Visual analytics for uncovering complex learning: Part II

Figure 1. Most students performed well throughout.

In an early article, I have introduced the idea of using graph theory to visualize learning. In this follow-up article, I will give an example based on our recent classroom study with our mixed-reality technology that engaged 65 high school students in three chemistry classes to learn the kinetic molecular theory through exploring gas laws. The primary task of the study is to investigate how a novel kind of mixed-reality technology that combines actions in and across real and virtual worlds can help students develop molecular reasoning skills to explain macroscopic phenomena with microscopic theories.

Figure 2. Less successful than Figure 1.

The key is to study students' interactions with the technology in great details in order to shed light on how the mixed-reality experience may help students make micro-macro connections. For most students, there is a huge gap between their macro perception and their micro conception. We want to fill the gap using this powerful real-time technology to provide just-in-time, integrated instruction.

Figure 3. Performed well initially and then went astray.

The challenge is to find a way to describe both the interaction process and the reasoning process. Our software logs everything students are doing behind the scenes (i.e., we can track students' hands). Meanwhile, through our embedded assessment using causality maps (a type of concept map), we can track students' minds. Our objective is to find any conceptual changes from the "mind-tracking" data and attribute them to the "hand-tracking" data. Only after we establish this association can we evaluate the worth of the mixed-reality technology in fostering science learning.

Both the "hand-tracking" and "mind-tracking" data are being studied using graph theory. In this article, we will show the analysis of the "mind-tracking" data first. The data were generated by students connecting 6-7 provided molecular concepts between a cause and an effect to compose an explanation. Some of the concepts are irrelevant and used as distractors. For example, increasing the number of molecules has nothing to do with the molecular mass.

Given the fact that there are 65 students and four causality maps, we need a way to quickly visualize student learning. Our solution is to construct graphs based on these data. It becomes immediately obvious that this visual analytics can provide extremely informative graphics that shows student learning on a statistical basis. The three images in this article show that: 1) Most students in Class E performed well throughout Graph 3; 2) Students are less successful in Class B in Graph 1; and 3) Students in Class D performed well in Graph 4 at the beginning and then went astray -- largely because this is a more difficult challenge that involves multiple reasoning paths. (The gray bands in the three images represent the correct reasoning chains in each case. The thickness of an edge represents the number of students who drew the link in the concept map.) These graphs can potentially show the weaknesses of the students at the scale of the whole class. I could not help thinking how useful this would be to the teacher if such informative feedback is provided to her.

This graph-based analytics is NOT just visual -- It is also interactive. You can examine all the student data we have collected at this page, programmed in Dart -- the latest Web programming language from Google. The page provides filters for you to examine and compare any number of students' work. You can also drag the nodes around for clarity. This kind of feedback tool is what teachers would want to have and what we should create for them. The world has spent billions of dollars building visual business analytics to assist business executives to make investment decisions day to day. How about spending a bit money to build an infographics system to assist teachers to make instructional decisions day to day?

Simulating thermal radiation with Energy2D

Figure 1: Stefan's Law in action.

The original ray-tracing radiation solver in our Energy2D software suffers from performance problems as well as inaccuracies (no, light particles do not travel that slowly as shown in it). After some sleepless nights, I finally implemented a real radiation solver, coupled it with the heat and fluid solvers, and supported both the convex and concave shapes (see this short paper for the mathematics and the algorithms). At last, Energy2D is capable of simulating all three heat transfer mechanisms in a decent way.

Figure 2: Radiation in a box.

Able to simulate heat, fluid, radiation, particles, and any combination of them, Energy2D is now one step closer towards a full multiphysics capacity. Despite the fact that all these complex calculations are done in real time on a single computer, the software still runs at a pretty amazing speed on an average Windows tablet (such as the Surface Pro). I guess this is why our industry friends love it (although Energy2D is mostly designed for K-12 students, to my surprise, quite a number of engineers are using it to do conceptual product design). Who doesn't like a CFD tool for dummies that can save time from the long preprocessor-solver-postprocessor cycle?

Figure 1 shows a simulation that illustrates radiation heat transfer. As you can see, energy can "jump" from a high-temperature object (a radiator) to a low-temperature one without heating the medium between them (unlike the cases of conduction and convection). Users

Figure 3. Radiation in a circle.

can adjust the temperature of the radiator on the left and investigate how the radiation heat transfer increases with respect to the temperature, as per Stefan-Boltzmann's Law. The image also shows the view factor field used in the computation. The simulation provides many subtleties. For example, if you observe carefully, you can find that the radiation barrier used to separate the left compartment from the right one increases the heating on the right side of the upper left object and the left side of the upper right object -- because it reflects the radiation from the two radiators at the lower part of the box to the two sides!

Figures 2 and 3 show radiation among different shapes in an enclosed space. They show how accurate the radiation solver may be. The radiation heating on the side walls seems to make sense. In Figure 2, the upper one gets the most radiation energy because it is the closest to the radiator. The right one gets the least because part of it is blocked from the radiator by the other object in a box. A further test case using a symmetric setup shows its accuracy.

Simulating PTC and NTC heating elements with Energy2D

Figure 1: A demo simulation.

A heating element converts electricity into heat through Joule heating: Electric current passing through the element encounters resistance, causing the temperature of the element to rise. A thermistor is a type of resistor whose resistance changes significantly with temperature. In a heating element that uses a thermistor with a positive temperature coefficient (PTC), called a PTC heating element, the temperature increases rapidly. In a heating element that uses a thermistor with a negative temperature coefficient (NTC), called a NTC heating element, the heating will gradually weaken when the temperature increases.

Figure 2: Setting the temperature coefficient.

Several Energy2D users have requested adding PTC/NTC controls to the software. So this was added last night. You can now set the temperature coefficient while defining a power source, as shown in Figure 2.

Figure 1 shows the comparison of the temperature increasing in a PTC heater, a constant-power heater, and a NTC heater, with the temperature coefficients being 0.1, 0, and -0.1, respectively. Note that in the case of constant power, the temperature increases linearly in time (as per the definition of constant power), whereas PTC and NTC exhibit nonlinear behaviors.

You can click the link under the image to run the simulation yourself.

Design replay: Reconstruction of students' engineering design processes from Energy3D logs

One of the useful features of our Energy3D software is the ability to record the entire design process of a student behind the scenes. We call the reconstruction of a design process from fine-grained process data design replay.

Design replay is not a screencast technology. The main difference is that it records a sequence of CAD models, not in any video format such as MP4. This sequence is played back in the original CAD tool that generated it, not in a video player. As such, every snapshot model is fully functional and editable. For instance, a viewer can pause the replay and click on the user interface of the CAD tool to obtain or visualize more information, if necessary. In this sense, design replay can provide far richer information than screencast (which records as much information as the pixels in the recording screen permit).

Design replay provides a convenient method for researchers and teachers to quickly look into students' design work. It compresses hours of student work into minutes of replay without losing any important information for analyses. Furthermore, the reconstructed sequence of design can be post-processed in many ways to extract additional information that may shed light on student learning, as we can use any model in the recorded sequence to calculate any of its properties.

The three videos embedded in this post show the design replays of three students' work from a classroom study that we just completed yesterday in a Massachusetts high school. Sixty-seven students spent approximately two weeks designing zero-energy houses -- a zero-energy house is a highly energy-efficient house that consumes net zero (or even negative) energy over a year due to its use of passive and active solar technologies to conserve and generate energy. These videos may give you a clue how these three students solved the design challenge.