Sorting Accuracy Guide | AISORT
Technical Guide
Understanding Sorting Accuracy: Purity, Recovery, and Efficiency Metrics
Sorting accuracy is the defining performance metric for any optical sorter — but "accuracy" means different things in different contexts. A sorter achieving 98% purity with 85% recovery operates very differently from one delivering 95% purity with 98% recovery, even though both might be described as "high accuracy." Understanding the distinction between these metrics is essential for specifying equipment, evaluating vendor claims, and optimizing line performance.
Core Sorting Metrics
Purity (Grade)
Definition: The percentage of the accept fraction that is correctly classified target material.
Formula: Purity = (Correctly Accepted Target) / (Total Accepted Material) × 100%
Example: If the accept bin contains 980kg of target PET and 20kg of non-PET contaminants, purity = 980/(980+20) = 98%.
When it matters most: Food-grade applications where contaminant limits are regulatory (EFSA/FDA); high-value material where contamination downgrades the entire bale; specifications where contract penalties apply per percentage point below target purity.
Recovery (Yield)
Definition: The percentage of target material in the feedstock that is correctly directed to the accept fraction.
Formula: Recovery = (Correctly Accepted Target) / (Total Target Material in Feedstock) × 100%
Example: If the feedstock contains 1,000kg of target PET and 980kg ends up in accept, recovery = 980/1,000 = 98%. The missing 20kg was falsely ejected into the reject fraction (false negative).
When it matters most: High-value target material where every percentage point of lost recovery directly reduces revenue; positive sorting applications where the objective is to extract maximum value from a mixed stream.
The Purity-Recovery Trade-off
Purity and recovery trade off against each other. Increasing the ejection threshold to remove more contaminants also ejects more target material (reducing recovery). Relaxing the threshold to capture more target material also allows more contaminants through (reducing purity). The optimal operating point depends on the economic value of purity vs. recovery for the specific application.
| Operating Mode | Purity | Recovery | Best For |
|---|---|---|---|
| Purity-priority | Maximum achievable | 80-90% | Food-grade rPET, pharmaceutical-grade materials, contract-specified purity requirements |
| Balanced | 95-98% | 90-95% | Most commercial recycling applications; bottle-grade rPET; engineering plastics recovery |
| Recovery-priority | 85-92% | 95-99% | Low-value target where volume matters more than purity; pre-concentration before secondary sorting; mining pre-concentration |
Factors Affecting Sorting Accuracy
Feedstock Presentation
The single most controllable factor in sorting accuracy is how material is presented to the sensors. A sorter can only classify what it can see:
- Monolayer vs. multilayer: Material must be spread in a single layer across the chute. Overlapping particles hide each other from sensors. Monolayer presentation typically limits throughput to 0.5-1.5 t/h per meter of chute width (material-dependent).
- Particle spacing: Particles must be separated enough for the ejection system to target individual items. If two particles pass the detection zone too close together, a single valve firing may eject both — a good particle and a contaminant together.
- Chute speed and angle: Faster chute speed increases throughput but reduces time in the detection zone, potentially degrading classification accuracy. Steeper angle improves singulation but increases particle velocity at the detection point.
- Dust and fines: Fines coat sensor windows, scatter light, and create false signals. Pre-screening material to remove the <2mm fraction improves both detection accuracy and sensor maintenance intervals.
Sensor Resolution
Sensor resolution determines the minimum detectable feature size:
- Spatial resolution: Pixel size at the material plane. Typically 0.1-1.0mm/pixel for optical sorters. Smaller pixels detect smaller contaminants but increase data volume and processing requirements.
- Spectral resolution: Number of wavelength bands (for hyperspectral systems). More bands = better material discrimination but higher cost and slower processing.
- Temporal resolution: Scan rate (Hz). Must be fast enough that particles don't move more than one pixel between scans. At 3 m/s belt speed with 0.2mm pixels, minimum scan rate is 15,000 Hz.
Ejection System Performance
The ejection system — typically an array of high-speed pneumatic valves — must physically remove contaminants identified by the sensors:
- Valve pitch: Spacing between adjacent valves. Smaller pitch = more precise ejection targeting. Typical: 6-25mm pitch, with finer pitch for small-particle applications like flake sorting.
- Valve response time: Time from detection signal to full valve opening. Typical: 0.5-2.0 ms. At 3 m/s particle velocity, a 1ms delay means the air jet hits 3mm behind the intended point.
- Valve cycle time: Minimum time between successive firings. Determines maximum ejection frequency. Typical: 3-5 ms minimum cycle, allowing 200-300 ejections per second per valve.
Measuring Accuracy: Challenge Tests
The gold standard for verifying sorting accuracy is a challenge test — running a known-quantity feedstock with deliberately introduced contaminants through the sorter and measuring the outcome:
- Prepare 100-500kg of clean target material (verified by hand-sort).
- Add a known quantity of contaminants (e.g., 100 pieces of PVC of known size and color).
- Run the material through the sorter at the target throughput rate.
- Hand-sort the accept and reject fractions, counting correctly and incorrectly classified items.
- Calculate purity (% of accept that is target), recovery (% of target in accept), and false-eject rate (% of target sent to reject).
A proper challenge test uses contaminants representative of the real feedstock and tests multiple contaminant types and sizes. It should be repeated at two or three throughput levels to establish the throughput-accuracy curve.
Continuous Accuracy Monitoring
Sorting accuracy drifts over time due to sensor window fouling, valve wear, belt/chute surface changes, and feedstock variation. Best practices for maintaining accuracy:
- Daily check sort: Run a small reference sample (50-100kg with known composition) through the sorter at the start of each shift. Compare results to baseline.
- Automated sensor cleaning: Program automatic sensor window cleaning cycles (compressed air purge or wiper) at regular intervals — typically every 15-60 minutes depending on dust load.
- Valve health monitoring: Track individual valve response times. A valve that slows down by 0.5ms may not fail outright but will progressively degrade ejection accuracy.
- Trend logging: Log purity and recovery trends over time. A gradual decline often indicates sensor window fouling or belt wear. A sudden drop suggests a valve failure, sensor malfunction, or feedstock change.