Sensing & Characterization¶
Updated to manuscript V3
The review emphasizes in-situ sensing that replaces separate characterization: viscosity from pipetting mass flow [22] or dispense time [23], from stirring forces [31] or filmed flow [28]; extrusion monitored by computer vision [33]; powder volume verified by vision during scooping [12]. A formulation experiment also records time series (balance, pressure, torque, self-powered sensor signals [131]) and spectra such as the acoustic-to-electrical response of energy-harvesting composites [132]; see Data & Decision Models.
Overview¶
Before any control strategy can work, you must measure what you're trying to control. This page covers sensors and characterization techniques for material properties and process state.
Sensor Categories¶
1. Material Property Sensors (Offline)¶
Used to characterize materials before the main task. Typically take 30 seconds to 5 minutes per measurement.
Viscosity/Rheology¶
- Rotational rheometer (Anton Paar, TA Instruments, Brookfield)
- Cost: $50k–$150k
- Measurement time: 2–10 minutes
- Accuracy: ±1–2%
- Output: viscosity vs. shear rate curve (full flow behavior)
- Vibrational viscometer (falling ball, tuning fork)
- Cost: $5k–$20k
- Measurement time: 30 seconds
- Accuracy: ±2–5%
- Output: single viscosity value at defined shear rate
Particle Size & Flowability¶
- Laser diffraction (Malvern Mastersizer)
- Cost: $30k–$80k
- Measurement time: 2 minutes
- Accuracy: ±2–5%
- Output: particle size distribution (PSD)
- Powder rheometer (FT4 Freeman Technology)
- Cost: $60k–$100k
- Measurement time: 5 minutes
- Accuracy: Empirical (not absolute)
- Output: flowability indices (BFE, SI, HR)
- Sieve analysis (manual)
- Cost: $500
- Measurement time: 30 minutes
- Accuracy: ±5%
- Output: cumulative PSD at discrete size bins
Moisture Content¶
- Moisture analyzer (Karl Fischer, loss-on-drying)
- Cost: $5k–$20k
- Measurement time: 5 minutes
- Accuracy: ±0.01%
- Output: water content (%w/w)
Spectroscopy¶
- UV-Vis spectrometer (PerkinElmer, Shimadzu)
- Cost: $10k–$50k
- Measurement time: 1 minute
- Accuracy: ±1%
- Output: absorbance at wavelengths (→ concentration via Beer-Lambert)
- FTIR (Fourier Transform Infrared)
- Cost: $20k–$80k
- Measurement time: 2 minutes
- Accuracy: Qualitative or semi-quantitative
- Output: chemical composition fingerprint
2. Process State Sensors (Online)¶
Used during the main task to monitor progress and enable feedback control. Typical sampling rates 100 Hz – 1 kHz.
Gravimetric (Weight)¶
- Precision balance (Mettler-Toledo XPE)
- Cost: $3k–$10k
- Range: 0–610 g
- Accuracy: ±0.001 g (1 mg)
- Bandwidth: 10 Hz (slow for real-time, but typical)
- Use: Powder dosing feedback
- Load cell (compression or tension)
- Cost: $100–$1k
- Range: 1 kg – 500 kg
- Accuracy: ±0.1% of range
- Bandwidth: 100–500 Hz (fast)
- Use: Real-time weight measurement, gripper force detection
Force & Torque¶
- 6-axis F/T sensor (ATI, Schunk, JR3)
- Cost: $2k–$10k
- Range: ±500 N, ±50 N·m (various models)
- Accuracy: ±1% of full scale
- Bandwidth: 500 Hz – 2 kHz
- Use: Extrusion pressure feedback, gripper force control, contact detection
Pressure¶
- Pressure transducer (Honeywell, Hydac)
- Cost: $200–$500
- Range: 0–10 bar (typical)
- Accuracy: ±0.5% FS
- Bandwidth: 100–500 Hz
- Use: Syringe pressure, pump outlet pressure, pneumatic feedback
Flow Rate¶
- Coriolis flowmeter (Micro Motion)
- Cost: $5k–$15k
- Range: 0.001–50 kg/min
- Accuracy: ±0.5%
- Bandwidth: 50 Hz
- Use: Extrusion rate monitoring
- Turbine flowmeter
- Cost: $500–$2k
- Range: Tunable per design
- Accuracy: ±1–2%
- Bandwidth: 100–500 Hz
- Use: Liquid pumping feedback
Displacement & Position¶
- Linear encoder / LVDT (Linear Variable Differential Transformer)
- Cost: $300–$1k
- Range: Typically 0–100 mm
- Accuracy: ±0.1 mm
- Bandwidth: 100 Hz – 10 kHz
- Use: Syringe plunger position, gripper finger displacement
- Potentiometer (rotary or linear)
- Cost: $50–$200
- Accuracy: ±2–5%
- Bandwidth: Very high (> 10 kHz)
- Use: Budget feedback, less critical measurements
3. Vision Sensors¶
Standard Camera¶
- USB camera (generic)
- Cost: $50–$500
- Resolution: 640×480 – 4K
- Frame rate: 30–120 fps
- Latency: 50–100 ms
- Use: Object detection, fill level monitoring
- Industrial camera (Basler, FLIR)
- Cost: $1k–$5k
- Resolution: 1K – 4K
- Frame rate: 60–500 fps (high-speed options)
- Latency: 10–20 ms
- Use: High-speed process monitoring, real-time feedback
Depth Camera¶
- RealSense / Kinect (structured light)
- Cost: $100–$500
- Depth range: 0.1–10 m
- Accuracy: ±2–5 mm
- Frame rate: 30–90 fps
- Use: 3D object detection, bin picking
- LiDAR (time-of-flight)
- Cost: $500–$5k
- Depth range: 0.1–100 m
- Accuracy: ±5–50 mm (depends on model)
- Frame rate: 10–20 Hz
- Use: 3D scene reconstruction, large-scale mapping
4. Environmental Sensors¶
Temperature¶
- Thermistor / RTD (Resistance Temperature Detector)
- Cost: $10–$100
- Accuracy: ±0.1–1°C
- Bandwidth: 10 Hz
- Use: Viscosity correction (viscosity drops ~2% per °C)
Humidity¶
- Capacitive humidity sensor
- Cost: $20–$100
- Accuracy: ±2–5% RH
- Bandwidth: 10 Hz
- Use: Moisture-sensitive powder properties
Sensor Accuracy Requirements¶
Key rule: Sensor accuracy should be 5–10× better than your control tolerance.
Example: - Target dosing accuracy: ±1% - Required sensor accuracy: ±0.1–0.2% - Implication: A 5 g target requires ±5–10 mg sensor accuracy - Load cell with ±1% accuracy (typically ±50 mg at 5 kg range) is marginal - Better: Use a precision scale with ±1 mg accuracy
Sensor Fusion¶
Combining multiple sensors improves reliability:
Example: Powder Dispensing¶
Approach 1: Weight only - Use load cell - Update frequency: 100 Hz - Latency: 10 ms - Robustness: Fails if scale drifts
Approach 2: Weight + vision (fill level) - Load cell provides primary feedback - Camera measures fill height as secondary check - If weight and fill level don't match, flag error - Robustness: Redundancy catches scale drift
Approach 3: Weight + flow meter - Load cell measures total accumulated mass - Flow meter measures instantaneous flow rate - Fuse both: if flow meter says "too slow" but weight is on target, feeder may be slowing down - Robustness: Predictive error detection
Calibration and Maintenance¶
Load Cell Calibration¶
- Zero calibration: With no load, set reading to 0
- Span calibration: With known weight (e.g., 5 kg), set reading to 5 kg
- Linearity check: Test at 25%, 50%, 75%, 100% of range
- Frequency: Weekly for production equipment, monthly for lab
Vision System Calibration¶
- Camera intrinsics: Focal length, principal point, distortion (one-time, or if lens changes)
- Camera extrinsics: Position and orientation relative to robot base (one-time or weekly)
- Lighting: Consistent illumination (check if shadows change with time of day)
Rheometer Calibration¶
- Geometry calibration: Empty spindle, verify torque = 0
- Temperature control: Peltier element stabilized to ±0.1°C
- Standard material: Test on reference fluid (e.g., silicone oil, known viscosity)
Sensor Selection Checklist¶
For your material handling task, answer:
- What variable am I trying to control? (weight, force, fill level, etc.)
- What is my target accuracy? (±0.1%, ±1%, ±5%, etc.)
- What is my process duration? (seconds vs. minutes)
- What sensor bandwidth do I need? (10 Hz, 100 Hz, 1 kHz?)
- What is my budget per sensor? ($100, $1k, $10k?)
- Do I need multiple sensors (redundancy/fusion)?
- What is my calibration maintenance plan? (weekly, monthly, quarterly)
References¶
- Rastegardoost, M. M.; Bao, Z.; Sadr, S.; Abdollahi, N.; Darvish, K.; Clasky, A.; Ohsaka, H.; Le Devedec, F. Self-Driving Laboratories for Formulation: Handling Liquids, Pastes, Powders and Objects, from Robots to Digital Twins and AI Agents. Manuscript in preparation, 2026.
- Key references from the review: [12], [22], [23], [28], [31], [33], [131], [132]. Full list on the References page.
Further reading (from the earlier wiki)¶
- ISO 13320: Particle size analysis – Laser diffraction method
- ISO 6440: Vegetable fats and oils – Determination of viscosity
- NIST Traceability Certificate (for calibration standards)