Biomass boiler · final-year project with Thermax · simulation and model
A rice-straw boiler you can turn the dials on
Two 3D views of the same boiler, driven live by a heat-loss model built from its design workbook: the plant seen from outside, and the flow of gas, water and steam inside. Change a dial and both views, and every chart, change at once. At the design point the boiler is 81.3 percent efficient.
1. Dials
2. The plant from outside
Drag to turn, scroll to zoom. The fuel conveyor and the feed fan speed follow the load and the air, the sight glass glows with the flame, the stack lamp and plume show alarms, smoke and steam. Details like platforms and the filter are for scale and are illustrative.
straw steam smoke
3. Inside: gas, water and steam
The same boiler cut open. Hot flue gas (colour = temperature) passes the furnace, settling chamber, the 127-tube convection bank and the economizer, then leaves by the stack. Feed water (blue) is heated in the economizer, rises to the steam drum and leaves as steam (white). The furnace, settling chamber and bank temperatures are the workbook's iterated values; the stack temperature and air are yours.
hot gas cooled gas feed water steam
4. Live charts
Live chart 1. Efficiency and useful heat (MW) over the last 60 seconds. Move a dial and watch the line react.Live chart 2. Where the fuel heat goes over the last 60 seconds: flue gas, unburnt fuel (plus the assumed low-air penalty) and radiation, stacked.Live chart 3. Stack temperature and oxygen in the dry flue gas over the last 60 seconds.Live chart 4. Efficiency against excess air for the current stack temperature, fuel water and penalty setting. The dot is where the boiler is now. The dashed line is the same curve without the assumed low-air penalty.Live chart 5. Which dial matters most right now: the change in efficiency for a step in each dial (air ±10 points, stack ±20 °C, unburnt ±1 point, fuel water ±25 kg/t).
5. Run a 12-hour shift
The fuel gets wetter and drier. A policy sets the excess air every 30 minutes and the 3D boiler follows. The score is the gap to the best possible setting.
Assumption, not measurement: the workbook has no penalty for too little air, so a stylised rule adds one: below a minimum air (35 % plus 0.15 % per kg of extra fuel water per tonne) the efficiency drops 0.25 points for every point of air missing. Under that rule the best setting at each step is exactly that minimum. Replace it with a plant curve before drawing conclusions.
Extra water in the fuel (kg per tonne). Wetter fuel needs more air.Excess air chosen (solid) against the assumed minimum air (dashed).Efficiency of each step against the provable optimum under the assumed rule.
6. Scores over 200 shifts
Mean over 200 seeded scenarios, computed in Python (simulation/evaluate.py). The live run above uses a JavaScript random generator, so one scenario is not identical to the same seed in Python; averages agree closely.
Policy
Mean efficiency (%)
Gap to optimum (points)
Mean score
The same table as a chart: the gap to the optimum, smaller is better.
7. What the heat-loss model says
efficiency at the design point
90 % range with uncertain inputs
6.96 kgflue gas per kg of fuel
Where the fuel heat goes at the design point (70 percent excess air, 200 °C stack).How far efficiency moves when one input changes over a plausible range. Stack temperature and unburnt loss matter most (about 4 points each).Efficiency from 5,000 random draws of the uncertain inputs (my assumed ranges). The middle 90 percent lies between 78.7 and 83.6 percent.
The model reproduces the workbook's flue gas mass (6,958 kg per tonne of fuel) and its oxygen share (7.77 %).
It is a design calculation, not plant data. The unburnt and radiation losses are fixed values from the workbook; the 3D plant details (platforms, filter) are illustrative.
An earlier machine-learning test found nothing better than a straight line, because the data came from a straight-line formula. That is why the heat-loss model is the core.