A practical, no-nonsense guide for plant operators, procurement managers, and investors evaluating sorting equipment. Based on real deployment experience across 40+ countries.
Understanding the landscape is the first step. Sorting equipment falls into six major categories. Most modern recycling lines use a combination.
| Technology | How It Works | Best For | Typical Purity | Throughput Range |
|---|---|---|---|---|
| AI Optical Sorting Recommended | RGB cameras + NIR sensors + deep learning models identify materials by visual and spectral signatures. Ejectors remove targeted items. | Plastics by polymer/color, e-waste, mixed recyclables, high-purity rPET, brand-level sorting | 90–97% | 1–30 t/h per unit |
| NIR Optical Sorting | Near-infrared light reflected off materials; each polymer has a unique spectral fingerprint. | Plastic polymer separation (PET vs HDPE vs PP), paper/cardboard | 82–92% | 1–25 t/h per unit |
| Hyperspectral Imaging | Captures hundreds of spectral bands (vs 4–6 for NIR). Distinguishes visually identical materials. | Multi-layer packaging, challenging polymer blends | 85–93% | 0.5–8 t/h per unit |
| X-Ray Transmission (XRT) | X-rays pass through materials; density and atomic number determine absorption. | Metals from non-metals, aluminum from heavy metals, PVC from other plastics | 88–96% | 2–20 t/h per unit |
| Eddy Current Separator | Rotating magnetic field induces repulsive force in conductive non-ferrous metals. | Aluminum cans, copper, brass from mixed waste | 90–97% | 3–35 t/h per unit |
| Magnetic Separation | Permanent magnets or electromagnets attract ferrous metals. | Steel cans, iron scrap from MSW and C&D waste | 95–99% | 5–50+ t/h per unit |
| Robotic Picking Complementary | Robot arms with suction grippers, guided by AI vision, pick targeted items from conveyor. | Last-chance quality control, positive picking of high-value items | 85–95% per pick | 40–80 picks/min per robot |
| Ballistic / Mechanical Separation | Screens, discs, and paddles separate by size, shape, and density. | 2D (film/paper) vs 3D (containers) separation, size fractionation | 80–90% | 10–60 t/h per unit |
AI optical sorting has become the default choice for new installations in developed markets. The cost premium over conventional NIR has narrowed to ~20–30%, while purity gains of 5–15 percentage points translate to significantly higher bale prices. If you're buying new equipment today and not evaluating AI options, you're leaving money on the table.
Don't evaluate sorting equipment in a vacuum. Use this framework to compare options systematically.
The core metrics. Purity = how clean the output stream is. Recovery = what % of target material was captured. They're a tradeoff — higher purity usually means lower recovery. Ask vendors for both numbers on your material, not their best-case demo.
Match throughput to your line speed. Oversized equipment wastes capital; undersized creates bottlenecks. Always ask: throughput at what purity? A machine that runs faster but dirtier may not help.
Can the sorter handle your current material mix and likely future changes? If you process packaging today but may add e-waste tomorrow, a flexible AI system costs more upfront but avoids replacement later.
False positive = wrong material in output (contamination). False negative = target material sent to residue (lost revenue). Ask for confusion matrices on your material types. Vendors who can't provide these haven't tested thoroughly.
Ask existing customers, not just the vendor's references (they pick the happy ones). What's the mean time between failures (MTBF)? Mean time to repair (MTTR)? How many unscheduled downtime hours last year?
Can your existing operators run it, or do you need a data scientist? Modern AI sorters should have intuitive interfaces. If the vendor needs a PhD on-site, keep looking.
What's the recommended maintenance schedule? Are spare parts locally available? What's the annual maintenance cost as % of CAPEX? Budget 2–5% of CAPEX annually for maintenance on optical equipment.
Does the vendor provide regular AI model updates? Is there a subscription fee? Can the system learn from your specific material stream over time? A sorter that improves with use is worth a premium over one that stays static.
Vendor spec sheets are full of jargon. Here's what matters and what to ignore.
| Specification | What It Actually Means | What's Good |
|---|---|---|
| Detection Resolution | Smallest object the sensor can reliably identify. Usually in mm per pixel. | <2mm/pixel for fine materials (flakes, e-waste chips); 3–5mm for whole containers |
| Ejector Pitch / Nozzle Spacing | Distance between air nozzles. Closer = more precise ejection = less compressed air waste. | 8–12mm pitch for precision sorting; 16–25mm acceptable for bulk sorting |
| Spectral Range | Wavelengths the sensor covers. NIR typically 900–1700nm. Hyperspectral covers 400–2500nm. | Wider range = more material discrimination. For plastics: VIS+NIR minimum; SWIR extends the detectable wavelength range |
| Frame Rate / Acquisition Speed | How fast the sensor captures and processes images. Higher = handles faster belts. | >200 fps for belt speeds above 3 m/s; >1000 fps for high-speed lines |
| AI Inference Time | Time from image capture to ejection decision. Must be < time for material to travel from sensor to ejector bank. | <5ms for high-speed lines; <15ms for standard lines |
| Compressed Air Consumption | The hidden OPEX driver. Air is expensive — about 20–30% of total operating cost. | <80 m³/h at 6 bar for a standard unit; ask for specific consumption per ton processed |
| Working Width | Width of the conveyor/sorting channel. Dictates how many units you need in parallel. | Match to your line width. 600mm, 1000mm, 1400mm, 2000mm are common standards |
| IP Rating | Dust and water protection. Recycling environments are harsh. | IP65 minimum for optical components; IP66 for washdown areas |
| Equipment Type | Price Range (2026, USD) | Installation & Commissioning |
|---|---|---|
| AI Optical Sorter (1-chute) | $180,000 – $450,000 | +15–25% of equipment cost |
| NIR Optical Sorter | $120,000 – $320,000 | +15–25% |
| Hyperspectral Sorter | $250,000 – $600,000 | +20–30% |
| X-Ray (XRT) Sorter | $300,000 – $700,000 | +20–30% (+ radiation safety compliance) |
| Eddy Current Separator | $40,000 – $150,000 | +10–15% |
| Magnetic Separator (Overband/Drum) | $15,000 – $60,000 | +5–10% |
| Robotic Picker (per robot + vision) | $120,000 – $250,000 | +10–20% |
| Ballistic Separator | $50,000 – $180,000 | +10–15% |
| Cost Category | Typical Range | Notes |
|---|---|---|
| Compressed Air | $12,000 – $45,000/yr per sorter | Often underestimated. Get specific consumption from vendor |
| Electricity | $3,000 – $12,000/yr per sorter | Depends on local rates and sensor type |
| Spare Parts & Consumables | $5,000 – $20,000/yr per sorter | Ejector valves, belts, filters, lamps/lasers (finite life) |
| Preventive Maintenance | $8,000 – $25,000/yr per sorter | Includes technician visits, calibration, cleaning |
| Software / AI Subscription | $0 – $18,000/yr per sorter | Some vendors bundle; others charge separately for AI updates |
| Labor (Operators) | $30,000 – $80,000/yr per shift | AI sorters reduce labor; budget 1 operator per 3–5 machines |
Buying a great sorter that doesn't fit your line is an expensive mistake. Address these points before signing.
A $50k cheaper sorter that produces 5% lower purity loses far more in reduced bale revenue over its lifetime. Calculate cost per ton of salable output, not equipment price.
Vendors run demos with clean, pre-sorted, dry material at optimal belt speed. Your material is dirtier, wetter, and more variable. Insist on testing your actual infeed material.
A sorter that uses 20% more compressed air than a competitor can cost $8,000–$15,000 more per year in electricity alone. Always compare specific air consumption (m³ per ton processed).
A $300k sorter fed by inconsistent, overlapping, or oversized material will underperform a $150k sorter with perfect infeed. Invest in upstream prep — ballistic separators, screens, and spreaders.
Does the sorter support adding sensors later (e.g., adding hyperspectral to NIR)? Can AI models be retrained for new materials? Buying a locked-down system saves money now but costs more when your material stream changes.
Optical sorters are precision equipment. A $25k/year service contract that prevents 3 days of downtime pays for itself. But read the SLA: "best effort" response is not the same as "technician on-site within 48 hours."
A sorter achieving 99% purity by ejecting 40% of the target material as "residue" is losing you money. Always ask: purity AND recovery, together, on your material.
Use this quick framework to estimate whether upgrading sorting equipment makes financial sense for your facility. For a detailed interactive calculator, see our Recycling ROI Calculator.
| Factor | Manual Sorting | Basic Optical | AI Optical |
|---|---|---|---|
| Labor cost per ton processed | $18–35 | $6–12 | $2–5 |
| Typical purity (mixed plastics) | 65–75% | 80–88% | 92–97% |
| Bale price premium per 1% purity gain | — | $8–15/ton | $8–15/ton |
| Recovery rate (good material captured) | 68–75% | 78–88% | 90–96% |
| Annual throughput per worker/unit | 150–400 tons | 8,000–25,000 tons | 6,000–30,000 tons |
| Typical payback period | Baseline | 2–4 years | 1.5–3 years |
Annual Benefit = (Labor Savings) + (Purity Premium × Volume) + (Recovery Gain × Material Value) − (Increased OPEX)
For a typical 20,000 t/yr mixed plastics line upgrading from manual to AI optical:
The recycling industry is changing fast. Regulations are tightening (EU PPWR, extended producer responsibility schemes), brands are demanding higher recycled content, and material streams are becoming more complex. When you buy sorting equipment with a 10–15 year lifespan, you're betting on what the industry will look like in 2035.
The best sorting equipment is the one that makes you money every day for 10+ years, not the one with the flashiest demo or the lowest price tag. Invest in reliability, flexibility, and a vendor who'll answer the phone at 2am when your line is down. Your operators — and your P&L — will thank you.