Nov 16, 2026 | 1958 words | 20 min read
17.6. Milestone 2 - Insulation and the Heat-Loss Model (Team)#
At a Glance
What you will do: Measure the heat loss of GHU-2’s wall in the sandbox, compare two regression models of it at four times of day in Python or Excel, and check the wall against its rated U-value. Then choose GHU-2’s glazing and heater, and size its battery.
What you will submit: Per team, one technical write-up (PDF) containing your results table, the regression plots, and the design justification, your sandbox data file, and your Python code or Excel workbook.
17.6.1. The Heat-Loss Model#
The crop needs a stable temperature inside the greenhouse to survive. Mars is extremely cold, so the greenhouse loses heat to the Martian atmosphere through its outer walls and glazing at all times.
This relationship models the rate of conductive heat loss, \(Q\), in watts (\(\text{W}\)).
where:
\(Q\) is the heat-loss rate (\(\text{W}\)).
\(U\) is the thermal transmittance, or U-value (\(\text{W}/(\text{m}^2\cdot\text{K})\)). The U-value is an insulation constant that represents how easily heat passes through a material. A lower U-value means better insulation.
\(A\) is the total structural surface area of the greenhouse (\(\text{m}^2\)). For the GHU-2 design, this area is fixed at \(43.2\text{ m}^2\).
\(\Delta T\) is the temperature difference across the glazing (\(\text{K}\) or \(^\circ\text{C}\)), calculated as \(\Delta T = T_{\text{in}} - T_{\text{out}}\).
The heat-loss rate \(Q\) is directly proportional to \(\Delta T\). The model predicts no heat loss (\(Q = 0\)) when there is no temperature difference (\(\Delta T = 0\)), so the model is a line through the origin.
Plot \(Q\) on the vertical (\(y\)) axis and \(\Delta T\) on the horizontal (\(x\)) axis. The terms of the two equations then correspond as follows.
The independent variable (\(x\)) is \(\Delta T\).
The dependent variable (\(y\)) is \(Q\).
The model’s \(y\)-intercept (\(b\)) is \(0\).
The slope (\(m\)) of the best-fit line is equal to the product of the insulation constant and the surface area.
\[\text{slope} = U \cdot A\]
Measured data rarely fits a model exactly. Two fits test this model against data: one with the intercept fixed at zero, as the model predicts, and one with a fitted intercept. Random error scatters the points about either line. A systematic error that shifts every point by the same amount moves the fitted intercept away from zero and leaves the slope unchanged.
17.6.2. Measuring the Heat Loss of GHU-2#
The hardware menu below lists a rated U-value for each glazing. GHU-2’s as-built twin-wall glazing is rated at \(U = 0.6\ \text{W}/(\text{m}^2\cdot\text{K})\). Before your team compares glazing options, measure whether the as-built wall meets its rating.
At night the greenhouse receives no sunlight. With the grow LEDs off, the heater is the only source of heat. When the greenhouse temperature stops changing, the heater supplies heat at the same rate the wall loses it, so the heater power is a measurement of \(Q\). During the day, sunlight through the glazing also heats the greenhouse. Running the heater at several power levels and recording the temperature difference each level sustains produces the data for the regression. Building engineers use the same method, called a co-heating test, to measure the heat loss of a finished house.
Run the Test in the Sandbox#
From inside the unzipped sandbox folder from Operating the Greenhouse, launch the sandbox with the as-built hardware and your team number as the seed. For team 12:
python run_sandbox.py 12 --glazing twin_wall --heater 2000
Set the LED, CO2 valve, and pump sliders to 0. Then complete one run for each heater setting: 0, 25, 50, 75, and 100 %.
Press Reset. The clock returns to Sol 0 00:00 LMST and the greenhouse returns to \(15\ ^\circ\text{C}\). The sliders keep their positions.
Confirm that the title above the greenhouse reads
Mode: MANUAL. Reset does not clear the Auto checkbox, and in Auto mode the rules overwrite the heater slider.Set the heater slider to the setting for this run.
Enter Sol 0 and 12:00 in the Run to fields above the greenhouse, then press Run to. The sandbox runs at the selected speed and pauses when the clock reaches 12:00 LMST on Sol 0.
After the fifth run, press Save data (CSV) and save the file. The file holds every
reading from every run since the sandbox opened, one row per simulated minute, with a
run column that numbers the runs. If a run goes wrong, press Reset and repeat it. The
file keeps every run, so use the mode and heater_pct columns to choose the five good
ones.
Each run starts from the same state, so at any clock time the outside conditions are identical in every run, and the heater setting is the only difference between runs. The crop dies in every run. This test measures the wall, not the crop.
If you cannot run the sandbox
Use ghu2_coheating_log.csv, saved from the sandbox
with this procedure at seed 42.
Fit the Models#
From the file, take one row per run at each of four clock times in the lmst column:
01:00, 02:00, 08:00, and 12:00. If two rows of a run share a clock time, use the first.
Convert heater_pct to watts with the heater’s \(2000\ \text{W}\) rating, and calculate
\(\Delta T\) from t_in_c and t_out_c.
At each clock time, fit two linear models of \(Q\) against \(\Delta T\): one with a fitted intercept, and one with the intercept fixed at zero. For both models, calculate the sum of squared errors (SSE) from the residuals, and calculate the coefficient of determination with the total sum of squares (SST) taken about the mean of \(Q\).
Choose either the Excel Workflow or the Python Workflow below to complete the analysis.
Excel Workflow: Review the Part 02 Excel M4 module. Filter the data to the four clock times. Use
=LINESTwith its third argumentTRUEfor the model with an intercept andFALSEfor the model through the origin. When the third argument isFALSE, LINEST reports an \(R^2\) with SST taken about zero instead of about the mean, so calculate both \(R^2\) values yourselves from SSE and SST.Python Workflow: Review the Part 04 Python M5 module. Load the file with
pandasand filter it to the four clock times. For each time, letxbe \(\Delta T\) andybe \(Q\). Fit the model with a fitted intercept usingnp.polyfit(x, y, 1); it returns the slope and intercept. For the model through the origin, set the intercept to 0 and calculate the slope asnp.sum(x * y) / np.sum(x**2). For both models, calculate predicted values, SSE, and \(R^2\) from the residuals.
Interpret the Fits#
Report the following in your write-up.
A table with one row per clock time and model: slope, intercept, SSE, \(R^2\), and the U-value from the slope.
Plots of \(Q\) against \(\Delta T\) at 02:00 and at 12:00, each with both fitted lines.
Which model describes the data at each clock time, and what SSE and \(R^2\) each show about the comparison.
Why the 01:00 slope differs from the later slopes.
How the fitted intercept changes from 02:00 to 12:00, what the 12:00 intercept represents physically, and how the 0 % run shows the same effect.
Whether the as-built wall meets its rated U-value, and by what margin.
If the fitted-intercept U-values at 02:00, 08:00, and 12:00 differ from the rating by more than 5 %, check the procedure before you continue: Manual mode on every run, a Reset before every run, and all five heater settings present at each clock time. The trade study below uses the rated U-value of each glazing.
17.6.3. Greenhouse Hardware Design Choice#
GHU-2’s hull, sensors, and grow LEDs are final. Your team chooses two other parts of its hardware, the wall glazing and the heater. Mission Control will fly the combination your team specifies. Your choice sets the physics your controller must handle. In Milestone 4 you declare the choice in your controller package, and the acceptance test runs your controller on the hardware you chose.
Glazing |
U-value |
Light transmittance |
Areal mass |
Material cost |
|---|---|---|---|---|
Single-layer polycarbonate |
\(1.8\text{ W}/(\text{m}^2\cdot\text{K})\) |
88 % |
\(2\ \text{kg}/\text{m}^2\) |
200 USD per square meter |
Twin-wall polycarbonate (as built) |
\(0.6\text{ W}/(\text{m}^2\cdot\text{K})\) |
80 % |
\(3.5\ \text{kg}/\text{m}^2\) |
400 USD per square meter |
Aerogel multi-layer |
\(0.3\text{ W}/(\text{m}^2\cdot\text{K})\) |
50 % |
\(8\ \text{kg}/\text{m}^2\) |
1,500 USD per square meter |
Vacuum-insulated panel |
\(0.15\text{ W}/(\text{m}^2\cdot\text{K})\) |
0 % (opaque) |
\(12\ \text{kg}/\text{m}^2\) |
4,000 USD per square meter |
Light transmittance is the fraction of outside sunlight that reaches the crop. Better insulation generally passes less light. Any light the glazing blocks must come from the grow LEDs instead.
Heater |
Rated power |
Mass |
|---|---|---|
Standard (as built) |
\(2000\ \text{W}\) |
\(6\ \text{kg}\) |
Upgraded |
\(3000\ \text{W}\) |
\(9\ \text{kg}\) |
GHU-2 draws its power from solar arrays and a battery. Size the battery for the worst case, a sol inside a dust storm. During that sol the arrays deliver almost nothing, and the battery alone must run the heater and the LEDs for a full sol (\(24.66\) hours). Space-rated battery packs store \(125\ \text{Wh}/\text{kg}\), and only \(80\%\) of a pack’s rated capacity may be used. The battery mass is therefore
Mission Control budgets launch at roughly 50,000 USD per kilogram delivered to the Martian surface. It sets a ceiling of \(600\ \text{kg}\) on the configurable launch mass, which is the glazing, heater, and battery together.
Design Trade-offs and Calculations#
Evaluate all four glazing options at the representative Martian cold-night condition. At that condition the inside setpoint is \(18^\circ\text{C}\) and the outside low is \(-70^\circ\text{C}\) (\(\Delta T = 88\text{ K}\)). Use the surface area \(A = 43.2\text{ m}^2\) to compute the following for each option.
The heat-loss rate (\(Q\)).
The heater capacity usage percentage, for both heater ratings. If an option’s heat loss exceeds what a heater can supply, say so, and state what happens to the crop on a cold night.
The nightly heating energy over a Martian night of \(12.33\) hours (half of one sol), assuming the cold-night heat loss holds all night. This overstates the energy, because the outside temperature is not at its minimum all night. State that assumption in your write-up.
The glazing mass for the greenhouse, and its launch cost.
Then choose your two strongest glazing and heater combinations. For each one, estimate the energy of a dust-storm sol. In that sol, the heater must cover the heat loss through the whole sol with almost no solar gain, and the LEDs must supply most of the crop’s light. State every assumption you make, including the LED power level and hours and the average outside temperature. Then convert the energy into a battery mass and a total launch mass.
Your estimate will not match the simulator exactly. In Milestone 4, the acceptance test reports the worst-sol energy your controller draws. Comparing the two is part of your gap analysis.
Use this check to catch a spreadsheet error early. The four options span a wide range of insulation quality, so your four heat-loss numbers should bracket the \(2000\ \text{W}\) heater. At least one option should fall below \(100\%\) of its capacity and at least one above it. If all four fall on the same side, re-check your arithmetic before you interpret the result.
In your write-up, justify a provisional choice by weighing launch mass, light, and heating margin against one another. No option is best on all three, so your justification should state which trade-off your team accepts and why. This analysis feeds into the Idea Generation and Weighted Decision Matrix sections of your final Design Project Technical Report.
Also, in the write-up, name who ran the sandbox test, who ran the regression, who verified the regression independently (with the second tool or a hand calculation), and who authored the design justification.
Important
Your hardware choice is provisional until Milestone 3. In Milestone 3, your team weighs hardware and controller architecture together and may revise the choice. Record any change and the reason in your decision matrix. From Milestone 4 on, the acceptance test runs your controller on the hardware you declare. The test fails any design whose launch mass exceeds \(600\ \text{kg}\).
17.6.4. Submission#
Submit one PDF write-up named
m2_heat_loss_model_write_up_team_#.pdfand either one Excel workbook namedm2_heat_loss_model_analysis_team_#.xlsxor one Python script namedm2_heat_loss_model_analysis_team_#.py, depending on which tool your team used for its analysis, per team to 17.6 Milestone 2 - Insulation and the Heat-Loss Model (Team) in Gradescope, with all members added.Submit before the deadline on the Course Schedule.
Learning Objectives
- DV01:
Table and Plot Formatting - Prepare tables and plots for technical presentation with proper formatting. Include descriptive titles, axis labels with units, appropriately scaled axes, and visible legends.
- DV04:
Regression Model Construction - Construct regression models that describe relationships between variables and support prediction.
- DV05:
Regression Model Evaluation - Evaluate and compare regression models using quantitative metrics (SSE, SST, and R²) and contextual reasoning.
- UC03:
Mathematical Modeling of Systems - Define engineering systems or processes using mathematical models that represent inputs, outputs, and accumulations.
- UC05:
System Efficiency and Performance - Calculate and interpret the efficiency of a system, product, or process in terms of energy, cost, or other engineering performance measures.
- EB01:
Testing and Comparison - Test prototypes and analyze results to compare alternative solutions.
- EB04:
Metrics and Weighting - Justify evaluation metrics and weighting based on stakeholder priorities.
- SQ01:
Technical Accuracy - Use accurate scientific, mathematical, and technical concepts, units, and data.
- EB02:
Assumptions Under Uncertainty - Identify assumptions made when information is limited or unavailable.
- PR01:
Programming Standards - Develop code that follows established programming standards and conventions.
- TM02:
Document Contributions - Provide clear evidence of individual contributions to the team’s work, including technical, written, or supportive tasks.