq2o6789DDh3hD8cwpQ/viewform?usp=sf_link"><em>here</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8f262eab16ad" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Using Deep Reinforcement Learning To Play Atari Space Invaders]]></title>
            <link>https://chloeewang.medium.com/using-deep-reinforcement-learning-to-play-atari-space-invaders-8d5159aa69ed?source=rss-25cb921bfe24------2</link>
            <guid isPermaLink="false">https://medium.com/p/8d5159aa69ed</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[reinforcement-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[deepreinforcementlearning]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Chloe Wang]]></dc:creator>
            <pubDate>Fri, 19 Nov 2021 04:30:20 GMT</pubDate>
            <atom:updated>2021-11-19T04:30:20.831Z</atom:updated>
            <content:encoded><![CDATA[<p><em>If you aren’t familiar with DRL, I would recommend reading </em><a href="https://chloeewang.medium.com/a-look-into-neural-networks-and-deep-reinforcement-learning-2d5a9baef3e3"><em>this article</em></a><em> before this one. It provides context on many concepts that were used in this model that I do not discuss here.</em></p><p>Deep reinforcement learning (DRL) is a subset of machine learning that essentially combines neural networks with reinforcement learning to help software agents learn to achieve their goals. It brings together function approximation and target optimization, allowing agents to map states and actions to the rewards they lead to.</p><p>Based on this concept, I was able to teach a DRL agent how to play Atari Space Invaders.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/284/0*DMHeY0rDf7dCmKAE" /><figcaption>My agent playing Atari Space Invaders</figcaption></figure><h3>TL;DR</h3><ul><li>I was able to teach an RL agent how to play Atari Space Invaders using concepts from both RL and DL.</li><li>I used <a href="https://openai.com/blog/gym-retro/">OpenAI Gym Retro</a> to create the environment that my agent played in. It’s from an initiative that encouraged DRL design across many different but similar environments.</li><li>The neural network in this model is used to process frames from the game to understand where objects are and what the agent is doing. It uses 3 convolutional layers and 3 dense (fully-connected) layers to do so.</li><li>My model trained for 10k steps (~4 hrs) and played decently, but Google DeepMind recommends training for 10M-40M steps for optimal playing.</li><li>Here is a <a href="https://youtu.be/9ZGMPWzNleg">video</a> of me explaining the project and watching the agent play.</li><li>This project was a replicate of <a href="https://github.com/nicknochnack/KerasRL-OpenAI-Atari-SpaceInvadersv0/blob/main/Space%20Invaders%20Walkthrough.ipynb">Nicholas Renotte’s DRL model</a>.</li></ul><h3>Model Walkthrough</h3><p>This model used TensorFlow 2.3.1, keras-r12 (a reinforcement learning library), and <a href="https://openai.com/blog/gym-retro/">OpenAI Gym Retro</a>. OpenAI Gym Retro was released a few years ago and can work in over 1000 retro game environments.</p><h4>Creating The Environment</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/871/1*FbcOH-lIu0eGrbM7tLAFRA.png" /><figcaption>‘NOOP’ is no action.</figcaption></figure><p>Here, I created the environment. Then I pulled the height, width, and channels out of a particular environment frame (equivalent to part of a state) so that I could pass it through a neural network. I also grabbed the actions and unwrapped them so I could see what they were.</p><p>Then I tested the environment with 5 episodes (5 games) to see how it performed. In each episode, I reset the environment and set the score to 0. As the episode was running, the agent chose a random action and provided information like the reward and state. I then used this information to output the score.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/880/1*TK3pzgMDgQc_-qD_ccbTUA.png" /><figcaption>In general, the agent does okay, but that’s just because it’s taking random actions.</figcaption></figure><h4>Building The DL Model</h4><p>After that I started building the deep learning model with Keras, NumPy, and TensorFlow. I defined a model that passes through the height, width, channels (pulled from observation space) and actions (pulled from action space). The shape of these parameters defines the model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*88FVtXk8pJwN1lA24yQI3A.png" /></figure><p>Inside the function I assigned <strong><em>Sequential()</em></strong> to <strong><em>model, </em></strong>which passes the frames through sequentially. Then using <strong><em>model.add</em></strong>, I could stack layers in the DL network.</p><p>The first layer is<strong> <em>Convolution2D</em></strong>, where I passed through a few parameters, including:</p><ul><li>The number of filters (32). The goal was to train the filters so that they could detect objects in the frames, like the enemies.</li><li>The size of the filters (8x8 units) and the number of strides.</li><li>The ReLU activation function and the shape of the frame.</li></ul><p>Note that in this first layer, I actually pass through multiple frames from the memory data frames. The second and third layers are also convolutional layers, with some different filters, filter sizes, and strides.</p><p>Then I flattened all the layers into a single layer, allowing me to pass it through a dense (fully-connected) layer. The first dense layer has 512 units. The second dense layer compresses this slightly and is 256 units. The third dense layer has the number of actions, which is 6 units.</p><p>Finally, I build the model by assigning the <strong><em>build_model</em></strong> function to <strong><em>model</em></strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/884/1*QjQnq-sjUQSA3AjifNVhHQ.png" /></figure><p>Here, you can see the full model in a more readable form. Notice how the size of the outputs in the convolutional layers decreases, since the goal is to basically compress the inputs. The flatten layer then transforms it, allowing the dense layers to fully compress the image to just 6 actions. The right column shows the parameters, and at the bottom, the total number of parameters; 34,812,326. This goes to show why it’s so useful to incorporate the data interpretation abilities of deep learning in reinforcement learning models.</p><h4>Creating The RL Agent</h4><p>Finally, I built the RL agent with Keras. I imported <strong><em>SequentialMemory</em></strong>, which allows the agent to retain memory from previous games. I also imported <strong><em>LinearAnnealedPolicy</em></strong>, which added decay as we got closer to the optimal strategy, and<strong><em> EpsGreedyPolicy</em></strong>, which allowed the agent to find the best reward outcome.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DRuoKNlULK2gJR0i2TZy7A.png" /></figure><p>I defined the agent and passed through the model and actions. Then I defined the policy by passing the epsilon greedy policy through the linear anneal policy. I defined the memory using sequential memory with a buffer limit of 1000 episodes with a window length of (meaning for 1000 episodes, the model stores the past 3 windows to capture what our previous steps look like).</p><p>Then I defined the agent by passing through the model, memory, and policy. I also added a dueling network, which splits the value and advantage and helps the model learn when to take actions and when not to take actions. I also added which actions to take and the number of steps the model should take.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/880/1*MBeUqBInvlGBvXMP0LIehg.png" /></figure><h4>Training and Testing</h4><p>To start training the model, I assigned the <strong><em>build_agent</em></strong> function to <strong><em>dqn</em></strong>. Then I added an optimizer for the neural network and a learning rate of .0001. I fitted the model and passed through how long I want to train the model (10000 steps), and specified that I don’t want to visualize the training to speed up the process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E5OeXBjmu4OmakX8A5d6HQ.png" /></figure><p>Finally, I tested the model for 10 steps and output the reward from each episode. At this point in its training, it was playing decently and started to develop a strategy by targeting the mothership (the purple ship at the top). Essentially, it was learning to play the game.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/889/1*N2Zt0iq9axmwKNl3qWp3FQ.png" /></figure><p>Ideally, the model should actually be trained for 10M-40M steps according to Google DeepMind, but I didn’t exactly have the time for that (considering that training for only 10k steps took 14675 seconds, or 4 hours). But if the model was trained for that many steps, it would reach the optimal playing state.</p><p>Here is a <a href="https://youtu.be/9ZGMPWzNleg">video</a> of me explaining the model and watching the agent play.</p><p><em>Have feedback or questions? Send me an email at chloewang.lv@gmail.com and I’ll be happy to respond!</em></p><p><em>You can check out more of what I’m up to in my quarterly newsletters. Sign up </em><a href="https://docs.google.com/forms/d/e/1FAIpQLSeAEWbVJplZ_njSIDfs9CWegCagrhYfq2o6789DDh3hD8cwpQ/viewform?usp=sf_link"><em>here</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8d5159aa69ed" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building An Ion Thruster: Model I]]></title>
            <link>https://chloeewang.medium.com/building-an-ion-thruster-model-i-54603e934850?source=rss-25cb921bfe24------2</link>
            <guid isPermaLink="false">https://medium.com/p/54603e934850</guid>
            <category><![CDATA[space-travel]]></category>
            <category><![CDATA[space]]></category>
            <category><![CDATA[ionic]]></category>
            <dc:creator><![CDATA[Chloe Wang]]></dc:creator>
            <pubDate>Wed, 28 Jul 2021 22:57:44 GMT</pubDate>
            <atom:updated>2021-07-28T22:58:37.012Z</atom:updated>
            <content:encoded><![CDATA[<p>One of the many technologies being considered for future space flight programs are <strong>ion thrusters</strong>, which come from a subset of propulsion known as electric propulsion.</p><p>They’re particularly popular for their <strong>high efficiency and long accelerating times</strong>. Unlike chemical propulsion systems, which only have enough fuel to launch the rocket out of orbit, ion thrusters have fuel that can last many years. By inertia, constantly excreting this fuel allows them to accelerate bodies to extremely high speeds over time.</p><p>In short, they <strong>operate by ionizing propellant </strong>(usually xenon)<strong> and producing plasma, which can be manipulated by electric and magnetic fields to form thrust</strong>. If you’d like an in-depth explanation of ion thruster operation, you can read “<a href="https://chloeewang.medium.com/hall-thrusters-an-efficient-alternative-to-chemical-propulsion-bf19d44ee2fa">2.1 Basic Ion Thruster Operation<em>” </em>from my review paper on Hall thrusters.</a></p><h3>Building An Ion Thruster</h3><p>I decided that I wanted to build an ion thruster — or at least a demonstration of the concept. It’s definitely not rocketship-worthy, but it uses similar mechanisms to show how ionic thrust/wind can be produced.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*pC_i4jIsEZsfE0WjOnXEAg.jpeg" /></figure><p>For my supplies I used:</p><ul><li>Battery container from a flashlight (mine held 3 AAA batteries)</li><li>Boost step-up power module (DC6–12V to 1000kV)</li><li>3 metal screws (~6 cm in length is ideal)</li><li>3 aluminum cans (mine were Brisk Tea cans)</li><li>Copper wire</li><li>Crocodile clip</li><li>Electrical tape (or heat shrinks)</li><li>Binder clip</li><li>Receipt (or another type of thin paper)</li><li>3D Printer (I used the Ender 3 V2)+ filament</li><li>Hot glue</li><li>Double-sided tape</li></ul><h4>3D Printing The Parts</h4><p>I first started off by designing the components that would hold the thruster in <strong>TinkerCAD</strong>, an easy-to-use software for designing 3D prints.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*U9G-yYO8iMauG8_HPWHQLQ.png" /></figure><ul><li>Platform A holds the stand and the can holder.</li><li>Platform B holds the screw holder and has a lid. This platform is hot glued to platform A when it’s printed.</li><li>The stand is made to have a receipt attached to it via binder clip.</li><li>The screw holder fits into a flat component that can be moved on platform B. This allows me to adjust the distance between the screws and the cans.</li><li>The platform lid goes onto the right platform after the screw holder is placed in.</li><li>The can holder holds 3 cans.</li></ul><blockquote><strong>Pro tip:</strong> When designing components made to fit into each other, make sure that the part that goes inside the hole is slightly smaller than the hole. I made the mistake of making them the same size and had to reprint some components a few times.</blockquote><p>I printed these components over the course of a few days (they generally took between 5–13 hours to print) using <strong>white PLA filament</strong>. I used Ultimaker Cura as the slicer, with<strong> regular settings at 10% infill</strong>. If you want to print the components faster, I printed the lid and can holder at 150 mm/s instead of 50 mm/s, and they still turned out fine.</p><h4>Assembly</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/384/1*qZ1kQN5jWFu3I5eoDrjUCw.png" /></figure><p>The battery container served as the <strong>power source</strong> and was (manually) connected to the power module through the green and red wires to achieve greater voltage (Approxiamately 4.5 volts to 1000 kV). I secured the power module and battery container on platform B using double-sided tape.</p><p><strong>This is the same power module that’s used in tasers, so you have to be careful while using it.</strong> I made sure that whenever I was moving components that the battery container was far from the green and red wires that connected it to the power module. I also made sure to not touch conductive components (wiring, nails, metal, etc.) while it was operating.</p><p>One wire on the opposite end of the power module was connected to a copper wire using electrical tape. That copper wire was then wrapped around the three screws in the screw holder. Wrapping the copper wire around the three screws allowed all of them to receive voltage.</p><blockquote><strong>Pro tip:</strong> I would recommend copper-plating the screws to make them more conductive. You can do this by putting a lemon’s worth of lemon juice + 2 pinches of salt + 20 pennies in a bowl with your screws. After waiting a few hours, your screws will be darker/copper-colored. (I ran out of lemons, so I couldn’t copper-plate my screws)</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/384/1*hhg8TvtuZ72iEWOdH5cIUw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/384/1*h9AXSQORhB6xWRpNrvngWg.jpeg" /></figure><p>The cans were placed on the can holder in a stacked position using no adhesives (they balanced on each other). Then I connected the other wire from the power module to one of the cans using a crocodile clip and some electrical tape.</p><p>The last part is putting in the stand. Then I attached a receipt paper to it using a binder clip. The movement of this paper will signify the production of ionic wind.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/384/1*9dx2k4w4qPIp9wnZhS21XQ.jpeg" /></figure><p>Fully assembled, the ion thruster should look something like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*pC_i4jIsEZsfE0WjOnXEAg.jpeg" /></figure><h3>How Does It Work?</h3><p>This model demonstrates how ionic wind can be produced using the basic mechanisms used in ion thrusters.</p><p>Providing voltage to the <strong>screws</strong> causes them to act as positively charged <strong>anodes</strong>, while providing voltage to the<strong> cans</strong> causes them to act as negatively charged <strong>cathodes</strong>. This generates an <strong>electric field.</strong></p><p>The screws make good anodes because they’re sharp, and the cans make good cathodes because they’re round. The point of the anode allows the charges to become <strong>packed together</strong>, making the electric field stronger around that location. In fact, it becomes strong enough to <strong>rip negatively charged electrons from the surrounding air, turning them into positively charged ions.</strong></p><p>These positively charged ions are now repelled from the positively charged anodes (screws) and attracted to the negatively charged cathodes (cans). As they’re moving towards the cathodes,<strong> they also bump into and knock electrons out from other air molecules, creating more ions</strong> that move towards the cathodes. These ions move through the cathodes as ionic wind, which you can see pushing the receipt when the ion thruster is operating.</p><h3>Operation</h3><p>Here are some videos of the ion thruster in operation. You can see that I manually connect the power module wires to the battery container, although you could design a switch if you wanted to. In the third video you can see the plasma at the tips of the screws as air molecules are being ionized.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FqNbVY1bH4hY%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DqNbVY1bH4hY&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FqNbVY1bH4hY%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/cbfad46669091aeccd12601dc445ff86/href">https://medium.com/media/cbfad46669091aeccd12601dc445ff86/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FzYVo_XjFou4&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DzYVo_XjFou4&amp;image=http%3A%2F%2Fi.ytimg.com%2Fvi%2FzYVo_XjFou4%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/ab555052d109c69699a889f9f06f4ce9/href">https://medium.com/media/ab555052d109c69699a889f9f06f4ce9/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FXEnqhdMDffo%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DXEnqhdMDffo&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/39918b034a1c1b03f645c417f7fc2ace/href">https://medium.com/media/39918b034a1c1b03f645c417f7fc2ace/href</a></iframe><h3>Potential Improvements</h3><p>This model is not perfect, and I definitely think that there are a few things that could be improved, such as:</p><ul><li>The screws could have been copper-plated and centered better. Centering the screws better would require some alterations to the original design of the screw holder.</li><li>Certain parts of the 3D printed components could have been made smaller to save printing time and filament.</li><li>A switch could have been added to the power part of this model so that the wires from the power module did not need to be manually attached to the battery container. It would have also made using this model safer.</li></ul><p>Overall, however, I’m happy with the product, since it successfully produced ionic wind and even visible plasma.</p><p><em>Have feedback or questions? Send me an email at chloewang.lv@gmail.com and I’ll be happy to respond!</em></p><p><em>You can check out more of what I’m up to in my monthly newsletters. Sign up </em><a href="https://docs.google.com/forms/d/e/1FAIpQLSeAEWbVJplZ_njSIDfs9CWegCagrhYfq2o6789DDh3hD8cwpQ/viewform?usp=sf_link"><em>here</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=54603e934850" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Hall Thrusters: An Efficient Alternative To Chemical Propulsion]]></title>
            <link>https://chloeewang.medium.com/hall-thrusters-an-efficient-alternative-to-chemical-propulsion-bf19d44ee2fa?source=rss-25cb921bfe24------2</link>
            <guid isPermaLink="false">https://medium.com/p/bf19d44ee2fa</guid>
            <category><![CDATA[space-travel]]></category>
            <category><![CDATA[space-exploration]]></category>
            <category><![CDATA[space-technology]]></category>
            <dc:creator><![CDATA[Chloe Wang]]></dc:creator>
            <pubDate>Fri, 02 Apr 2021 19:14:54 GMT</pubDate>
            <atom:updated>2021-04-03T04:11:09.499Z</atom:updated>
            <content:encoded><![CDATA[<h3>Abstract</h3><p>Chemical rocket propulsion is not an efficient method of long-term space travel due to its requirements for large amounts of fuel for a small payload. Hall thruster systems are a subset of electric propulsion systems that may serve as an efficient alternative to chemical propulsion systems. Hall thrusters operate by bombarding neutral atoms with electrons to form ions and ejecting those ions at high velocities. They can operate for extended periods of time, allowing their rocket bodies to accelerate to speeds much higher than that of a chemical rocket. This article reviews critical Hall thruster parameters, basic Hall thruster operation, comparisons to traditional ion thrusters, types of Hall thrusters, drawbacks, and current missions.</p><p><strong>Keywords: </strong>Hall thruster, electric propulsion</p><h3>Outline</h3><p><strong>1. Preliminaries</strong></p><p>1.1. Lorentz Force and ExB</p><p>1.2. Hall effect</p><p>1.3. Critical Thruster Parameters</p><p><strong>2. Ion Thrusters</strong></p><p>2.1. Basic Ion Thruster Operation</p><p><strong>3. Hall Thrusters</strong></p><p>3.1. Basic Hall Thruster Operation</p><p>3.2. Hall Thrusters Vs Ion Thrusters</p><p>3.3. Propellants</p><p>3.3.1. Xenon Propellant</p><p>3.3.2. Iodine Propellant</p><p><strong>4. Types of Hall Thrusters</strong></p><p>4.1. Dielectric Wall Thrusters</p><p>4.2. Magnetic Layer Thrusters</p><p>4.3. Anode Layer Thrusters</p><p><strong>5. Drawbacks</strong></p><p><strong>6. Current Missions</strong></p><p>6.1. Raijin 94 Hall Thruster</p><p>6.2. THT VI Hall Thruster</p><p>6.3. AEPS 13.3 kW Hall Thruster</p><p>6.4. XR-100 Electric Propulsion System and X3 Hall Thruster</p><h3>Introduction</h3><p>Current traditional means of entering and traveling through space are typically done through chemical thrust, propulsion, and inertia. The standard chemical rocket includes a propulsion system containing a rocket engine, fuel, and oxidizer that may take up as much as 80% of the main rocket body. These rockets produce large amounts of thrust by causing fuel and oxidizer to react with each other in a combustion chamber, ejecting the product at high velocities. The payload is typically carried at the top of the rocket and makes up a very small percentage of the rocket body and weight.</p><p>To increase the payload taken to space, the amount of fuel needed to launch the payload must also increase. The increased amount of fuel adds more weight to the rocket body, which needs to be accommodated by additional fuel. This cycle is described by the Tsiolkovsky rocket equation.</p><p>This fuel only accounts for escaping the Earth’s gravity. Many rockets expel all their fuel once they escape the Earth’s gravity, and travel to their destination by inertia afterward. As a result, they travel very slow and often need to rely on gravitational pushes from other large bodies in space to continue forward.</p><p>Hence, the use of traditional chemical rockets is not practical for long-term space travel and exploration. An efficient alternative to chemical propulsion systems are electric propulsion systems. Hall thrusters are a subset of electric propulsion systems that present themselves as another efficient system for traveling through space.</p><h3>1. Preliminaries</h3><p>Hall thrusters are a subset of electrical propulsion systems that use the Hall effect, named after Edwin Hall, to operate. Preliminaries to understanding the Hall effect are required to understand the operation of Hall thrusters.</p><h4>1.1. Lorentz Force and ExB Drift</h4><p>The Lorentz force describes an event that occurs when a force is exerted on a charged particle moving through a magnetic field, such as an electron. When a magnetic field is directed perpendicular to the direction of an electron’s motion, a force is exerted on the electron that is perpendicular to both the magnetic field and the movement of the electron.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*3ljBghnJA9F2UM-W" /><figcaption>Lorentz force diagram, by <a href="https://www.allaboutcircuits.com/technical-articles/understanding-and-applying-the-hall-effect/">Robert Keim from All About Circuits</a></figcaption></figure><p>The direction of the electron is determined by the right-hand rule. The resulting force directs motion in a curve such that if we view along the direction of the magnetic field, positive particles will gyrate counter-clockwise while negative particles will gyrate clockwise. Negatively charged electrons will gyrate clockwise.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*siozMDpiTMCC6zvm.jpg" /><figcaption>ExB drift visual, by <a href="https://svs.gsfc.nasa.gov/4265">NASA Scientific Visualization Studio</a></figcaption></figure><p>When an electrical field is added, a counter-intuitive effect is created. One could expect that positive particles drift in the direction of the electrical field while negative particles drift opposite the electrical field direction. However, both particles drift in a direction perpendicular to the electrical and magnetic field directions. This is known as ExB drift, where E<strong> </strong>is the electrical field and B is the magnetic field.</p><h4>1.2. Hall Effect</h4><p>The Hall effect refers to when the Lorentz force acts upon an electron moving through a conductor, causing differences in electric potential to develop along both sides of the conductor.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JsfRWpiH1RoUZ7fl" /><figcaption>Hall effect diagram, by <a href="https://www.allaboutcircuits.com/technical-articles/understanding-and-applying-the-hall-effect/">Robert Keim from All About Circuits</a></figcaption></figure><p>In a conventional direct current, electrons travel in the direction opposite of the flow. In the diagram above, the Lorentz force is acting downward due to the direction of the electron flow, causing electrons to accumulate near the bottom of the conductor and therefore causing a difference in electric potential. The accumulation of electrons is relevant to the limited electron motion that is required during Hall thruster operation.</p><h4>1.3 Critical Thruster Parameters</h4><p>Specific impulse is defined as the thrust produced per unit rate of consumption of propellant and is a measure of the efficiency of a rocket engine. A greater specific impulse means greater engine efficiency.</p><p>In standard chemical rocket engines, one may assume that the specific impulse is governed solely by the rocket equation. However, this is not the case.</p><p>A traditional spacecraft applies a force through the expulsion of propellant at a high velocity, which can be modeled in a one-dimensional situation absent of gravitational forces.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/377/1*diSvC_wAI2eaernL8KjqeA.png" /><figcaption>By <a href="https://pepl.engin.umich.edu/pdf/2004_Hofer_thesis.pdf">Richard R. Hofer at NASA</a></figcaption></figure><p>Where <em>M(t)</em> is the function of time with mass, and <em>U(t)</em> is the function of velocity with mass. By the conservation of momentum, the acceleration of the spacecraft results from the application of <em>T</em>, thrust, given by the product of the mass flow rate <em>ṁ</em> and effective propellant exit velocity <em>c</em>. The presence of <em>c</em> accounts for unbalanced pressure forces and the nonuniform distribution of exhaust velocities over the exit cross-section of the rocket.</p><p>As propellant is released, the mass of the spacecraft is reduced over time. Assuming the mass flow rate of the propellant is constant, the spacecraft mass can be modeled by the following function, where Mₒ is the initial mass.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/332/1*-l16qhFcBzXflbe-JYDDoA.png" /><figcaption>By <a href="https://pepl.engin.umich.edu/pdf/2004_Hofer_thesis.pdf">Richard R. Hofer at NASA</a></figcaption></figure><p>By differentiating the function and substituting it in the previous one-dimensional model, we find that</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/331/1*8UxNa9luj0ox06WMmiReSA.png" /><figcaption>By <a href="https://pepl.engin.umich.edu/pdf/2004_Hofer_thesis.pdf">Richard R. Hofer at NASA</a></figcaption></figure><p>By further canceling the time differentials and integrating over an initial to final spacecraft velocity and mass, we can obtain a ratio of the final to initial vehicle mass as a function of the velocity increment <strong><em>Δ</em></strong><em>U</em> and effective exit velocity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/237/1*-Qe6bUHSPwFh5RddGRwmeg.png" /><figcaption>By <a href="https://pepl.engin.umich.edu/pdf/2004_Hofer_thesis.pdf">Richard R. Hofer at NASA</a></figcaption></figure><p>This derivation of the Tsiolkovsky rocket equation illustrates the impact of exhaust velocity on the mass ratio of a spacecraft. This implies that the exhaust velocity should be on the order of the velocity increment to deliver a useful mass fraction at its destination. This explains why exhaust velocity is often used as a first-order standard in space vehicle design. However, there are competing criteria that diminish the importance of exhaust velocity, such that optimizing technology with the greatest exhaust velocity isn’t always the most efficient.</p><p>Total impulse is used to report rocket lifetime. For example, the lifetime of a rocket with a total impulse capability of 3*10⁶ N-s and a constant thrust of 83 mN would be about 10,000 hours.</p><p>The following function integrates total impulse delivered to a spacecraft over time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/211/1*tMfAUeJ2sBs1K54164tI2g.png" /><figcaption>By <a href="https://pepl.engin.umich.edu/pdf/2004_Hofer_thesis.pdf">Richard R. Hofer at NASA</a></figcaption></figure><p>Specific impulse can also be thought of as the total impulse per unit weight of a given propellant, as given in the following function.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/315/1*T0DJmcL09A8LKfyocVCyRA.png" /><figcaption>By <a href="https://pepl.engin.umich.edu/pdf/2004_Hofer_thesis.pdf">Richard R. Hofer at NASA</a></figcaption></figure><p>Specific impulse is reduced to the following is the thrust and mass flow rate of the vehicle are constant over time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/490/1*O4yrVhrVGQmP1ob7rFN3Fw.png" /><figcaption>By <a href="https://pepl.engin.umich.edu/pdf/2004_Hofer_thesis.pdf">Richard R. Hofer at NASA</a></figcaption></figure><p>This makes the specific impulse is ~1/10 of the effective exit velocity, meaning that specific impulse and effective exit velocity are essentially interchangeable. However, don’t rely on this, since the constant can be arbitrary and lead to overestimation of the specific impulse. Yet, it provides the argument that specific impulse is another major factor that should be considered when designing propulsion systems.</p><p>Thus, Hall thruster systems rely immensely on optimizing specific impulses, efficiency, and thrust.</p><h3>2. Ion Thrusters</h3><p>Electric propulsion is an alternative to chemical propulsion, and there are many subsets to it. To understand why Hall thrusters are optimal, one should familiarize themself with its competitor, the standard ion thruster.</p><h4>2.1 Basic Ion Thruster Operation</h4><p>An ion thruster operates by ionizing a propellant through the addition or removal of electrons.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/850/0*MU6XLk2VVzvAaoBp.png" /><figcaption>Ion thruster diagram, by <a href="https://www.researchgate.net/figure/Diagram-of-an-Ion-engine_fig12_259902732">Taher Muhhammad</a></figcaption></figure><p>Standard ion thrusters ionize the propellant through electron bombardment. In this process, a high-energy electron collides with a neutral propellant atom, releasing electrons from the propellant atom and forming a positively charged ion. On a large scale, plasma is produced, since the mix of positively charged ions and negatively charged electrons forms a neutrally charged gas. The plasma can be manipulated by electric and magnetic fields.</p><p>The electrons needed to activate this process are distributed by a discharge cathode and flow into the discharge chamber. A neutral propellant is injected into the discharge chamber with the electrons to allow bombardment to occur.</p><p>Once ions are produced, they begin to migrate toward grids containing thousands of apertures at the aft end of the ion thruster. The first grid is a positively charged electrode known as a screen grid. The high positive voltage applied to the screen grid forces plasma to reside at a high voltage.</p><p>The second grid is a negatively charged electrode known as an accelerator grid. Positive ions are attracted to this negative grid, accelerating them to speeds up to 90,000 mph.</p><p>The ions accelerate out as an ion beam, producing thrust. To prevent ions from being attracted back to the main spacecraft, a hollow cathode neutralizer expels an equal amount of electrons to make the total charge of the exhaust beam neutral.</p><p>The efficiency of ion thrusters allows fuel to be used over a long period of time. The ion beam constantly accelerates the rocket body and allows it to travel at much higher velocities than chemical rockets.</p><h3>3. Hall Thrusters</h3><p>Hall thrusters are an efficient alternative system to chemical propulsion. The following covers basic operation, thruster comparisons, and propellants.</p><h4>3.1. Basic Hall Thruster Operation</h4><p>Similarly to ion thrusters, Hall thrusters achieve propulsive thrust through electron bombardment and an ionized inert gas beam. However, the methods and efficiency of doing so differ from that of an ion thruster, since the ionized propellant gas in Hall thrusters is accelerated by electric and magnetic fields.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/592/0*LrEuJDEoe-wO_NN4" /><figcaption>Hall thruster diagram, by <a href="https://core.ac.uk/reader/322446284">Sukhmander Singh</a></figcaption></figure><p>Propellant enters the discharge chamber through dispersion by a positively charged anode. The discharge chamber is typically a cylindrical shape and made of a metallic material, such as boron nitride.</p><p>A magnetic field on the order of ~150 Gs is applied to produce a closed drift of electrons inside the channel. Additionally, a strong electric field of ~1000 V/m is generated by the magnetic circuit inside the discharge channel along the axial direction of the device. The applied magnetic field is strong enough to magnetize the electrons, causing them to gyrate in the discharge chamber. The magnetic field does not affect propellant ions due to their large size.</p><p>Cross field configurations cause the electrons to effectively remain trapped in ExB drifts around the channel and slowly drift back towards the anode through the Hall effect. Because the electrons are trapped in a constantly gyrating motion and reside in the channel longer, neutral propellant atoms are more likely to be bombarded and ionized. The closed-drift region is the portion of the discharge chamber where the electron ExB drift is the greatest.</p><p>The establishment of strict axial electron mobility also creates a self-consistent electric field. The electric field must sharply rise in the region of maximum magnetic field intensity in order to maintain current continuity.</p><p>Ions remain unaffected by the magnetic field due to their greater mass, but are still accelerated by the electric field to produce thrust by an ion beam. Like the ion thruster, the plasma in closed-drift regions is neutral due to the mix of positively charged ions and negatively charged electrons.</p><p>As in ion thrusters, the efficiency of Hall thrusters allows fuel to be used over a long period of time. The ion beam constantly accelerates the rocket body and allows it to travel at much higher velocities than chemical rockets.</p><h4>3.2. Hall Thrusters Vs Ion Thrusters</h4><p>What makes Hall and ion thrusters distinctly different from chemical thrusters is that the energy that propels the rocket comes from an external power source instead of the molecular bonds of a propellant.</p><p>In chemical rockets, this dependence on molecular bonds for energy limits the specific impulse to about 450 s, whereas in electric thrusters, specific impulses of over 17,000 s have been achieved in labs.</p><p>Longer lifetimes are desirable as they allow the ion beams to accelerate the rocket bodies for longer periods of time.</p><p>Generally, ion thrusters have low thrust densities (~0.1 mN/cm²) because electrons between grids are excluded, limiting the beam current. Traditional ion thrusters have specific impulses of ~3000 s, efficiencies of ~60%, and lifetimes of about 10,000 h.</p><p>On the other hand, Hall thrusters accelerate ions in a quasineutral plasma such that the thrust density is only limited by material and thermal limitations. They have much higher thrust densities that are generally on the order of 1.0 mN/cm². They also have higher specific power because they require fewer power supplies and have less stringent voltage isolation requirements. Traditional Hall thrusters have specific impulses of ~1600 s, efficiencies of ~50%, and lifetimes of ~7000 h.</p><p>If a Hall thruster and ion thruster were of equal power, the ion thruster would:</p><ul><li>Be larger due to a low specific power and thrust density</li><li>Produce less thrust because a higher specific impulse means a lower thrust-to-power ratio</li></ul><p>But will have a higher specific impulse, efficiency, and lifetime.</p><p>On the contrary, the Hall thruster would:</p><ul><li>Be smaller due to a higher specific power and thrust density</li><li>Produce more thrust because a lower specific impulse means a higher thrust-to-power ratio</li></ul><p>But will have a lower specific impulse, efficiency, and lifetime.</p><p>For the Hall thruster to outweigh its counterpart in benefits, it must achieve a specific impulse between 2000–3000 s. Once this specific impulse is achieved, then the benefits may include greater payloads, launch vehicle step-downs, reduced trip times, and lower required power, making it the ideal electric thruster.</p><h4>3.3. Propellants</h4><p>A desirable propellant must have a high atomic mass and gaseous electronic properties that allow for efficient ionization (low ionization potential).</p><p>In general, molecular propellants are not considered because of efficiency losses due to vibrational and rotational degrees of freedom. Additionally, they may accommodate a significant amount of internal energy that results in frozen flow power losses. Substances stored in molecular form may also have low dissociation energy and are readily and efficiently converted into atomic form. Some molecular species are also subject to rapid dissociative electron recombination reactions, which serve as a major sink for low-energy electrons.</p><p>Atomic gases are preferred over molecular species for these reasons. However, some substances stored in molecular form may also have low dissociation energy and are readily and efficiently converted into atomic form.</p><p><strong>3.3.1. Xenon Propellant</strong></p><p>Xenon is the 54th element on the periodic table and is the most commonly used propellant in Hall thruster systems. Xenon is located in the 18th column, indicating its non-reactive properties as a noble gas. It is the most preferred propellant because it is easily ionized and has a high atomic mass. This high atomic mass allows the xenon atoms to generate a desirable level of thrust when ions are accelerated. Xenon has a high storage density, which is suitable for spacecraft storage. This is highly preferred over the storage of cryogenic propellants, such as hydrogen. Xenon also has excellent discharge properties.</p><p>The main disadvantage of using xenon propellant is that it’s extremely expensive due to its small abundance in the Earth’s crust. 1 kg of solid xenon (99.995%) costs about $4000 USD, which will not be sustainable if Hall thrusters become more widely used in the future.</p><p><strong>3.3.2 Iodine Propellant</strong></p><p>Iodine is the most favorable alternative to xenon, although it is a molecular propellant. Iodine is the 53rd element on the periodic table, located directly next to xenon. This gives it many similar properties to xenon.</p><p>Since Iodine is a molecular propellant, it would be stored in solid form. In solid form, it has a high density (4.94g cm³) with a room temperature vapor pressure of 0.3 Torr and a boiling point of 558 K. Moderate heating can raise its vapor pressure considerably. The low vapor pressure of iodine in addition to its high density would allow the tank to be substantially smaller and of a significantly lower mass than that of a high-pressure gas cylinder required for gaseous propellants like xenon. However, the system created would be more complex than a xenon propellant system, since a frit will be needed to prevent iodine crystals from migrating into the discharge chamber in zero-gravity environments.</p><p>The significant difference between iodine and xenon is that iodine has a large electron affinity that allows it to form stable negative ions, while xenon does not. Additionally, 1 kg of iodine (99.999%) is $400, which is 10x cheaper than Xenon. This would be far more sustainable in the occurrence of mass Hall Thruster use.</p><p>The primary disadvantage of iodine is that it is substantially more corrosive than xenon. In the long term, this can pose a serious threat to anode surfaces, especially if negative ion currents are important.</p><h3>4. Types of Hall Thrusters</h3><p>There are multiple types of Hall thrusters, including dielectric wall thrusters, magnetic layer thrusters, and anode layer thrusters.</p><h4>4.1. Dielectric Wall Thruster</h4><p>Dielectric wall thrusters, also known as stationary plasma thrusters, are thrusters with closed electron drifts and an extended acceleration zone. The walls are made of dielectric materials, such as boron nitride or silicon carbide. This allows for collisions of electrons and ions within the wall to generate low-energy secondary electrons.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/291/0*b2LNiQ48kd8lHJRi.jpg" /><figcaption>Dielectric wall thruster diagram, from <a href="http://Stationary plasma thruster simulation">Stationary plasma thruster simulation</a></figcaption></figure><p>These secondary electrons help maintain low electron temperatures inside the discharge plasma. By reducing the discharge electron energy, one can achieve smooth and continuous variation in plasma potential between the anode and cathode.</p><p>Nonconductive dielectric walls let charge accumulate along the length of the acceleration channel that leads to a variable potential profile along its length.</p><h4>4.2. Magnetic Layer Thruster</h4><p>Magnetic layer thrusters are single-stage Hall thrusters with a coaxial ceramic discharge chamber. It is the most common Hall thruster type and therefore the common basic design.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/699/1*9XH39SgG4tcXHHXo5bo62A.jpeg" /><figcaption>Magnetic layer thruster diagram, by <a href="https://pepl.engin.umich.edu/pdf/2004_Hofer_thesis.pdf">Richard R. Hofer at NASA</a></figcaption></figure><p>A magnetic layer thruster has an anode, cathode, magnetic circuit, and discharge chamber.</p><p>The anode is found inside the discharge chamber and ejects a neutral propellant gas. The walls are typically made of boron nitride or boron nitride mixed with silicon dioxide (borosil). The anode doubles as a positive electrode, which attracts electrons gyrating in ExB drift.</p><p>The cathode is responsible for supplying electrons to the discharge chamber and plasma plume for neutralization of ion exhaust. They tend to be mounted outside the thruster, but can also be mounted on the thruster centerline if the inner core is hollow.</p><p>An orifice hollow cathode tends to be the cathode of choice. To operate such a cathode, propellant gas is passed over a thermionic emitter, such as lanthanum hexaboride or barium oxide, which when heated emits electrons and initiates a plasma breakdown via electron-neutral collisions. The electrons are extracted through a small orifice with a positively biased electrode known as a keeper. Cathodes are typically capable of heater-less and keeperless operation after the initial plasma breakdown occurs.</p><p>The magnetic circuit supplies the electric field that confines plasma in discharge chambers and acts as a support structure for other thruster components. The basic design is made of a collection of electromagnetic coils used to create a magnetic flux and magnetic pole pieces used to channel the magnetic flux into the discharge chamber.</p><h4>4.3. Anode Layer Thruster</h4><p>Anode layer thrusters are two-stage Hall thrusters where the thruster uses a metallic discharge chamber separated into an ionization and acceleration stage by an intermediate electrode.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/713/1*BoKHhSadj-atI2PuTQdCWw.jpeg" /><figcaption>Anode layer thruster diagram, by <a href="https://pepl.engin.umich.edu/pdf/2004_Hofer_thesis.pdf">Richard R. Hofer at NASA</a></figcaption></figure><p>Ionization potential is known as the amount of energy required to remove an electron from an isolated atom or molecule. Applied potential is known as the difference of potential measured between two identical metallic leads to two electrodes of an electrochemical cell.</p><p>In the ionization stage of an anode layer thruster, the applied potential is 5–10x that of the ionization potential of the propellant. The applied potential in the acceleration stage, also known as acceleration-stage voltage, accelerates ions created in the ionization stage.</p><p>In comparison to the magnetic layer thruster, the components are primarily the same besides the intermediate electrode found in the anode layer thruster.</p><p>The main purpose of anode layer thrusters are to improve ionization efficiency, hence why it tends to be used in larger spacecraft. However, this comes at the expense of complexity due to an additional electrode. The benefits felt by the anode layer thruster only begin to outweigh that of the magnetic layer thruster when the acceleration potential is on the order of several kilovolts. This is because sub-kilovolt magnetic layer thrusters are capable of efficiently ionizing propellants without the hindrance of an ionization stage.</p><p>Anode layer thruster also have higher electron temperatures than magnetic layer thrusters. This means that in order to maintain thruster efficiency, the two-stage approach in an anode layer thruster becomes necessary at lower voltages.</p><h3>5. Drawbacks</h3><p>Although Hall thrusters have a number of benefits over other types of propulsion systems, there are still drawbacks that are still being studied. The following are two major drawbacks.</p><p>The first drawback is that the divergence angle of these devices is ~60°. This angle is relatively large and causes problems related to the erosion of channel walls and outer surfaces of the Hall thruster. It’s ideal to avoid the erosion of walls, since it decreases the lifetime of a device.</p><p>Another major drawback is that in many Hall thrusters, the plasma does not stay uniform. When an inhomogeneous plasma is immersed in the external electric and magnetic fields, it’s not in a thermodynamical equilibrium state, which leads to plasma instabilities. The amplitudes of the waves and instabilities are attributed by the density scale lengths of plasma, the magnetic field, and other parameters. These waves and instabilities may negatively affect the efficiency of the device.</p><h3>6. Current Missions</h3><p>Today, a number of space agencies are developing Hall thrusters for different missions, with the primary leaders being the National Aeronautics and Space Administration (NASA) and the Japan Aerospace Exploration Agency (JAXA). Many agencies share a common goal of developing a powerful electrical propulsion system for planetary exploration and all-electric satellites. The following reviews a few of the missions in development, including the Raijin 94 Hall Thruster, THT VI Hall Thruster, AEPS 13.3 kW Hall Thruster, and X3 Hall Thruster from the XR-100 Electric Propulsion System.</p><h4>6.1. Raijin 94 Hall Thruster</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/625/1*ifHACIEWcEB95HSp2vIlWA.png" /><figcaption>Raijin 94 Hall Thruster visuals, from <a href="https://www.jstage.jst.go.jp/article/tjsass/60/5/60_T-16-49/_article">Hall Thruster Development for Japanese Space Propulsion Programs</a></figcaption></figure><p>The Raijin 94 Hall Thruster is one of the thrusters currently being developed by JAXA. It is a 5 kW anode layer Hall thruster, with an inner acceleration channel that is 60 mm in diameter and an outer acceleration channel that is 94 mm in diameter.</p><p>A solenoid is a cylindrical coil of wire that acts as a magnet when carrying an electric current. The Raijin 94 contains one inner solenoid and four outer solenoids. This creates a predominantly radial magnetic field in the acceleration channel.</p><p>There is also a hollow annular cathode, as found in many Hall thrusters. It’s made of two cylindrical rings that propellant gas, xenon, is fed through. The gap between the tip of the anode and the exit of the acceleration channel is fixed at 3 mm.</p><p>The Raijin 94 is currently in testing stages and is set to be in use in all-electrical satellites.</p><h4>6.2. THT VI Hall Thruster</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/721/0*glaRgPAFGjtdWBdG" /><figcaption>THT VI Hall Thruster visuals, from <a href="https://www.jstage.jst.go.jp/article/tjsass/60/5/60_T-16-49/_article">Hall Thruster Development for Japanese Space Propulsion Programs</a></figcaption></figure><p>The THT VI Hall Thruster is a magnetic layer hall thruster being developed by JAXA. The inner diameter of the discharge channel is 56 mm while the outer diameter is 100 mm. The channel is 22 mm wide and 40 mm long, which is the same as another hall thruster being developed by Russia (SPT-100). The walls are made of boron nitride.</p><p>As in most hall thrusters, a hollow cathode serves as the electron source, while an anode expells xenon propellant. The thruster mass flow rate is fixed at 3.0 mg/s. The hollow cathode mass flow rate is 0.1 mg/s. In high discharge voltage operation, the mass flow rate is 0.2 mg/s, when discharge voltage is between 950–1000 V for stable operation.</p><p>JAXA has plans to also use this thruster in all-electrical satellites.</p><h4>6.3. AEPS 13.3 kW Hall Thruster</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/561/1*DEjwHIYF-6XEhvXJPWmtug.png" /><figcaption>AEPS Hall Thruster visuals, from <a href="http://electricrocket.org/2019/692.pdf">13kW Advanced Electric Propulsion Flight System Development and Qualification</a></figcaption></figure><p>The AEPS 13.3 kW Hall Thruster is a thruster being developed by NASA, and will serve as a building block for a 40 kW class solar electric propulsion vehicle. The architecture is based on an evolutionary human exploration strategy that focuses on flight testing and validation of exploration capabilities in cislunar space prior to conducting missions to Mars. As such, the 40 kW class Hall thruster propulsion system along with a flexible blanket array technology has already presented scalable technology with a clear path to much higher power systems.</p><p>The AEPS thruster is designed to deliver a maximum specific impulse of 2600 s with a propellant throughput capability of 1700 kg. Each power processing unit is designed to operate over input voltages ranging from 95–140 V.</p><h4>6.4. XR-100 Electric Propulsion System and X3 Hall Thruster</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/875/0*tLTPriGBIi-Me-L8" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/595/1*t6kZPUi-xxavCVboDLpZFQ.png" /><figcaption>X3 Hall Thruster visuals, by NASA</figcaption></figure><p>The X3 Hall Thruster is a 200 kW nested hall thruster that is a part of the XR-100 Electric Propulsion System. The primary goal of this system is to serve as a high-power electric propulsion system with the potential to participate in large-scale cargo transportation missions to support human missions to the Moon and Mars. This is the leading project in the field of Hall thrusters and is a top priority technology at NASA.</p><p>The thruster is about 80 cm in diameter and weighs about 230 kg. It has the largest throttling capability of any Hall thruster to date, with 7 firing configurations and power levels ranging from 2 to 200 kW. It’s the primary driver to reaching the 600–700 kW mark, which is needed for electric propulsion to have the same thrust as chemical propulsion.</p><p>The X3 Hall Thruster is a nested channel Hall thruster (NHT), which can increase to power levels above 100 kW while maintaining an acceptable device size and mass. It does this by adding channels that circumscribe the centrally mounted cathode. NHTs also enable higher total power and power density compared to single channel Hall thrusters. Using NHTs also allows heat to be passively radiated during operation and requires no active or conductive cooling, simplifying temperature control systems.</p><p>This thruster is such a valuable project because in critical missions, such as cargo missions to Mars, the spacecraft based on electric propulsion systems mass can be reduced by up to 80% compared to spacecraft based on chemical propulsion systems. This is because the thruster only requires propellant and enough heat during periods of eclipse to keep the components above their qualified temperatures.</p><p>Currently, NASA is working on development efforts for components experiencing challenges with increased power and propellant flow rates.</p><h3>Conclusion</h3><p>Chemical rockets are not an efficient method of space travel due to their need for large amounts of fuel and therefore increased body weight for small payloads. Hall thrusters are an efficient alternative system to chemical propulsion systems that use ions to continuously accelerate through space.</p><p>Hall thrusters use ExB drift and the Hall effect to aid in accumulating electrons. This helps increase electron-propellant bombardment to produce more ions and therefore increase thrust and efficiency.</p><p>Although Hall thrusters tend to have lower specific impulses, efficiencies, and lifetimes compared to their primary competitor, the standard ion thruster, they are smaller due to their higher specific power and thrust density. Additionally, their lower specific impulses allow for a higher thrust-to-power ratio.</p><p>The most commonly used propellant is xenon due to its high atomic mass and non-reactive properties. However, iodine is another element being considered for use in Hall thrusters.</p><p>There are three main types of Hall thrusters — dielectric wall thrusters, magnetic layer thrusters, and anode layer thrusters. Each have their own unique architecture that are advantageous in different contexts.</p><p>The primary drawbacks to Hall thruster systems are that they have large divergence angles that lead to erosion and inhomogenous plasma.</p><p>The main leaders in Hall thruster development are NASA and JAXA. JAXA is developing the Raijin 94 Hall Thruster and THT VI Hall Thruster, while NASA is developing the AEPS 13.3 kW Hall Thruster and X3 Hall Thruster.</p><p>Hall thruster applications are likely to take place (but not limited to) in deep space missions, including missions to the Moon and Mars.</p><h3>Sources</h3><p><a href="https://core.ac.uk/reader/322446284">Hall Thruster: An ElectricPropulsion through Plasmas</a></p><p><a href="https://www.jstage.jst.go.jp/article/tjsass/60/5/60_T-16-49/_article">Hall Thruster Development for Japanese Space Propulsion Programs</a></p><p><a href="https://trs.jpl.nasa.gov/bitstream/handle/2014/48291/CL%2318-2282.pdf?sequence=1&amp;isAllowed=y">Electric Propulsion Research and Development at NASA</a></p><p><a href="https://pepl.engin.umich.edu/pdf/2004_Hofer_thesis.pdf">Development and Characterization of High-Efficiency, High Specific Impulse Xenon Hall Thrusters</a></p><p><a href="https://core.ac.uk/reader/161999685">100 kW Nested Hall Thruster System Development</a></p><p><a href="https://www.researchgate.net/publication/269068626_Propellant_alternatives_for_ion_and_Hall_effect_thrusters">Propellant Alternatives for ion and Hall effect thrusters</a></p><p><a href="https://pepl.engin.umich.edu/project/x3-nested-channel-hall-thruster/">X3 — Nested Channel Hall Thruster</a></p><p><a href="https://www.nasa.gov/centers/glenn/about/fs21grc.html">NASA — Ion Propulsion</a></p><p><em>Have feedback or questions? Send me an email at chloewang.lv@gmail.com and I’ll be happy to respond!</em></p><p><em>You can check out more of what I’m up to in my monthly newsletters. Sign up </em><a href="https://docs.google.com/forms/d/e/1FAIpQLSeAEWbVJplZ_njSIDfs9CWegCagrhYfq2o6789DDh3hD8cwpQ/viewform?usp=sf_link"><em>here</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bf19d44ee2fa" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>