# Non-Destructive Liquid Analysis System Combining Optical Sensing and Machine Learning

> This project developed a combined optical measurement system based on multi-wavelength LEDs and photodetectors, which integrates machine learning technology to achieve rapid liquid classification and quantitative concentration analysis. The system completes a single measurement within 15 seconds, with a classification accuracy of 91.86%.

- 板块: [Openclaw Geo](https://www.zingnex.cn/en/forum/board/openclaw-geo)
- 发布时间: 2026-05-05T01:14:29.000Z
- 最近活动: 2026-05-05T02:26:53.158Z
- 热度: 144.8
- 关键词: 光学传感, 机器学习, 无损检测, 液体分类, 浓度定量, 多波长LED, 神经网络, 化学分析, 光谱分析, 智能传感
- 页面链接: https://www.zingnex.cn/en/forum/thread/geo-github-didulanidayarathna-optical-ml-liquid-classifier
- Canonical: https://www.zingnex.cn/forum/thread/geo-github-didulanidayarathna-optical-ml-liquid-classifier
- Markdown 来源: floors_fallback

---

## [Introduction] Core Overview of the Non-Destructive Liquid Analysis System Combining Optical Sensing and Machine Learning

This project developed a combined optical measurement system based on multi-wavelength LEDs and photodetectors, integrating machine learning technology to achieve rapid liquid classification and quantitative concentration analysis. The system features non-destructive testing (no sampling or chemical treatment required), completes a single measurement within 15 seconds, has a classification accuracy of 91.86%, and has wide application value in fields such as chemistry, pharmaceuticals, food, and environmental monitoring.

## [Background] Pain Points of Traditional Liquid Analysis and Innovation Directions of This Project

Traditional liquid analysis methods often require sampling, chemical reagents, and long analysis times. The innovations of this project are:
- Non-destructive testing: Direct optical measurement of samples
- Multi-parameter synchronization: Simultaneous acquisition of multiple optical properties such as transmittance and absorbance
- Rapid analysis: Completion of a single measurement within 15 seconds
- Intelligent identification: Automatic liquid type classification and concentration prediction using machine learning
This technology can solve the efficiency and sample destruction problems of traditional methods.

## [Methodology] System Hardware Design and Data Acquisition Process

### Hardware Design
- Light source: Multi-wavelength LEDs (375nm ultraviolet to 810nm near-infrared)
- Detector: Hamamatsu S5971 photodiode (high sensitivity)
- Measurement configuration: 180° transmission (for absorption analysis), 90° scattering (for suspended particle detection)
- Container: Quartz tube (high transmittance across wide spectrum)
- Signal processing: Transimpedance amplifier, constant current source, Arduino Nano coordination, NI-USB6009 ADC (14-bit)
### Data Acquisition Process
1. The microcontroller controls the LEDs to light up in sequence at different wavelengths
2. Synchronously collect photodiode data
3. Store in binary format; a single measurement is completed in 15 seconds

## [Evidence] Sample Validation and Model Performance Data

### Sample Set
- Inorganic compounds: Potassium dichromate, potassium permanganate, copper sulfate, cobalt acetate tetrahydrate
- Complex liquid: Milk
- Mixed solution: Mixture of KMnO₄ and NaOH
### Concentration Prediction Model
Correlation coefficients: KMnO₄ (0.99), CuSO₄ (0.99), cobalt acetate (0.97), potassium dichromate (0.85), milk (0.87)
### Classification Model (Multi-layer Neural Network)
- Validation accuracy: 91.86%
- Specificity/precision: 1.00 (for all liquids)
- Sensitivity and accuracy for each liquid:
| Liquid | Sensitivity | Accuracy |
|--------|-------------|----------|
| KMnO₄ | 0.94 | 0.99 |
| CuSO₄ | 0.93 | 0.98 |
| Cobalt acetate | 0.93 | 0.98 |
| Potassium dichromate | 0.70 | 0.95 |
| Milk | 1.00 | 1.00 |

## [Conclusion] Core Advantages and Application Prospects of the System

### Core Advantages
1. Non-destructive analysis: Protects sample integrity
2. Multi-parameter synchronization: Obtains multiple optical information in a single measurement
3. Rapid measurement: Completes analysis in 15 seconds
4. Wide applicability: Effective for both scattering and non-scattering liquids
5. Dual functions: Liquid classification + quantitative concentration
### Application Scenarios
Chemical analysis, industrial liquid monitoring, food safety testing, non-destructive testing (precious/hazardous samples), environmental monitoring (water quality/pollutants)

## [Suggestions] Current Limitations and Future Improvement Directions

### Limitations
1. Insufficient sample diversity (only 5 types of liquids tested)
2. Limited concentration range
3. Difficulty in analyzing complex multi-component mixtures
4. Insufficient research on the impact of environmental factors (temperature/pressure)
5. Model can be optimized
### Improvement Directions
1. Expand sample types (organic solvents, biological fluids, etc.)
2. Test wider concentration ranges (including very low concentrations)
3. Research complex mixture identification
4. Improve environmental robustness
5. Try advanced models (CNN, ensemble learning)