Green Living
ISDN2001/2002: Second Year Design Project
Near Infrared Spectroscopy
In our year-long project RealCycle, we would like to identify PET and non-PET plastic bottles using discrete infrared spectroscopy. This informs the sorting mechanism whether it should sort the incoming bottle into the PET or the non-PET bin.
Common types of plastic
PET (Polyethylene terephthalate)
widely used in food packaging
HDPE (High-density polyethylene)
commonly used in gallon milk jugs
PP (Polypropylene)
heat resistant,
durable, resistant to fatigue
Most recycling service providers focus on processing one type of plastic
Two modes of operation
9 near-infrared LEDs and 1 InGaAs photodiode are used in our NIR spectrometer.
Reflectance
-
Infrared light emitted from frontend A.
-
NIR LEDs driven by BJT transistors.
-
-
Some light rays are reflected by the material.
-
Frontend A picks up the reflected light.
-
InGaAs photodiode current amplified by AD8606 transimpedance amplifier
-
read by ADC of ESP32
-
Transmittance
-
Infrared light emitted from frontend B.
-
NIR LEDs driven by BJT transistors.
-
Some light rays pass through the material.
-
Frontend A picks up the reflected light.
-
InGaAs photodiode current amplified by AD8606 transimpedance amplifier
-
read by ADC of ESP32
-
The iterative prototyping process
To understand the behaviour of the invisible near-infrared light, we began by constructng an experimental optical setup.
Raw data visualisation
The measured intensities of the 9 reflected wavelengths are plotted as raw data.
The dual mode generation
A mid-fidelity setup was made to evaluate the feasibility of simultaneously measuring reflectance and transmittance.
Classification
On the bottom left corner the actual type is shown while on the graph the classified type is shown. RED is PET and BLUE is non-PET.
The third spectrometer
Constructed with acrylic and PLA, the latest generation of spectrometer is refined to a all-in-one coherent design.
Data processing
What is PCA?
Quartl, CC BY-SA 3.0 <https://creativecommons.org/licenses/by-sa/3.0>, via Wikimedia Commons
Principal Component Analysis
Variance
Data looks too mangled together? Maximise the separation of data.
Dimensionality reduction
Too much data to handle? Give up some of them by jumping onto a lower dimension.
Projection
It's like casting a shadow. But with mathematics.
Linear combination
PCA is not strictly reversible since data is discarded. But it can reproduce as close as possible with linear combination.
Classification
After extracting the high variance features through PCA, support vector machine is used as a classifier to draw the boundaries between the points produced by PCA, thus collapsing the coordinates of the points into discrete classes in terms of the type of plastic.
On the bottom left corner, the actual type of plastic is indicated.
Final verdict
Although RealCycle’s discrete NIR spectrometer is no panacea for sorting plastic types, it serves as a proof of concept that a low cost and lightweight alternative to the expensive and bulky laboratory counterpart can be implemented.