Case Study 1: Exceeding Bale-Spec for Municipal Waste Sorting
Written by Georgia Crowther, Founder and CEO
Municipal Waste Sorting Goal - Reaching Bale Spec.
In order to actually recycle post-consumer plastics, waste must be sorted into bales by resin number. The purity of recycled plastic bales is critical, as contaminants affect the quality of the recycled product and can damage recycling equipment. Therefore, our goal for this case study was to train our Acoustic AI model to reach the bale purity specifications for municipal plastic waste as defined by the Association of Plastic Recycler’s Model Bales Specifications.
Generally, plastic waste must be sorted with no more than between 2-4% contamination from other plastic resins to be considered a salable bale. In order to achieve this benchmark, our Acoustic AI model should reach an average 97% precision.
Acoustic Data Collection Method
Reclamation Factory collected thousands of samples of municipal plastic recycling during our summer 2025 pilot with the City of Pittsburgh’s Department of Public Works through PGHLab. Of these samples, we built a dataset containing hundreds of acoustic samples from each of main four consumer resin types (#1-PET, #2-HDPE, #5-PP, and #6-PS). Audio features extracted from these acoustic samples were used to train our Acoustic AI for municipal plastic sorting.
Acoustic Sorting Results
Our final municipal waste Acoustic AI sorting model achieves an average of over 97% precision across mixed municipal waste classes from the four main resin types. Importantly, bale-spec precisions were exceeded for the two most common consumer resins, PP and PET (nearly 99% and 97% precision respectively).
These results are form agnostic, meaning that our model works across all consumer form-factors (bottles, clamshells, tubs, etc.) for each resin, and can achieve these results using acoustic analysis as the sole sensing modality.
Compared to other state-of-the-art plastic sorting technologies, like visual and hyper-spectral imaging, Acoustic AI can be deployed onto sorting lines at much lower costs. This advancement can make achieving bale precision a new possibility for a wide array of material recovery facilities.
Next Steps and Further Improvements
Reclamation Factory’s Municipal Waste Acoustic AI model results can be further improved with a larger (and more diverse) data set. Audio samples collected through municipal recycling pilots will continue to increase both our precision and throughput rates. We are currently welcoming new pilot customers to try Acoustic AI on their material streams.
Our initial studies have also shown that fusing visual inference with acoustic inference can increase precision to nearly 100% for many municipal form-factors and resin types. Our previous success with visual waste form-factor recognition and our existing visual dataset will accelerate this multi-modal approach and enable even more detailed sorting specifications, including features like color, shape, and cleanliness.
Join a Municipal Waste Sorting Pilot Program Today
For recyclers, manufacturers and secondary-material partners that struggle with waste sorting, we invite you to join a pilot study to test our process on your waste stream. Contact our team today to explore how acoustic sensing can advance your business and improve bale grades for your operation.